Autosegmentation for thoracic radiation treatment planning: A grand challenge at AAPM 2017

Towards more precise automatic analysis: a systematic review of deep learning-based multi-organ segmentation

Accurate segmentation of multiple organs in the head, neck, chest, and abdomen from medical images is an essential step in computer-aided diagnosis, surgical navigation, and radiation therapy. In the past few years, with a data-driven feature extraction approach and end-to-end training, automatic deep learning-based multi-organ segmentation methods have far outperformed traditional methods and become a new research topic. This review systematically summarizes the latest research in this field. We searched Google Scholar for papers published from January 1, 2016 to December 31, 2023, using keywords "multi-organ segmentation" and "deep learning", resulting in 327 papers. We followed the PRISMA guidelines for paper selection, and 195 studies were deemed to be within the scope of this review. We summarized the two main aspects involved in multi-organ segmentation: datasets and methods. Regarding datasets, we provided an overview of existing public datasets and conducted an in-depth analysis. Concerning methods, we categorized existing approaches into three major classes: fully supervised, weakly supervised and semi-supervised, based on whether they require complete label information. We summarized the achievements of these methods in terms of segmentation accuracy. In the discussion and conclusion section, we outlined and summarized the current trends in multi-organ segmentation.

Accurate object localization facilitates automatic esophagus segmentation in deep learning

BACKGROUND

Currently, automatic esophagus segmentation remains a challenging task due to its small size, low contrast, and large shape variation. We aimed to improve the performance of esophagus segmentation in deep learning by applying a strategy that involves locating the object first and then performing the segmentation task.

METHODS

A total of 100 cases with thoracic computed tomography scans from two publicly available datasets were used in this study. A modified CenterNet, an object location network, was employed to locate the center of the esophagus for each slice. Subsequently, the 3D U-net and 2D U-net_coarse models were trained to segment the esophagus based on the predicted object center. A 2D U-net_fine model was trained based on the updated object center according to the 3D U-net model. The dice similarity coefficient and the 95% Hausdorff distance were used as quantitative evaluation indexes for the delineation performance. The characteristics of the automatically delineated esophageal contours by the 2D U-net and 3D U-net models were summarized. Additionally, the impact of the accuracy of object localization on the delineation performance was analyzed. Finally, the delineation performance in different segments of the esophagus was also summarized.

RESULTS

The mean dice coefficient of the 3D U-net, 2D U-net_coarse, and 2D U-net_fine models were 0.77, 0.81, and 0.82, respectively. The 95% Hausdorff distance for the above models was 6.55, 3.57, and 3.76, respectively. Compared with the 2D U-net, the 3D U-net has a lower incidence of delineating wrong objects and a higher incidence of missing objects. After using the fine object center, the average dice coefficient was improved by 5.5% in the cases with a dice coefficient less than 0.75, while that value was only 0.3% in the cases with a dice coefficient greater than 0.75. The dice coefficients were lower for the esophagus between the orifice of the inferior and the pulmonary bifurcation compared with the other regions.

CONCLUSION

The 3D U-net model tended to delineate fewer incorrect objects but also miss more objects. Two-stage strategy with accurate object location could enhance the robustness of the segmentation model and significantly improve the esophageal delineation performance, especially for cases with poor delineation results.

SAROS: A dataset for whole-body region and organ segmentation in CT imaging

The Sparsely Annotated Region and Organ Segmentation (SAROS) dataset was created using data from The Cancer Imaging Archive (TCIA) to provide a large open-access CT dataset with high-quality annotations of body landmarks. In-house segmentation models were employed to generate annotation proposals on randomly selected cases from TCIA. The dataset includes 13 semantic body region labels (abdominal/thoracic cavity, bones, brain, breast implant, mediastinum, muscle, parotid/submandibular/thyroid glands, pericardium, spinal cord, subcutaneous tissue) and six body part labels (left/right arm/leg, head, torso). Case selection was based on the DICOM series description, gender, and imaging protocol, resulting in 882 patients (438 female) for a total of 900 CTs. Manual review and correction of proposals were conducted in a continuous quality control cycle. Only every fifth axial slice was annotated, yielding 20150 annotated slices from 28 data collections. For the reproducibility on downstream tasks, five cross-validation folds and a test set were pre-defined. The SAROS dataset serves as an open-access resource for training and evaluating novel segmentation models, covering various scanner vendors and diseases.

Cascaded-TOARNet: A cascaded framework based on mixed attention and multiscale information for thoracic OARs segmentation

BACKGROUND

Accurate and automated segmentation of thoracic organs-at-risk (OARs) is critical for radiotherapy treatment planning of thoracic cancers. However, this has remained a challenging task for four major reasons: (1) thoracic OARs have diverse morphologies; (2) thoracic OARs have low contrast with the background; (3) boundaries of thoracic OARs are blurry; (4) class imbalance issue caused by small organs.

PURPOSE

To overcome the above challenges and achieve accurate and automated segmentation of thoracic OARs on thoracic CT.

METHODS

A novel cascaded framework based on mixed attention and multiscale information for thoracic OARs segmentation, called Cascaded-TOARNet. This cascaded framework comprises two stages: localization and segmentation. During the localization stage, TOARNet locates each organ to crop the regions of interest (ROIs). During the segmentation stage, TOARNet accurately segments the ROIs, and the segmentation results are merged into a complete result.

RESULTS

We evaluated our proposed method and other common segmentation methods on two public datasets: the AAPM Thoracic Auto-Segmentation Challenge dataset and the Segmentation of Thoracic Organs at Risk (SegTHOR) dataset. Our method demonstrated superior performance, achieving a mean Dice score of 92.6% on the SegTHOR dataset and 90.8% on the AAPM dataset.

CONCLUSIONS

This segmentation method holds great promise as an essential tool for enhancing the efficiency of thoracic radiotherapy planning.

NRG Oncology Assessment of Artificial Intelligence Deep Learning-Based Auto-segmentation for Radiation Therapy: Current Developments, Clinical Considerations, and Future Directions

Deep learning neural networks (DLNN) in Artificial intelligence (AI) have been extensively explored for automatic segmentation in radiotherapy (RT). In contrast to traditional model-based methods, data-driven AI-based models for auto-segmentation have shown high accuracy in early studies in research settings and controlled environment (single institution). Vendor-provided commercial AI models are made available as part of the integrated treatment planning system (TPS) or as a stand-alone tool that provides streamlined workflow interacting with the main TPS. These commercial tools have drawn clinics' attention thanks to their significant benefit in reducing the workload from manual contouring and shortening the duration of treatment planning. However, challenges occur when applying these commercial AI-based segmentation models to diverse clinical scenarios, particularly in uncontrolled environments. Contouring nomenclature and guideline standardization has been the main task undertaken by the NRG Oncology. AI auto-segmentation holds the potential clinical trial participants to reduce interobserver variations, nomenclature non-compliance, and contouring guideline deviations. Meanwhile, trial reviewers could use AI tools to verify contour accuracy and compliance of those submitted datasets. In recognizing the growing clinical utilization and potential of these commercial AI auto-segmentation tools, NRG Oncology has formed a working group to evaluate the clinical utilization and potential of commercial AI auto-segmentation tools. The group will assess in-house and commercially available AI models, evaluation metrics, clinical challenges, and limitations, as well as future developments in addressing these challenges. General recommendations are made in terms of the implementation of these commercial AI models, as well as precautions in recognizing the challenges and limitations.

A Deformable Constraint Transport Network for Optimal Aortic Segmentation From CT Images

Aortic segmentation from computed tomography (CT) is crucial for facilitating aortic intervention, as it enables clinicians to visualize aortic anatomy for diagnosis and measurement. However, aortic segmentation faces the challenge of variable geometry in space, as the geometric diversity of different diseases and the geometric transformations that occur between raw and measured images. Existing constraint-based methods can potentially solve the challenge, but they are hindered by two key issues: inaccurate definition of properties and inappropriate topology of transformation in space. In this paper, we propose a deformable constraint transport network (DCTN). The DCTN adaptively extracts aortic features to define intra-image constrained properties and guides topological implementation in space to constrain inter-image geometric transformation between raw and curved planar reformation (CPR) images. The DCTN contains a deformable attention extractor, a geometry-aware decoder and an optimal transport guider. The extractor generates variable patches that preserve semantic integrity and long-range dependency in long-sequence images. The decoder enhances the perception of geometric texture and semantic features, particularly for low-intensity aortic coarctation and false lumen, which removes background interference. The guider explores the geometric discrepancies between raw and CPR images, constructs probability distributions of discrepancies, and matches them with inter-image transformation to guide geometric topology in space. Experimental studies on 267 aortic subjects and four public datasets show the superiority of our DCTN over 23 methods. The results demonstrate DCTN's advantages in aortic segmentation for different types of aortic disease, for different aortic segments, and in the measurement of clinical indexes.

External validation, radiological evaluation, and development of deep learning automatic lung segmentation in contrast-enhanced chest CT

OBJECTIVES

There is a need for CT pulmonary angiography (CTPA) lung segmentation models. Clinical translation requires radiological evaluation of model outputs, understanding of limitations, and identification of failure points. This multicentre study aims to develop an accurate CTPA lung segmentation model, with evaluation of outputs in two diverse patient cohorts with pulmonary hypertension (PH) and interstitial lung disease (ILD).

