Practical guidelines for handling head and neck computed tomography artifacts for quantitative image analysis

The Promise and Future of Radiomics for Personalized Radiotherapy Dosing and Adaptation

Quantitative image analysis, also known as radiomics, aims to analyze large-scale quantitative features extracted from acquired medical images using hand-crafted or machine-engineered feature extraction approaches. Radiomics has great potential for a variety of clinical applications in radiation oncology, an image-rich treatment modality that utilizes computed tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET) for treatment planning, dose calculation, and image guidance. A promising application of radiomics is in predicting treatment outcomes after radiotherapy such as local control and treatment-related toxicity using features extracted from pretreatment and on-treatment images. Based on these individualized predictions of treatment outcomes, radiotherapy dose can be sculpted to meet the specific needs and preferences of each patient. Radiomics can aid in tumor characterization for personalized targeting, especially for identifying high-risk regions within a tumor that cannot be easily discerned based on size or intensity alone. Radiomics-based treatment response prediction can aid in developing personalized fractionation and dose adjustments. In order to make radiomics models more applicable across different institutions with varying scanners and patient populations, further efforts are needed to harmonize and standardize the acquisition protocols by minimizing uncertainties within the imaging data.

Radiomics in veterinary medicine: Overview, methods, and applications

Radiomics, or quantitative image analysis from radiographic image data, borrows the suffix from other emerging -omics fields of study, such as genomics, proteomics, and metabolomics. This report provides an overview of the general principles of how radiomic features are computed, describes major types of morphological, first order, and texture features, and the applications, challenges, and opportunities of radiomics as applied in veterinary medicine. Some advantages radiomics has over traditional semantic radiological features include standardized methodology in computing semantic features, the ability to compute features in multi-dimensional images, their newfound associations with genomic and pathological abnormalities, and the number of perceptible and imperceptible features available for regression or classification modeling. Some challenges in deploying radiomics in a clinical setting include sensitivity to image acquisition settings and image artifacts, pre- and post-image reconstruction and calculation settings, variability in feature estimates stemming from inter- and intra-observer contouring errors, and challenges with software and data harmonization and generalizability of findings given the challenges of small sample size and patient selection bias in veterinary medicine. Despite this, radiomics has enormous potential in patient-centric diagnostics, prognosis, and theragnostics. Fully leveraging the utility of radiomics in veterinary medicine will require inter-institutional collaborations, data harmonization, and data sharing strategies amongst institutions, transparent and robust model development, and multi-disciplinary efforts within and outside the veterinary medical imaging community.

Graph-Based Motion Artifacts Detection Method from Head Computed Tomography Images

Computed tomography (CT) images play an important role due to effectiveness and accessibility, however, motion artifacts may obscure or simulate pathology and dramatically degrade the diagnosis accuracy. In recent years, convolutional neural networks (CNNs) have achieved state-of-the-art performance in medical imaging due to the powerful learning ability with the help of the advanced hardware technology. Unfortunately, CNNs have significant overhead on memory usage and computational resources and are labeled 'black-box' by scholars for their complex underlying structures. To this end, an interpretable graph-based method has been proposed for motion artifacts detection from head CT images in this paper. From a topological perspective, the artifacts detection problem has been reformulated as a complex network classification problem based on the network topological characteristics of the corresponding complex networks. A motion artifacts detection method based on complex networks (MADM-CN) has been proposed. Firstly, the graph of each CT image is constructed based on the theory of complex networks. Secondly, slice-to-slice relationship has been explored by multiple graph construction. In addition, network topological characteristics are investigated locally and globally, consistent topological characteristics including average degree, average clustering coefficient have been utilized for classification. The experimental results have demonstrated that the proposed MADM-CN has achieved better performance over conventional machine learning and deep learning methods on a real CT dataset, reaching up to 98% of the accuracy and 97% of the sensitivity.