METHODS

This retrospective study develops an nnU-Net-based segmentation model using data from two specialist centres (UK and USA). Model was trained (n = 37), tested (n = 12), and clinically evaluated (n = 176) on a diverse 'real-world' cohort of 225 PH patients with volumetric CTPAs. Dice score coefficient (DSC) and normalised surface distance (NSD) were used for testing. Clinical evaluation of outputs was performed by two radiologists who assessed clinical significance of errors. External validation was performed on heterogenous contrast and non-contrast scans from 28 ILD patients.

RESULTS

A total of 225 PH and 28 ILD patients with diverse demographic and clinical characteristics were evaluated. Mean accuracy, DSC, and NSD scores were 0.998 (95% CI 0.9976, 0.9989), 0.990 (0.9840, 0.9962), and 0.983 (0.9686, 0.9972) respectively. There were no segmentation failures. On radiological review, 82% and 71% of internal and external cases respectively had no errors. Eighteen percent and 25% respectively had clinically insignificant errors. Peripheral atelectasis and consolidation were common causes for suboptimal segmentation. One external case (0.5%) with patulous oesophagus had a clinically significant error.

CONCLUSION

State-of-the-art CTPA lung segmentation model provides accurate outputs with minimal clinical errors on evaluation across two diverse cohorts with PH and ILD.

CLINICAL RELEVANCE

Clinical translation of artificial intelligence models requires radiological review and understanding of model limitations. This study develops an externally validated state-of-the-art model with robust radiological review. Intended clinical use is in techniques such as lung volume or parenchymal disease quantification.

KEY POINTS

• Accurate, externally validated CT pulmonary angiography (CTPA) lung segmentation model tested in two large heterogeneous clinical cohorts (pulmonary hypertension and interstitial lung disease). • No segmentation failures and robust review of model outputs by radiologists found 1 (0.5%) clinically significant segmentation error. • Intended clinical use of this model is a necessary step in techniques such as lung volume, parenchymal disease quantification, or pulmonary vessel analysis.

Ultrasound image segmentation of renal tumors based on UNet++ with fusion of multiscale residuals and dual attention

Objective. Laparoscopic renal unit-preserving resection is a routine and effective means of treating renal tumors. Image segmentation is an essential part before tumor resection. The current segmentation method mainly relies on doctors manual delineation, which is time-consuming, labor-intensive, and influenced by their personal experience and ability. And the image quality of segmentation is low, with problems such as blurred edges, unclear size and shape, which are not conducive to clinical diagnosis.Approach. To address these problems, we propose an automated segmentation method, i.e. the UNet++ algorithm fusing multiscale residuals and dual attention (MRDA_UNet++). It replaces two consecutive 3 × 3 convolutions in UNet++ with the 'MultiRes block' module, which incorporates coordinate attention to fuse features from different scales and suppress the impact of background noise. Furthermore, an attention gate is also added at the short connections to enhance the ability of the network to extract features from the target area.Main results. The experimental results show that MRDA_UNet++ achieves 93.18%, 92.87%, 93.66%, and 92.09% on the real-world dataset for MIoU, Dice, Precision, and Recall, respectively. Compared to the baseline model UNet++ on three public datasets, the MIoU, Dice, and Recall metrics improved by 6.00%, 7.90% and 18.09% respectively for BUSI, 0.39%, 0.27% and 1.03% for Dataset C, and 1.37%, 1.75% and 1.30% for DDTI.Significance. The proposed MRDA_UNet++ exhibits obvious advantages in feature extraction, which can not only significantly reduce the workload of doctors, but also further decrease the risk of misdiagnosis. It is of great value to assist doctors diagnosis in the clinic.

Zero-Shot Medical Image Translation via Frequency-Guided Diffusion Models

Recently, the diffusion model has emerged as a superior generative model that can produce high quality and realistic images. However, for medical image translation, the existing diffusion models are deficient in accurately retaining structural information since the structure details of source domain images are lost during the forward diffusion process and cannot be fully recovered through learned reverse diffusion, while the integrity of anatomical structures is extremely important in medical images. For instance, errors in image translation may distort, shift, or even remove structures and tumors, leading to incorrect diagnosis and inadequate treatments. Training and conditioning diffusion models using paired source and target images with matching anatomy can help. However, such paired data are very difficult and costly to obtain, and may also reduce the robustness of the developed model to out-of-distribution testing data. We propose a frequency-guided diffusion model (FGDM) that employs frequency-domain filters to guide the diffusion model for structure-preserving image translation. Based on its design, FGDM allows zero-shot learning, as it can be trained solely on the data from the target domain, and used directly for source-to-target domain translation without any exposure to the source-domain data during training. We evaluated it on three cone-beam CT (CBCT)-to-CT translation tasks for different anatomical sites, and a cross-institutional MR imaging translation task. FGDM outperformed the state-of-the-art methods (GAN-based, VAE-based, and diffusion-based) in metrics of Fréchet Inception Distance (FID), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index Measure (SSIM), showing its significant advantages in zero-shot medical image translation.

Framework for Radiation Oncology Department-wide Evaluation and Implementation of Commercial Artificial Intelligence Autocontouring

PURPOSE

Artificial intelligence (AI)-based autocontouring in radiation oncology has potential benefits such as standardization and time savings. However, commercial AI solutions require careful evaluation before clinical integration. We developed a multidimensional evaluation method to test pretrained AI-based automated contouring solutions across a network of clinics.

METHODS AND MATERIALS

Curated data included 121 patient planning computed tomography (CT) scans with a total of 859 clinically approved contours used for treatment from 4 clinics. Regions of interest (ROIs) were generated with 3 commercial AI-based automated contouring software solutions (AI1, AI2, AI3) spanning the following disease sites: brain, head and neck (H&N), thorax, abdomen, and pelvis. Quantitative agreement between AI-generated and clinical contours was measured by Dice similarity coefficient (DSC) and Hausdorff distance (HD). Qualitative assessment was performed by multiple experts scoring blinded AI-contours using a Likert scale. Workflow and usability surveying was also conducted.

RESULTS

AI1, AI2, and AI3 contours had high quantitative agreement in 27.8%, 32.8%, and 34.1% of cases (DSC >0.9), performing well in pelvis (median DSC = 0.86/0.88/0.91) and thorax (median DSC = 0.91/0.89/0.91). All 3 solutions had low quantitative agreement in 7.4%, 8.8%, and 6.1% of cases (DSC <0.5), performing worse in brain (median DSC = 0.65/0.78/0.75) and H&N (median DSC = 0.76/0.80/0.81). Qualitatively, AI1 and AI2 contours were acceptable (rated 1-2) with at most minor edits in 70.7% and 74.6% of ROIs (2906 ratings), higher for abdomen (AI1: 79.2%) and thorax (AI2: 90.2%), and lower for H&N (29.0/35.6%). An end-user survey showed strong user preference for full automation and mixed preferences for accuracy versus total number of structures generated.

CONCLUSIONS

Our evaluation method provided a comprehensive analysis of both quantitative and qualitative measures of commercially available pretrained AI autocontouring algorithms. The evaluation framework served as a roadmap for clinical integration that aligned with user workflow preference.

Gross failure rates and failure modes for a commercial AI-based auto-segmentation algorithm in head and neck cancer patients

PURPOSE

Artificial intelligence (AI) based commercial software can be used to automatically delineate organs at risk (OAR), with potential for efficiency savings in the radiotherapy treatment planning pathway, and reduction of inter- and intra-observer variability. There has been little research investigating gross failure rates and failure modes of such systems.

METHOD

50 head and neck (H&N) patient data sets with "gold standard" contours were compared to AI-generated contours to produce expected mean and standard deviation values for the Dice Similarity Coefficient (DSC), for four common H&N OARs (brainstem, mandible, left and right parotid). An AI-based commercial system was applied to 500 H&N patients. AI-generated contours were compared to manual contours, outlined by an expert human, and a gross failure was set at three standard deviations below the expected mean DSC. Failures were inspected to assess reason for failure of the AI-based system with failures relating to suboptimal manual contouring censored. True failures were classified into 4 sub-types (setup position, anatomy, image artefacts and unknown).

RESULTS

There were 24 true failures of the AI-based commercial software, a gross failure rate of 1.2%. Fifteen failures were due to patient anatomy, four were due to dental image artefacts, three were due to patient position and two were unknown. True failure rates by OAR were 0.4% (brainstem), 2.2% (mandible), 1.4% (left parotid) and 0.8% (right parotid).

CONCLUSION

True failures of the AI-based system were predominantly associated with a non-standard element within the CT scan. It is likely that these non-standard elements were the reason for the gross failure, and suggests that patient datasets used to train the AI model did not contain sufficient heterogeneity of data. Regardless of the reasons for failure, the true failure rate for the AI-based system in the H&N region for the OARs investigated was low (∼1%).

Deep learning for automated segmentation in radiotherapy: a narrative review

The segmentation of organs and structures is a critical component of radiation therapy planning, with manual segmentation being a laborious and time-consuming task. Interobserver variability can also impact the outcomes of radiation therapy. Deep neural networks have recently gained attention for their ability to automate segmentation tasks, with convolutional neural networks (CNNs) being a popular approach. This article provides a descriptive review of the literature on deep learning (DL) techniques for segmentation in radiation therapy planning. This review focuses on five clinical sub-sites and finds that U-net is the most commonly used CNN architecture. The studies using DL for image segmentation were included in brain, head and neck, lung, abdominal, and pelvic cancers. The majority of DL segmentation articles in radiation therapy planning have concentrated on normal tissue structures. N-fold cross-validation was commonly employed, without external validation. This research area is expanding quickly, and standardization of metrics and independent validation are critical to benchmarking and comparing proposed methods.

GPU-accelerated lung CT segmentation based on level sets and texture analysis

This paper presents a novel semi-automatic method for lung segmentation in thoracic CT datasets. The fully three-dimensional algorithm is based on a level set representation of an active surface and integrates texture features to improve its robustness. The method's performance is enhanced by the graphics processing unit (GPU) acceleration. The segmentation process starts with a manual initialisation of 2D contours on a few representative slices of the analysed volume. Next, the starting regions for the active surface are generated according to the probability maps of texture features. The active surface is then evolved to give the final segmentation result. The recent implementation employs features based on grey-level co-occurrence matrices and Gabor filters. The algorithm was evaluated on real medical imaging data from the LCTCS 2017 challenge. The results were also compared with the outcomes of other segmentation methods. The proposed approach provided high segmentation accuracy while offering very competitive performance.

Systematic Evaluation of the Impact of Lung Segmentation Methods on 4-Dimensional Computed Tomography Ventilation Imaging Using a Large Patient Database

PURPOSE

A novel form of lung functional imaging applied for functional avoidance radiation therapy has been developed that uses 4-dimensional computed tomography (4DCT) data and image processing techniques to calculate lung ventilation (4DCT-ventilation). Lung segmentation is a common step to define a region of interest for 4DCT-ventilation generation. The purpose of this study was to quantitatively evaluate the sensitivity of 4DCT-ventilation imaging using different lung segmentation methods.

METHODS AND MATERIALS

The 4DCT data of 350 patients from 2 institutions were used. Lung contours were generated using 3 methods: (1) reference segmentations that removed airways and pulmonary vasculature manually (Lung-Manual), (2) standard lung contours used for planning (Lung-RadOnc), and (3) artificial intelligence (AI)-based contours that removed the airways and pulmonary vasculature (Lung-AI). The AI model was based on a residual 3-dimensional U-Net and was trained using the Lung-Manual contours of 279 patients. We compared the Lung-RadOnc or Lung-AI with Lung-Manual contours for the entire 4DCT-ventilation functional avoidance process including lung segmentation (surface Dice similarity coefficient [Surface DSC]), 4DCT-ventilation generation (correlation), and subanalysis of 10 patients on a dosimetric endpoint (percentage of high functional volume of lung receiving ≥20 Gy [fV20{%}]).

RESULTS

Surface DSC comparing Lung-Manual/Lung-RadOnc and Lung-Manual/Lung-AI contours was 0.40 ± 0.06 and 0.86 ± 0.04, respectively. The correlation between 4DCT-ventilation images generated with Lung-Manual/Lung-RadOnc and Lung-Manual/Lung-AI were 0.48 ± 0.14 and 0.85 ± 0.14, respectively. The difference in fV20[%] between 4DCT-ventilation generated with Lung-Manual/Lung-RadOnc and Lung-Manual/Lung-AI was 2.5% ± 4.1% and 0.3% ± 0.5%, respectively.

CONCLUSIONS

Our work showed that using standard planning lung contours can result in significantly variable 4DCT-ventilation images. The study demonstrated that AI-based segmentations generate lung contours and 4DCT-ventilation images that are similar to those generated using manual methods. The significance of the study is that it characterizes the lung segmentation sensitivity of the 4DCT-ventilation process and develops methods that can facilitate the integration of this novel imaging in busy clinics.

Review and recommendations on deformable image registration uncertainties for radiotherapy applications

Deformable image registration (DIR) is a versatile tool used in many applications in radiotherapy (RT). DIR algorithms have been implemented in many commercial treatment planning systems providing accessible and easy-to-use solutions. However, the geometric uncertainty of DIR can be large and difficult to quantify, resulting in barriers to clinical practice. Currently, there is no agreement in the RT community on how to quantify these uncertainties and determine thresholds that distinguish a good DIR result from a poor one. This review summarises the current literature on sources of DIR uncertainties and their impact on RT applications. Recommendations are provided on how to handle these uncertainties for patient-specific use, commissioning, and research. Recommendations are also provided for developers and vendors to help users to understand DIR uncertainties and make the application of DIR in RT safer and more reliable.

Multimodal graph attention network for COVID-19 outcome prediction

When dealing with a newly emerging disease such as COVID-19, the impact of patient- and disease-specific factors (e.g., body weight or known co-morbidities) on the immediate course of the disease is largely unknown. An accurate prediction of the most likely individual disease progression can improve the planning of limited resources and finding the optimal treatment for patients. In the case of COVID-19, the need for intensive care unit (ICU) admission of pneumonia patients can often only be determined on short notice by acute indicators such as vital signs (e.g., breathing rate, blood oxygen levels), whereas statistical analysis and decision support systems that integrate all of the available data could enable an earlier prognosis. To this end, we propose a holistic, multimodal graph-based approach combining imaging and non-imaging information. Specifically, we introduce a multimodal similarity metric to build a population graph that shows a clustering of patients. For each patient in the graph, we extract radiomic features from a segmentation network that also serves as a latent image feature encoder. Together with clinical patient data like vital signs, demographics, and lab results, these modalities are combined into a multimodal representation of each patient. This feature extraction is trained end-to-end with an image-based Graph Attention Network to process the population graph and predict the COVID-19 patient outcomes: admission to ICU, need for ventilation, and mortality. To combine multiple modalities, radiomic features are extracted from chest CTs using a segmentation neural network. Results on a dataset collected in Klinikum rechts der Isar in Munich, Germany and the publicly available iCTCF dataset show that our approach outperforms single modality and non-graph baselines. Moreover, our clustering and graph attention increases understanding of the patient relationships within the population graph and provides insight into the network's decision-making process.

Evaluation of different algorithms for automatic segmentation of head-and-neck lymph nodes on CT images

PURPOSE

To investigate the performance of 4 atlas-based (multi-ABAS) and 2 deep learning (DL) solutions for head-and-neck (HN) elective nodes (CTVn) automatic segmentation (AS) on CT images.

MATERIAL AND METHODS

Bilateral CTVn levels of 69 HN cancer patients were delineated on contrast-enhanced planning CT. Ten and 49 patients were used for atlas library and for training a mono-centric DL model, respectively. The remaining 20 patients were used for testing. Additionally, three commercial multi-ABAS methods and one commercial multi-centric DL solution were investigated. Quantitative evaluation was assessed using volumetric Dice Similarity Coefficient (DSC) and 95-percentile Hausdorff distance (HD95%). Blind evaluation was performed for 3 solutions by 4 physicians. One recorded the time needed for manual corrections. A dosimetric study was finally conducted using automated planning.

RESULTS

Overall DL solutions had better DSC and HD95% results than multi-ABAS methods. No statistically significant difference was found between the 2 DL solutions. However, the contours provided by multi-centric DL solution were preferred by all physicians and were also faster to correct (1.1 min vs 4.17 min, on average). Manual corrections for multi-ABAS contours took on average 6.52 min Overall, decreased contour accuracy was observed from CTVn2 to CTVn3 and to CTVn4. Using the AS contours in treatment planning resulted in underdosage of the elective target volume.

CONCLUSION

Among all methods, the multi-centric DL method showed the highest delineation accuracy and was better rated by experts. Manual corrections remain necessary to avoid elective target underdosage. Finally, AS contours help reducing the workload of manual delineation task.

Feasibility evaluation of novel AI-based deep-learning contouring algorithm for radiotherapy

PURPOSE

To evaluate the clinical feasibility of the Siemens Healthineers AI-Rad Companion Organs RT VA30A (Organs-RT) auto-contouring algorithm for organs at risk (OARs) of the pelvis, thorax, and head and neck (H&N).

METHODS

Computed tomography (CT) datasets from 30 patients (10 pelvis, 10 thorax, and 10 H&N) were collected. Four sets of OARs were generated on each scan, one set by Organs-RT and the others by three experienced users independently. A physician (expert) then evaluated each contour by assigning a score from the following scale: 1-Must Redo, 2-Major Edits, 3-Minor Edits, 4-Clinically usable. Using the highest-scored OAR from the human users as a reference, the contours generated by Organs-RT were evaluated via Dice Similarity Coefficient (DSC), Hausdorff Distance (HDD), Mean Distance to Agreement (mDTA), Volume comparison, and visual inspection. Additionally, each human user recorded the time to delineate each structure set and time-saving efficiency was measured.

RESULTS

The average DSC obtained for the pelvic OARs ranged between (0.81 ± 0.06)Rectum and (0.94 ± 0.03)Bladder . (0.75 ± 0.09)Esophagus to( 0.96 ± 0.02 ) Rt . Lung ${( {0.96 \pm 0.02} )}_{{\mathrm{Rt}}.{\mathrm{\ Lung}}}$ for the thoracic OARs and (0.66 ± 0.07)Lips to (0.83 ± 0.04)Brainstem for the H&N. The average HDD in cm for the pelvis cohort ranged between (0.95 ± 0.35)Bladder to (3.62 ± 2.50)Rectum , (0.42 ± 0.06)SpinalCord to (2.09 ± 2.00)Esophagus for the thoracic set and( 0.53 ± 0.22 ) Cerv _ SpinalCord ${( {0.53 \pm 0.22} )}_{{\mathrm{Cerv}}\_{\mathrm{SpinalCord}}}$ to (1.50 ± 0.50)Mandible for the H&N region. The time-saving efficiency was 67% for H&N, 83% for pelvis, and 84% for thorax. 72.5%, 82%, and 50% of the pelvis, thorax, and H&N OARs were scored as clinically usable by the expert, respectively.

CONCLUSIONS

The highest agreement registered between OARs generated by Organs-RT and their respective references was for the bladder, heart, lungs, and femoral heads, with an overall DSC≥0.92. The poorest agreement was for the rectum, esophagus, and lips, with an overall DSC⩽0.81. Nonetheless, Organs-RT serves as a reliable auto-contouring tool by minimizing overall contouring time and increasing time-saving efficiency in radiotherapy treatment planning.

Towards deep-learning (DL) based fully automated target delineation for rectal cancer neoadjuvant radiotherapy using a divide-and-conquer strategy: a study with multicenter blind and randomized validation

PURPOSE

Manual clinical target volume (CTV) and gross tumor volume (GTV) delineation for rectal cancer neoadjuvant radiotherapy is pivotal but labor-intensive. This study aims to propose a deep learning (DL)-based workflow towards fully automated clinical target volume (CTV) and gross tumor volume (GTV) delineation for rectal cancer neoadjuvant radiotherapy.

MATERIALS & METHODS

We retrospectively included 141 patients with Stage II-III mid-low rectal cancer and randomly grouped them into training (n = 121) and testing (n = 20) cohorts. We adopted a divide-and-conquer strategy to address CTV and GTV segmentation using two separate DL models with DpuUnet as backend-one model for CTV segmentation in the CT domain, and the other for GTV in the MRI domain. The workflow was validated using a three-level multicenter-involved blind and randomized evaluation scheme. Dice similarity coefficient (DSC) and 95th percentile Hausdorff distance (95HD) metrics were calculated in Level 1, four-grade expert scoring was performed in Level 2, and head-to-head Turing test in Level 3.

RESULTS

For the DL-based CTV contours over the testing cohort, the DSC and 95HD (mean ± SD) were 0.85 ± 0.06 and 7.75 ± 6.42 mm respectively, and 96.4% cases achieved clinical viable scores (≥ 2). The positive rate in the Turing test was 52.3%. For GTV, the DSC and 95HD were 0.87 ± 0.07 and 4.07 ± 1.67 mm respectively, and 100% of the DL-based contours achieved clinical viable scores (≥ 2). The positive rate in the Turing test was 52.0%.

CONCLUSION

The proposed DL-based workflow exhibited promising accuracy and excellent clinical viability towards automated CTV and GTV delineation for rectal cancer neoadjuvant radiotherapy.

Human factors in the clinical implementation of deep learning-based automated contouring of pelvic organs at risk for MRI-guided radiotherapy

PURPOSE

Deep neural nets have revolutionized the science of auto-segmentation and present great promise for treatment planning automation. However, little data exists regarding clinical implementation and human factors. We evaluated the performance and clinical implementation of a novel deep learning-based auto-contouring workflow for 0.35T magnetic resonance imaging (MRI)-guided pelvic radiotherapy, focusing on automation bias and objective measures of workflow savings.

METHODS

An auto-contouring model was developed using a UNet-derived architecture for the femoral heads, bladder, and rectum in 0.35T MR images. Training data was taken from 75 patients treated with MRI-guided radiotherapy at our institution. The model was tested against 20 retrospective cases outside the training set, and subsequently was clinically implemented. Usability was evaluated on the first 30 clinical cases by computing Dice coefficient (DSC), Hausdorff distance (HD), and the fraction of slices that were used un-modified by planners. Final contours were retrospectively reviewed by an experienced planner and clinical significance of deviations was graded as negligible, low, moderate, and high probability of leading to actionable dosimetric variations. In order to assess whether the use of auto-contouring led to final contours more or less in agreement with an objective standard, 10 pre-treatment and 10 post-treatment blinded cases were re-contoured from scratch by three expert planners to get expert consensus contours (EC). EC was compared to clinically used (CU) contours using DSC. Student's t-test and Levene's statistic were used to test statistical significance of differences in mean and standard deviation, respectively. Finally, the dosimetric significance of the contour differences were assessed by comparing the difference in bladder and rectum maximum point doses between EC and CU before and after the introduction of automation.

RESULTS

Median (interquartile range) DSC for the retrospective test data were 0.92(0.02), 0.92(0.06), 0.93(0.06), 0.87(0.04) for the post-processed contours for the right and left femoral heads, bladder, and rectum, respectively. Post-implementation median DSC were 1.0(0.0), 1.0(0.0), 0.98(0.04), and 0.98(0.06), respectively. For each organ, 96.2, 95.4, 59.5, and 68.21 percent of slices were used unmodified by the planner. DSC between EC and pre-implementation CU contours were 0.91(0.05*), 0.91*(0.05*), 0.95(0.04), and 0.88(0.04) for right and left femoral heads, bladder, and rectum, respectively. The corresponding DSC for post-implementation CU contours were 0.93(0.02*), 0.93*(0.01*), 0.96(0.01), and 0.85(0.02) (asterisks indicate statistically significant difference). In a retrospective review of contours used for planning, a total of four deviating slices in two patients were graded as low potential clinical significance. No deviations were graded as moderate or high. Mean differences between EC and CU rectum max-doses were 0.1 ± 2.6 Gy and -0.9 ± 2.5 Gy for pre- and post-implementation, respectively. Mean differences between EC and CU bladder/bladder wall max-doses were -0.9 ± 4.1 Gy and 0.0 ± 0.6 Gy for pre- and post-implementation, respectively. These differences were not statistically significant according to Student's t-test.

CONCLUSION

We have presented an analysis of the clinical implementation of a novel auto-contouring workflow. Substantial workflow savings were obtained. The introduction of auto-contouring into the clinical workflow changed the contouring behavior of planners. Automation bias was observed, but it had little deleterious effect on treatment planning.

Geometric and dosimetric analysis of CT- and MR-based automatic contouring for the EPTN contouring atlas in neuro-oncology

PURPOSE

Atlas-based and deep-learning contouring (DLC) are methods for automatic segmentation of organs-at-risk (OARs). The European Particle Therapy Network (EPTN) published a consensus-based atlas for delineation of OARs in neuro-oncology. In this study, geometric and dosimetric evaluation of automatically-segmented neuro-oncological OARs was performed using CT- and MR-models following the EPTN-contouring atlas.

METHODS

Image and contouring data from 76 neuro-oncological patients were included. Two atlas-based models (CT-atlas and MR-atlas) and one DLC-model (MR-DLC) were created. Manual contours on registered CT-MR-images were used as ground-truth. Results were analyzed in terms of geometrical (volumetric Dice similarity coefficient (vDSC), surface DSC (sDSC), added path length (APL), and mean slice-wise Hausdorff distance (MSHD)) and dosimetrical accuracy. Distance-to-tumor analysis was performed to analyze to which extent the location of the OAR relative to planning target volume (PTV) has dosimetric impact, using Wilcoxon rank-sum tests.

RESULTS

CT-atlas outperformed MR-atlas for 22/26 OARs. MR-DLC outperformed MR-atlas for all OARs. Highest median (95 %CI) vDSC and sDSC were found for the brainstem in MR-DLC: 0.92 (0.88-0.95) and 0.84 (0.77-0.89) respectively, as well as lowest MSHD: 0.27 (0.22-0.39)cm. Median dose differences (ΔD) were within ± 1 Gy for 24/26(92 %) OARs for all three models. Distance-to-tumor showed a significant correlation for ΔDmax,0.03cc-parameters when splitting the data in ≤ 4 cm and > 4 cm OAR-distance (p < 0.001).

CONCLUSION

MR-based DLC and CT-based atlas-contouring enable high-quality segmentation. It was shown that a combination of both CT- and MR-autocontouring models results in the best quality.

Feasibility and Clinical Acceptability of Automation-Assisted 3D Conformal Radiotherapy Planning for Patients With Cervical Cancer in a Resource-Constrained Setting

PURPOSE

The Ocean Road Cancer Institute (ORCI) in Tanzania began offering 3D conformal radiation therapy (3DCRT) in 2018. Steep learning curves, high patient volume, and a limited workforce resulted in long radiation therapy (RT) planning workflows. We aimed to establish the feasibility of implementing an automation-assisted cervical cancer 3DCRT planning system.

MATERIALS AND METHODS

We performed chart abstractions on 30 patients with cervical cancer treated with 3DCRT at ORCI. The Radiation Planning Assistant (RPA) generated a new automated set of contours and plans on the basis of anonymized computed tomography images. Each were assessed for edit time requirements, dose-volume safety metrics, and clinical acceptability by two ORCI physician investigators. Dice similarity coefficient (DSC) agreement analysis was conducted between original and new contour sets.

RESULTS

The average time to manually develop treatment plans was 7 days. Applying RPA, automated same-day contours and plans were developed for 29 of 30 patients (97%). Of the 29 evaluable contours, all were approved with <2 minutes of edit time. Agreement between clinical and RPA contours was highest for the rectum (median DSC, 0.72) and bladder (DSC, 0.90). Agreement was lower with the primary tumor clinical target volume (CTVp; DSC, 0.69) and elective nodal clinical target volume (CTVn; DSC, 0.63). All RPA plans were approved with <4 minutes of edit time. RPA target coverage was excellent, covering the CTVp with median V45 Gy 100% and CTVn with median V45 Gy 99.9%.

CONCLUSION

Automation-assisted 3DCRT contouring yielded high levels of agreement for normal structures. The RPA met all planning safety metrics and sustained high levels of clinical acceptability with minimal edit times. This tool offers the potential to significantly decrease RT planning timelines while maintaining high-quality RT delivery in resource-constrained settings.

A clinical evaluation of the performance of five commercial artificial intelligence contouring systems for radiotherapy

Purpose/objectives

Auto-segmentation with artificial intelligence (AI) offers an opportunity to reduce inter- and intra-observer variability in contouring, to improve the quality of contours, as well as to reduce the time taken to conduct this manual task. In this work we benchmark the AI auto-segmentation contours produced by five commercial vendors against a common dataset.

Methods and materials

The organ at risk (OAR) contours generated by five commercial AI auto-segmentation solutions (Mirada (Mir), MVision (MV), Radformation (Rad), RayStation (Ray) and TheraPanacea (Ther)) were compared to manually-drawn expert contours from 20 breast, 20 head and neck, 20 lung and 20 prostate patients. Comparisons were made using geometric similarity metrics including volumetric and surface Dice similarity coefficient (vDSC and sDSC), Hausdorff distance (HD) and Added Path Length (APL). To assess the time saved, the time taken to manually draw the expert contours, as well as the time to correct the AI contours, were recorded.

Results

There are differences in the number of CT contours offered by each AI auto-segmentation solution at the time of the study (Mir 99; MV 143; Rad 83; Ray 67; Ther 86), with all offering contours of some lymph node levels as well as OARs. Averaged across all structures, the median vDSCs were good for all systems and compared favorably with existing literature: Mir 0.82; MV 0.88; Rad 0.86; Ray 0.87; Ther 0.88. All systems offer substantial time savings, ranging between: breast 14-20 mins; head and neck 74-93 mins; lung 20-26 mins; prostate 35-42 mins. The time saved, averaged across all structures, was similar for all systems: Mir 39.8 mins; MV 43.6 mins; Rad 36.6 min; Ray 43.2 mins; Ther 45.2 mins.

Conclusions

All five commercial AI auto-segmentation solutions evaluated in this work offer high quality contours in significantly reduced time compared to manual contouring, and could be used to render the radiotherapy workflow more efficient and standardized.

Applying a novel two-step deep learning network to improve the automatic delineation of esophagus in non-small cell lung cancer radiotherapy

Purpose

To introduce a model for automatic segmentation of thoracic organs at risk (OARs), especially the esophagus, in non-small cell lung cancer radiotherapy, using a novel two-step deep learning network.

Materials and methods

A total of 59 lung cancer patients' CT images were enrolled, of which 39 patients were randomly selected as the training set, 8 patients as the validation set, and 12 patients as the testing set. The automatic segmentations of the six OARs including the esophagus were carried out. In addition, two sets of treatment plans were made on the basis of the manually delineated tumor and OARs (Plan1) as well as the manually delineated tumor and the automatically delineated OARs (Plan2). The Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), and average surface distance (ASD) of the proposed model were compared with those of U-Net as a benchmark. Next, two groups of plans were also compared according to the dose-volume histogram parameters.

Results

The DSC, HD95, and ASD of the proposed model were better than those of U-Net, while the two groups of plans were almost the same. The highest mean DSC of the proposed method was 0.94 for the left lung, and the lowest HD95 and ASD were 3.78 and 1.16 mm for the trachea, respectively. Moreover, the DSC reached 0.73 for the esophagus.

Conclusions

The two-step segmentation method can accurately segment the OARs of lung cancer. The mean DSC of the esophagus realized preliminary clinical significance (>0.70). Choosing different deep learning networks based on different characteristics of organs offers a new option for automatic segmentation in radiotherapy.

Contouring quality assurance methodology based on multiple geometric features against deep learning auto-segmentation

BACKGROUND

Contouring error is one of the top failure modes in radiation treatment. Multiple efforts have been made to develop tools to automatically detect segmentation errors. Deep learning-based auto-segmentation (DLAS) has been used as a baseline for flagging manual segmentation errors, but those efforts are limited to using only one or two contour comparison metrics.

PURPOSE

The purpose of this research is to develop an improved contouring quality assurance system to identify and flag manual contouring errors.

METHODS AND MATERIALS

DLAS contours were used as a reference to compare with manually segmented contours. A total of 27 geometric agreement metrics were determined from the comparisons between the two segmentation approaches. Feature selection was performed to optimize the training of a machine learning classification model to identify potential contouring errors. A public dataset with 339 cases was used to train and test the classifier. Four independent classifiers were trained using five-fold cross validation, and the predictions from each classifier were ensembled using soft voting. The trained model was validated on a held-out testing dataset. An additional independent clinical dataset with 60 cases was used to test the generalizability of the model. Model predictions were reviewed by an expert to confirm or reject the findings.

RESULTS

The proposed machine learning multiple features (ML-MF) approach outperformed traditional nonmachine-learning-based approaches that are based on only one or two geometric agreement metrics. The machine learning model achieved recall (precision) values of 0.842 (0.899), 0.762 (0.762), 0.727 (0.842), and 0.773 (0.773) for Brainstem, Parotid_L, Parotid_R, and mandible contours, respectively compared to 0.526 (0.909), 0.619 (0.765), 0.682 (0.882), 0.773 (0.568) for an approach based solely on Dice similarity coefficient values. In the external validation dataset, 66.7, 93.3, 94.1, and 58.8% of flagged cases were confirmed to have contouring errors by an expert for Brainstem, Parotid_L, Parotid_R, and mandible contours, respectively.

CONCLUSIONS

The proposed ML-MF approach, which includes multiple geometric agreement metrics to flag manual contouring errors, demonstrated superior performance in comparison to traditional methods. This method is easy to implement in clinical practice and can help to reduce the significant time and labor costs associated with manual segmentation and review.

Quantification of liver-Lung shunt fraction on 3D SPECT/CT images for selective internal radiation therapy of liver cancer using CNN-based segmentations and non-rigid registration

PURPOSE

Selective internal radiation therapy (SIRT) has been proven to be an effective treatment for hepatocellular carcinoma (HCC) patients. In clinical practice, the treatment planning for SIRT using ⁹⁰Y microspheres requires estimation of the liver-lung shunt fraction (LSF) to avoid radiation pneumonitis. Currently, the manual segmentation method to draw a region of interest (ROI) of the liver and lung in 2D planar imaging of ^99mTc-MAA and 3D SPECT/CT images is inconvenient, time-consuming and observer-dependent. In this study, we propose and evaluate a nearly automatic method for LSF quantification using 3D SPECT/CT images, offering improved performance compared with the current manual segmentation method.

METHODS

We retrospectively acquired 3D SPECT with non-contrast-enhanced CT images (nCECT) of 60 HCC patients from a SPECT/CT scanning machine, along with the corresponding diagnostic contrast-enhanced CT images (CECT). Our approach for LSF quantification is to use CNN-based methods for liver and lung segmentations in the nCECT image. We first apply 3D ResUnet to coarsely segment the liver. If the liver segmentation contains a large error, we dilate the coarse liver segmentation into the liver mask as a ROI in the nCECT image. Subsequently, non-rigid registration is applied to deform the liver in the CECT image to fit that obtained in the nCECT image. The final liver segmentation is obtained by segmenting the liver in the deformed CECT image using nnU-Net. In addition, the lung segmentations are obtained using 2D ResUnet. Finally, LSF quantitation is performed based on the number of counts in the SPECT image inside the segmentations. Evaluations and Results: To evaluate the liver segmentation accuracy, we used Dice similarity coefficient (DSC), asymmetric surface distance (ASSD), and max surface distance (MSD) and compared the proposed method to five well-known CNN-based methods for liver segmentation. Furthermore, the LSF error obtained by the proposed method was compared to a state-of-the-art method, modified Deepmedic, and the LSF quantifications obtained by manual segmentation. The results show that the proposed method achieved a DSC score for the liver segmentation that is comparable to other state-of-the-art methods, with an average of 0.93, and the highest consistency in segmentation accuracy, yielding a standard deviation of the DSC score of 0.01. The proposed method also obtains the lowest ASSD and MSD scores on average (2.6 mm and 31.5 mm, respectively). Moreover, for the proposed method, a median LSF error of 0.14% is obtained, which is a statically significant improvement to the state-of-the-art-method (p=0.004), and is much smaller than the median error in LSF manual determination by the medical experts using 2D planar image (1.74% and p<0.001).

CONCLUSIONS

A method for LSF quantification using 3D SPECT/CT images based on CNNs and non-rigid registration was proposed, evaluated and compared to state-of-the-art techniques. The proposed method can quantitatively determine the LSF with high accuracy and has the potential to be applied in clinical practice.

A hybrid multi-particle approach to range assessment-based treatment verification in particle therapy

Particle therapy (PT) used for cancer treatment can spare healthy tissue and reduce treatment toxicity. However, full exploitation of the dosimetric advantages of PT is not yet possible due to range uncertainties, warranting development of range-monitoring techniques. This study proposes a novel range-monitoring technique introducing the yet unexplored concept of simultaneous detection and imaging of fast neutrons and prompt-gamma rays produced in beam-tissue interactions. A quasi-monolithic organic detector array is proposed, and its feasibility for detecting range shifts in the context of proton therapy is explored through Monte Carlo simulations of realistic patient models and detector resolution effects. The results indicate that range shifts of [Formula: see text] can be detected at relatively low proton intensities ([Formula: see text] protons/spot) when spatial information obtained through imaging of both particle species are used simultaneously. This study lays the foundation for multi-particle detection and imaging systems in the context of range verification in PT.

Assessment of Optimizers and their Performance in Autosegmenting Lung Tumors

Purpose

Optimizers are widely utilized across various domains to enhance desired outcomes by either maximizing or minimizing objective functions. In the context of deep learning, they help to minimize the loss function and improve model's performance. This study aims to evaluate the accuracy of different optimizers employed for autosegmentation of non-small cell lung cancer (NSCLC) target volumes on thoracic computed tomography images utilized in oncology.

Materials and Methods

The study utilized 112 patients, comprising 92 patients from "The Cancer Imaging Archive" (TCIA) and 20 of our local clinical patients, to evaluate the efficacy of various optimizers. The gross tumor volume was selected as the foreground mask for training and testing the models. Of the 92 TCIA patients, 57 were used for training and validation, and the remaining 35 for testing using nnU-Net. The performance of the final model was further evaluated on the 20 local clinical patient datasets. Six different optimizers, namely AdaDelta, AdaGrad, Adam, NAdam, RMSprop, and stochastic gradient descent (SGD), were investigated. To assess the agreement between the predicted volume and the ground truth, several metrics including Dice similarity coefficient (DSC), Jaccard index, sensitivity, precision, Hausdorff distance (HD), 95th percentile Hausdorff distance (HD95), and average symmetric surface distance (ASSD) were utilized.

Results

The DSC values for AdaDelta, AdaGrad, Adam, NAdam, RMSprop, and SGD were 0.75, 0.84, 0.85, 0.84, 0.83, and 0.81, respectively, for the TCIA test data. However, when the model trained on TCIA datasets was applied to the clinical datasets, the DSC, HD, HD95, and ASSD metrics showed a statistically significant decrease in performance compared to the TCIA test datasets, indicating the presence of image and/or mask heterogeneity between the data sources.

Conclusion

The choice of optimizer in deep learning is a critical factor that can significantly impact the performance of autosegmentation models. However, it is worth noting that the behavior of optimizers may vary when applied to new clinical datasets, which can lead to changes in models' performance. Therefore, selecting the appropriate optimizer for a specific task is essential to ensure optimal performance and generalizability of the model to different datasets.

Analytic calculations and synthetic shapes for validation of quantitative contour comparison software

A high level of variability in reported values was observed in a recent survey of contour similarity measures (CSMs) calculation tools. Such variations in the output measurements prevent meaningful comparison between studies. The purpose of this study was to develop a dataset with analytically calculated gold standard values to facilitate standardization and ensure accuracy of CSM implementations. The dataset was generated in the Digital Imaging and Communications in Medicine (DICOM) format. Both the dataset and the software used for its generation are made publicly available to encourage robust testing of CSM implementations for accuracy, improving consistency between different implementations.

Multiround Transfer Learning and Modified Generative Adversarial Network for Lung Cancer Detection

Lung cancer has been the leading cause of cancer death for many decades. With the advent of artificial intelligence, various machine learning models have been proposed for lung cancer detection (LCD). Typically, challenges in building an accurate LCD model are the small-scale datasets, the poor generalizability to detect unseen data, and the selection of useful source domains and prioritization of multiple source domains for transfer learning. In this paper, a multiround transfer learning and modified generative adversarial network (MTL-MGAN) algorithm is proposed for LCD. The MTL transfers the knowledge between the prioritized source domains and target domain to get rid of exhaust search of datasets prioritization among multiple datasets, maximizing the transferability with a multiround transfer learning process, and avoiding negative transfer via customization of loss functions in the aspects of domain, instance, and feature. In regard to the MGAN, it not only generates additional training data but also creates intermediate domains to bridge the gap between the source domains and target domains. 10 benchmark datasets are chosen for the performance evaluation and analysis of the MTL-MGAN. The proposed algorithm has significantly improved the accuracy compared with related works. To examine the contributions of the individual components of the MTL-MGAN, ablation studies are conducted to confirm the effectiveness of the prioritization algorithm, the MTL, the negative transfer avoidance via loss functions, and the MGAN. The research implications are to confirm the feasibility of multiround transfer learning to enhance the optimal solution of the target model and to provide a generic approach to bridge the gap between the source domain and target domain using MGAN.

Automated Contouring and Planning in Radiation Therapy: What Is 'Clinically Acceptable'?

Developers and users of artificial-intelligence-based tools for automatic contouring and treatment planning in radiotherapy are expected to assess clinical acceptability of these tools. However, what is 'clinical acceptability'? Quantitative and qualitative approaches have been used to assess this ill-defined concept, all of which have advantages and disadvantages or limitations. The approach chosen may depend on the goal of the study as well as on available resources. In this paper, we discuss various aspects of 'clinical acceptability' and how they can move us toward a standard for defining clinical acceptability of new autocontouring and planning tools.

Data synthesis and adversarial networks: A review and meta-analysis in cancer imaging

Despite technological and medical advances, the detection, interpretation, and treatment of cancer based on imaging data continue to pose significant challenges. These include inter-observer variability, class imbalance, dataset shifts, inter- and intra-tumour heterogeneity, malignancy determination, and treatment effect uncertainty. Given the recent advancements in image synthesis, Generative Adversarial Networks (GANs), and adversarial training, we assess the potential of these technologies to address a number of key challenges of cancer imaging. We categorise these challenges into (a) data scarcity and imbalance, (b) data access and privacy, (c) data annotation and segmentation, (d) cancer detection and diagnosis, and (e) tumour profiling, treatment planning and monitoring. Based on our analysis of 164 publications that apply adversarial training techniques in the context of cancer imaging, we highlight multiple underexplored solutions with research potential. We further contribute the Synthesis Study Trustworthiness Test (SynTRUST), a meta-analysis framework for assessing the validation rigour of medical image synthesis studies. SynTRUST is based on 26 concrete measures of thoroughness, reproducibility, usefulness, scalability, and tenability. Based on SynTRUST, we analyse 16 of the most promising cancer imaging challenge solutions and observe a high validation rigour in general, but also several desirable improvements. With this work, we strive to bridge the gap between the needs of the clinical cancer imaging community and the current and prospective research on data synthesis and adversarial networks in the artificial intelligence community.

Stress-testing pelvic autosegmentation algorithms using anatomical edge cases

Commercial autosegmentation has entered clinical use, however real-world performance may suffer in certain cases. We aimed to assess the influence of anatomic variants on performance. We identified 112 prostate cancer patients with anatomic variations (edge cases). Pelvic anatomy was autosegmented using three commercial tools. To evaluate performance, Dice similarity coefficients, and mean surface and 95% Hausdorff distances were calculated versus clinician-delineated references. Deep learning autosegmentation outperformed atlas-based and model-based methods. However, edge case performance was lower versus the normal cohort (0.12 mean DSC reduction). Anatomic variation presents challenges to commercial autosegmentation.

Medical image segmentation based on self-supervised hybrid fusion network

Automatic segmentation of medical images has been a hot research topic in the field of deep learning in recent years, and achieving accurate segmentation of medical images is conducive to breakthroughs in disease diagnosis, monitoring, and treatment. In medicine, MRI imaging technology is often used to image brain tumors, and further judgment of the tumor area needs to be combined with expert analysis. If the diagnosis can be carried out by computer-aided methods, the efficiency and accuracy will be effectively improved. Therefore, this paper completes the task of brain tumor segmentation by building a self-supervised deep learning network. Specifically, it designs a multi-modal encoder-decoder network based on the extension of the residual network. Aiming at the problem of multi-modal feature extraction, the network introduces a multi-modal hybrid fusion module to fully extract the unique features of each modality and reduce the complexity of the whole framework. In addition, to better learn multi-modal complementary features and improve the robustness of the model, a pretext task to complete the masked area is set, to realize the self-supervised learning of the network. Thus, it can effectively improve the encoder's ability to extract multi-modal features and enhance the noise immunity. Experimental results present that our method is superior to the compared methods on the tested datasets.

Strong semantic segmentation for Covid-19 detection: Evaluating the use of deep learning models as a performant tool in radiography

INTRODUCTION

With the increasing number of Covid-19 cases as well as care costs, chest diseases have gained increasing interest in several communities, particularly in medical and computer vision. Clinical and analytical exams are widely recognized techniques for diagnosing and handling Covid-19 cases. However, strong detection tools can help avoid damage to chest tissues. The proposed method provides an important way to enhance the semantic segmentation process using combined potential deep learning (DL) modules to increase consistency. Based on Covid-19 CT images, this work hypothesized that a novel model for semantic segmentation might be able to extract definite graphical features of Covid-19 and afford an accurate clinical diagnosis while optimizing the classical test and saving time.

METHODS

CT images were collected considering different cases (normal chest CT, pneumonia, typical viral causes, and Covid-19 cases). The study presents an advanced DL method to deal with chest semantic segmentation issues. The approach employs a modified version of the U-net to enable and support Covid-19 detection from the studied images.

RESULTS

The validation tests demonstrated competitive results with important performance rates: Precision (90.96% ± 2.5) with an F-score of (91.08% ± 3.2), an accuracy of (93.37% ± 1.2), a sensitivity of (96.88% ± 2.8) and a specificity of (96.91% ± 2.3). In addition, the visual segmentation results are very close to the Ground truth.

CONCLUSION

The findings of this study reveal the proof-of-principle for using cooperative components to strengthen the semantic segmentation modules for effective and truthful Covid-19 diagnosis.

IMPLICATIONS FOR PRACTICE

This paper has highlighted that DL based approach, with several modules, may be contributing to provide strong support for radiographers and physicians, and that further use of DL is required to design and implement performant automated vision systems to detect chest diseases.

Comparison of atlas-based and deep learning methods for organs at risk delineation on head-and-neck CT images using an automated treatment planning system

BACKGROUND AND PURPOSE

To investigate the performance of head-and-neck (HN) organs-at-risk (OAR) automatic segmentation (AS) using four atlas-based (ABAS) and two deep learning (DL) solutions.

MATERIAL AND METHODS

All patients underwent iodine contrast-enhanced planning CT. Fourteen OAR were manually delineated. DL.1 and DL.2 solutions were trained with 63 mono-centric patients and > 1000 multi-centric patients, respectively. Ten and 15 patients with varied anatomies were selected for the atlas library and for testing, respectively. The evaluation was based on geometric indices (DICE coefficient and 95th percentile-Hausdorff Distance (HD_95%)), time needed for manual corrections and clinical dosimetric endpoints obtained using automated treatment planning.

RESULTS

Both DICE and HD_95% results indicated that DL algorithms generally performed better compared with ABAS algorithms for automatic segmentation of HN OAR. However, the hybrid-ABAS (ABAS.3) algorithm sometimes provided the highest agreement to the reference contours compared with the 2 DL. Compared with DL.2 and ABAS.3, DL.1 contours were the fastest to correct. For the 3 solutions, the differences in dose distributions obtained using AS contours and AS + manually corrected contours were not statistically significant. High dose differences could be observed when OAR contours were at short distances to the targets. However, this was not always interrelated.

CONCLUSION

DL methods generally showed higher delineation accuracy compared with ABAS methods for AS segmentation of HN OAR. Most ABAS contours had high conformity to the reference but were more time consuming than DL algorithms, especially when considering the computing time and the time spent on manual corrections.

Generalizability of Machine Learning Models: Quantitative Evaluation of Three Methodological Pitfalls

Purpose

To investigate the impact of the following three methodological pitfalls on model generalizability: (a) violation of the independence assumption, (b) model evaluation with an inappropriate performance indicator or baseline for comparison, and (c) batch effect.

Materials and Methods

The authors used retrospective CT, histopathologic analysis, and radiography datasets to develop machine learning models with and without the three methodological pitfalls to quantitatively illustrate their effect on model performance and generalizability. F1 score was used to measure performance, and differences in performance between models developed with and without errors were assessed using the Wilcoxon rank sum test when applicable.

Results

Violation of the independence assumption by applying oversampling, feature selection, and data augmentation before splitting data into training, validation, and test sets seemingly improved model F1 scores by 71.2% for predicting local recurrence and 5.0% for predicting 3-year overall survival in head and neck cancer and by 46.0% for distinguishing histopathologic patterns in lung cancer. Randomly distributing data points for a patient across datasets superficially improved the F1 score by 21.8%. High model performance metrics did not indicate high-quality lung segmentation. In the presence of a batch effect, a model built for pneumonia detection had an F1 score of 98.7% but correctly classified only 3.86% of samples from a new dataset of healthy patients.

Conclusion

Machine learning models developed with these methodological pitfalls, which are undetectable during internal evaluation, produce inaccurate predictions; thus, understanding and avoiding these pitfalls is necessary for developing generalizable models.Keywords: Random Forest, Diagnosis, Prognosis, Convolutional Neural Network (CNN), Medical Image Analysis, Generalizability, Machine Learning, Deep Learning, Model Evaluation Supplemental material is available for this article. Published under a CC BY 4.0 license.

Multicenter comparison of measures for quantitative evaluation of contouring in radiotherapy

•

There is variation in contour measurement definitions used between institutions.

•

Differences in implementation can result in large discrepancies in reported results.

•

Many in-house implementations were found to have bugs in at least one measure.

•

Extreme care should be taken comparing results from different studies.

•

A dataset with known ground truth values would assist in validation of such tools.

Background and Purpose

A wide range of quantitative measures are available to facilitate clinical implementation of auto-contouring software, on-going Quality Assurance (QA) and interobserver contouring variation studies. This study aimed to assess the variation in output when applying different implementations of the measures to the same data in order to investigate how consistently such measures are defined and implemented in radiation oncology.

Materials and Methods

A survey was conducted to assess if there were any differences in definitions of contouring measures or their implementations that would lead to variation in reported results between institutions. This took two forms: a set of computed tomography (CT) image data with “Test” and “Reference” contours was distributed for participants to process using their preferred tools and report results, and a questionnaire regarding the definition of measures and their implementation was completed by the participants.

Results

Thirteen participants completed the survey and submitted results, with one commercial and twelve in-house solutions represented. Excluding outliers, variations of up to 50% in Dice Similarity Coefficient (DSC), 50% in 3D Hausdorff Distance (HD), and 200% in Average Distance (AD) were observed between the participant submitted results. Collaborative investigation with participants revealed a large number of bugs in implementation, confounding the understanding of intentional implementation choices.

Conclusion

Care must be taken when comparing quantitative results between different studies. There is a need for a dataset with clearly defined measures and ground truth for validation of such tools prior to their use.

Clinical evaluation of deep learning and atlas-based auto-segmentation for critical organs at risk in radiation therapy

INTRODUCTION

Contouring organs at risk (OARs) is a time-intensive task that is a critical part of radiation therapy. Atlas-based automatic segmentation has shown some success at reducing this time burden on practitioners; however, this method often requires significant manual editing to reach a clinically accurate standard. Deep learning (DL) auto-segmentation has recently emerged as a promising solution. This study compares the accuracy of DL and atlas-based auto-segmentation in relation to clinical 'gold standard' reference contours.

METHODS

Ninety CT datasets (30 head and neck, 30 thoracic, 30 pelvic) were automatically contoured using both atlas and DL segmentation techniques. Sixteen critical OARs were then quantitatively measured for accuracy using the Dice similarity coefficient (DSC) and Hausdorff distance (HD). Qualitative analysis was performed to visually classify the accuracy of each structure into one of four explicitly defined categories. Additionally, the time to edit atlas and DL contours to a clinically acceptable level was recorded for a subset of 9 OARs.

RESULTS

Of the 16 OARs analysed, DL delivered statistically significant improvements over atlas segmentation in 13 OARs measured with DSC, 12 OARs measured with HD, and 12 OARs measured qualitatively. The mean editing time for the subset of DL contours was 50%, 23% and 61% faster (all P < 0.05) than that of atlas segmentation for the head and neck, thorax, and pelvis respectively.

CONCLUSIONS

Deep learning segmentation comprehensively outperformed atlas-based contouring for the majority of evaluated OARs. Improvements were observed in geometric accuracy and visual acceptability, while editing time was reduced leading to increased workflow efficiency.

Clinical target volume automatic segmentation based on lymph node stations for lung cancer with bulky lump lymph nodes

Background

The lack of standardized delineation of lymph node station in lung cancer radiotherapy leads to nonstandard clinical target volume (CTV) contouring, especially in patients with bulky lump gross target volume lymph nodes (GTVnd). This study defines lymph node region boundaries in radiotherapy for lung cancer and automatically contours lymph node stations based on the International Association for the Study of Lung Cancer (IASLC) lymph node map.

Methods

Computed tomography (CT) scans of 200 patients with small cell lung cancer were collected. The lymph node zone boundaries were defined based on the IASLC lymph node map, with adjustments to meet radiotherapy requirements. Contours of lymph node stations were confirmed by two experienced oncologists. A model (DiUNet) was constructed by incorporating the contours of GTVnd to precisely contour the boundaries. Quantitative evaluation metrics and clinical evaluations were conducted.

Results

The mean 3D Dice similarity coefficient (Dice similarity coefficient) values of DiUNet in most lymph node stations was greater than 0.7, 98.87% of the lymph node station slices are accepted. The mean DiUNet score was not significantly different from that of the man contoured in the evaluation of lymph node stations and CTV.

Conclusion

This is the first study to propose a method that automatically contours lymph node regions station by station based on the IASLC lymph node map with bulky lump GTVnd. Delineation of lymph node stations based on the DiUNet model is a promising strategy to obtain accuracy and efficiency for CTV delineation in lung cancer patients, especially for bulky lump GTVnd.

Automated Segmentation of Spinal Muscles From Upright Open MRI Using a Multiscale Pyramid 2D Convolutional Neural Network

STUDY DESIGN

Randomized trial.

OBJECTIVE

To implement an algorithm enabling the automated segmentation of spinal muscles from open magnetic resonance images in healthy volunteers and patients with adult spinal deformity (ASD).

SUMMARY OF BACKGROUND DATA

Understanding spinal muscle anatomy is critical to diagnosing and treating spinal deformity.Muscle boundaries can be extrapolated from medical images using segmentation, which is usually done manually by clinical experts and remains complicated and time-consuming.

METHODS

Three groups were examined: two healthy volunteer groups (N = 6 for each group) and one ASD group (N = 8 patients) were imaged at the lumbar and thoracic regions of the spine in an upright open magnetic resonance imaging scanner while maintaining different postures (various seated, standing, and supine). For each group and region, a selection of regions of interest (ROIs) was manually segmented. A multiscale pyramid two-dimensional convolutional neural network was implemented to automatically segment all defined ROIs. A five-fold crossvalidation method was applied and distinct models were trained for each resulting set and group and evaluated using Dice coefficients calculated between the model output and the manually segmented target.

RESULTS

Good to excellent results were found across all ROIs for the ASD (Dice coefficient >0.76) and healthy (dice coefficient > 0.86) groups.

CONCLUSION

This study represents a fundamental step toward the development of an automated spinal muscle properties extraction pipeline, which will ultimately allow clinicians to have easier access to patient-specific simulations, diagnosis, and treatment.

Dose-volume-based evaluation of convolutional neural network-based auto-segmentation of thoracic organs at risk

•

Dice score and Hausdorff distance do not correlate with dose-volume-based results.

•

Auto-contours close to the tumor or in entry/exit beams should be checked.

•

Heart and esophagus must be checked for locally advanced non-small cell lung cancer.

•

Bronchi must be checked for peripheral early-stage non-small cell lung cancer.

•

Every treatment plan still passed the clinical goals for the manual organs at risk.

Background and purpose

The geometrical accuracy of auto-segmentation using convolutional neural networks (CNNs) has been demonstrated. This study aimed to investigate the dose-volume impact of differences between automatic and manual OARs for locally advanced (LA) and peripherally located early-stage (ES) non-small cell lung cancer (NSCLC).

Material and methods

A single CNN was created for automatic delineation of the heart, lungs, main left and right bronchus, esophagus, spinal cord and trachea using 55/10/40 patients for training/validation/testing. Dice score coefficient (DSC) and 95th percentile Hausdorff distance (HD95) were used for geometrical analysis. A new treatment plan based on the auto-segmented OARs was created for each test patient using 3D for ES-NSCLC (SBRT, 3–8 fractions) and IMRT for LA-NSCLC (24–35 fractions). The correlation between geometrical metrics and dose-volume differences was investigated.

Results

The average (±1 SD) DSC and HD95 were 0.82 ± 0.07 and 16.2 ± 22.4 mm, while the average dose-volume differences were 0.5 ± 1.5 Gy (ES) and 1.5 ± 2.8 Gy (LA). The geometrical metrics did not correlate with the observed dose-volume differences (average Pearson for DSC: −0.27 ± 0.18 (ES) and −0.09 ± 0.12 (LA); HD95: 0.1 ± 0.3 mm (ES) and 0.2 ± 0.2 mm (LA)).

Conclusions

After post-processing, manual adjustments of automatic contours are only needed for clinically relevant OARs situated close to the tumor or within an entry or exit beam e.g., the heart and the esophagus for LA-NSCLC and the bronchi for ES-NSCLC. The lungs do not need to be checked further in detail.

A Survey on Deep Learning for Precision Oncology

Precision oncology, which ensures optimized cancer treatment tailored to the unique biology of a patient’s disease, has rapidly developed and is of great clinical importance. Deep learning has become the main method for precision oncology. This paper summarizes the recent deep-learning approaches relevant to precision oncology and reviews over 150 articles within the last six years. First, we survey the deep-learning approaches categorized by various precision oncology tasks, including the estimation of dose distribution for treatment planning, survival analysis and risk estimation after treatment, prediction of treatment response, and patient selection for treatment planning. Secondly, we provide an overview of the studies per anatomical area, including the brain, bladder, breast, bone, cervix, esophagus, gastric, head and neck, kidneys, liver, lung, pancreas, pelvis, prostate, and rectum. Finally, we highlight the challenges and discuss potential solutions for future research directions.

Barriers and facilitators to clinical implementation of radiotherapy treatment planning automation: A survey study of medical dosimetrists

PURPOSE

Little is known about the scale of clinical implementation of automated treatment planning techniques in the United States. In this work, we examine the barriers and facilitators to adoption of commercially available automated planning tools into the clinical workflow using a survey of medical dosimetrists.

METHODS/MATERIALS

Survey questions were developed based on a literature review of automation research and cognitive interviews of medical dosimetrists at our institution. Treatment planning automation was defined to include auto-contouring and automated treatment planning. Survey questions probed frequency of use, positive and negative perceptions, potential implementation changes, and demographic and institutional descriptive statistics. The survey sample was identified using both a LinkedIn search and referral requests sent to physics directors and senior physicists at 34 radiotherapy clinics in our state. The survey was active from August 2020 to April 2021.

RESULTS

Thirty-four responses were collected out of 59 surveys sent. Three categories of barriers to use of automation were identified. The first related to perceptions of limited accuracy and usability of the algorithms. Eighty-eight percent of respondents reported that auto-contouring inaccuracy limited its use, and 62% thought it was difficult to modify an automated plan, thus limiting its usefulness. The second barrier relates to the perception that automation increases the probability of an error reaching the patient. Third, respondents were concerned that automation will make their jobs less satisfying and less secure. Large majorities reported that they enjoyed plan optimization, would not want to lose that part of their job, and expressed explicit job security fears.

CONCLUSION

To our knowledge this is the first systematic investigation into the views of automation by medical dosimetrists. Potential barriers and facilitators to use were explicitly identified. This investigation highlights several concrete approaches that could potentially increase the translation of automation into the clinic, along with areas of needed research.

Supervised and weakly supervised deep learning models for COVID-19 CT diagnosis: A systematic review

Artificial intelligence (AI) and computer vision (CV) methods become reliable to extract features from radiological images, aiding COVID-19 diagnosis ahead of the pathogenic tests and saving critical time for disease management and control. Thus, this review article focuses on cascading numerous deep learning-based COVID-19 computerized tomography (CT) imaging diagnosis research, providing a baseline for future research. Compared to previous review articles on the topic, this study pigeon-holes the collected literature very differently (i.e., its multi-level arrangement). For this purpose, 71 relevant studies were found using a variety of trustworthy databases and search engines, including Google Scholar, IEEE Xplore, Web of Science, PubMed, Science Direct, and Scopus. We classify the selected literature in multi-level machine learning groups, such as supervised and weakly supervised learning. Our review article reveals that weak supervision has been adopted extensively for COVID-19 CT diagnosis compared to supervised learning. Weakly supervised (conventional transfer learning) techniques can be utilized effectively for real-time clinical practices by reusing the sophisticated features rather than over-parameterizing the standard models. Few-shot and self-supervised learning are the recent trends to address data scarcity and model efficacy. The deep learning (artificial intelligence) based models are mainly utilized for disease management and control. Therefore, it is more appropriate for readers to comprehend the related perceptive of deep learning approaches for the in-progress COVID-19 CT diagnosis research.

Reduced Chest Computed Tomography Scan Length for Patients Positive for Coronavirus Disease 2019: Dose Reduction and Impact on Diagnostic Utility

METHODS

This study used the Personalized Rapid Estimation of Dose in CT (PREDICT) tool to estimate patient-specific organ doses from CT image data. The PREDICT is a research tool that combines a linear Boltzmann transport equation solver for radiation dose map generation with deep learning algorithms for organ contouring. Computed tomography images from 74 subjects in the Medical Imaging Data Resource Center-RSNA International COVID-19 Open Radiology Database data set (chest CT of adult patients positive for COVID-19), which included expert annotations including "infectious opacities," were analyzed. First, the full z-scan length of the CT image data set was evaluated. Next, the z-scan length was reduced from the left hemidiaphragm to the top of the aortic arch. Generic dose reduction based on dose length product (DLP) and patient-specific organ dose reductions were calculated. The percentage of infectious opacities excluded from the reduced z-scan length was used to quantify the effect on diagnostic utility.

RESULTS

Generic dose reduction, based on DLP, was 69%. The organ dose reduction ranged from approximately equal to 18% (breasts) to approximately equal to 64% (bone surface and bone marrow). On average, 12.4% of the infectious opacities were not included in the reduced z-coverage, per patient, of which 5.1% were above the top of the arch and 7.5% below the left hemidiaphragm.

CONCLUSIONS

Limiting z-scan length of chest CTs reduced radiation dose without significantly compromising diagnostic utility in COVID-19 patients. The PREDICT demonstrated that patient-specific organ dose reductions varied from generic dose reduction based on DLP.

Geometric and Dosimetric Evaluation of the Automatic Delineation of Organs at Risk (OARs) in Non-Small-Cell Lung Cancer Radiotherapy Based on a Modified DenseNet Deep Learning Network

Purpose

To introduce an end-to-end automatic segmentation model for organs at risk (OARs) in thoracic CT images based on modified DenseNet, and reduce the workload of radiation oncologists.

Materials and Methods

The computed tomography (CT) images of 36 lung cancer patients were included in this study, of which 27 patients’ images were randomly selected as the training set, 9 patients’ as the testing set. The validation set was generated by cross validation and 6 patients’ images were randomly selected from the training set during each epoch as the validation set. The autosegmentation task of the left and right lungs, spinal cord, heart, trachea and esophagus was implemented, and the whole training time was approximately 5 hours. Geometric evaluation metrics including the Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95) and average surface distance (ASD), were used to assess the autosegmentation performance of OARs based on the proposed model and were compared with those based on U-Net as benchmarks. Then, two sets of treatment plans were optimized based on the manually contoured targets and OARs (Plan1), as well as the manually contours targets and the automatically contoured OARs (Plan2). Dosimetric parameters, including Dmax, Dmean and Vx, of OARs were obtained and compared.

Results

The DSC, HD95 and ASD of the proposed model were better than those of U-Net. The differences in the DSC of the spinal cord and esophagus, differences in the HD95 of the spinal cord, heart, trachea and esophagus, as well as differences in the ASD of the spinal cord were statistically significant between the two models (P<0.05). The differences in the dose-volume parameters of the two sets of plans were not statistically significant (P>0.05). Moreover, compared with manual segmentation, autosegmentation significantly reduced the contouring time by nearly 40.7% (P<0.05).

Conclusions

The bilateral lungs, spinal cord, heart and trachea could be accurately delineated using the proposed model in this study; however, the automatic segmentation effect of the esophagus must still be further improved. The concept of feature map reuse provides a new idea for automatic medical image segmentation.