Recurrent Convolutional Neural Networks for 3D Mandible Segmentation in Computed Tomography

Purpose: Classic encoder–decoder-based convolutional neural network (EDCNN) approaches cannot accurately segment detailed anatomical structures of the mandible in computed tomography (CT), for instance, condyles and coronoids of the mandible, which are often affected by noise and metal artifacts. The main reason is that EDCNN approaches ignore the anatomical connectivity of the organs. In this paper, we propose a novel CNN-based 3D mandible segmentation approach that has the ability to accurately segment detailed anatomical structures. Methods: Different from the classic EDCNNs that need to slice or crop the whole CT scan into 2D slices or 3D patches during the segmentation process, our proposed approach can perform mandible segmentation on complete 3D CT scans. The proposed method, namely, RCNNSeg, adopts the structure of the recurrent neural networks to form a directed acyclic graph in order to enable recurrent connections between adjacent nodes to retain their connectivity. Each node then functions as a classic EDCNN to segment a single slice in the CT scan. Our proposed approach can perform 3D mandible segmentation on sequential data of any varied lengths and does not require a large computation cost. The proposed RCNNSeg was evaluated on 109 head and neck CT scans from a local dataset and 40 scans from the PDDCA public dataset. The final accuracy of the proposed RCNNSeg was evaluated by calculating the Dice similarity coefficient (DSC), average symmetric surface distance (ASD), and 95% Hausdorff distance (95HD) between the reference standard and the automated segmentation. Results: The proposed RCNNSeg outperforms the EDCNN-based approaches on both datasets and yields superior quantitative and qualitative performances when compared to the state-of-the-art approaches on the PDDCA dataset. The proposed RCNNSeg generated the most accurate segmentations with an average DSC of 97.48%, ASD of 0.2170 mm, and 95HD of 2.6562 mm on 109 CT scans, and an average DSC of 95.10%, ASD of 0.1367 mm, and 95HD of 1.3560 mm on the PDDCA dataset. Conclusions: The proposed RCNNSeg method generated more accurate automated segmentations than those of the other classic EDCNN segmentation techniques in terms of quantitative and qualitative evaluation. The proposed RCNNSeg has potential for automatic mandible segmentation by learning spatially structured information.

Reproducibility of radiomic features using network analysis and its application in Wasserstein k-means clustering

Purpose: The goal of this study is to develop innovative methods for identifying radiomic features that are reproducible over varying image acquisition settings. Approach: We propose a regularized partial correlation network to identify reliable and reproducible radiomic features. This approach was tested on two radiomic feature sets generated using two different reconstruction methods on computed tomography (CT) scans from a cohort of 47 lung cancer patients. The largest common network component between the two networks was tested on phantom data consisting of five cancer samples. To further investigate whether radiomic features found can identify phenotypes, we propose a k -means clustering algorithm coupled with the optimal mass transport theory. This approach following the regularized partial correlation network analysis was tested on CT scans from 77 head and neck squamous cell carcinoma (HNSCC) patients in the Cancer Imaging Archive (TCIA) and validated using an independent dataset. Results: A set of common radiomic features was found in relatively large network components between the resultant two partial correlation networks resulting from a cohort of lung cancer patients. The reliability and reproducibility of those radiomic features were further validated on phantom data using the Wasserstein distance. Further analysis using the network-based Wasserstein k -means algorithm on the TCIA HNSCC data showed that the resulting clusters separate tumor subsites as well as HPV status, and this was validated on an independent dataset. Conclusion: We showed that a network-based analysis enables identifying reproducible radiomic features and use of the selected set of features can enhance clustering results.

Automated detection of dental artifacts for large-scale radiomic analysis in radiation oncology

Background and purpose

Computed tomography (CT) is one of the most common medical imaging modalities in radiation oncology and radiomics research, the computational voxel-level analysis of medical images. Radiomics is vulnerable to the effects of dental artifacts (DA) caused by metal implants or fillings and can hamper future reproducibility on new datasets. In this study we seek to better understand the robustness of quantitative radiomic features to DAs. Furthermore, we propose a novel method of detecting DAs in order to safeguard radiomic studies and improve reproducibility.

Materials and methods

We analyzed the correlations between radiomic features and the location of dental artifacts in a new dataset containing 3D CT scans from 3211 patients. We then combined conventional image processing techniques with a pre-trained convolutional neural network to create a three-class patient-level DA classifier and slice-level DA locator. Finally, we demonstrated its utility in reducing the correlations between the location of DAs and certain radiomic features.

Results

We found that when strong DAs were present, the proximity of the tumour to the mouth was highly correlated with 36 radiomic features. We predicted the correct DA magnitude yielding a Matthews correlation coefficient of 0.73 and location of DAs achieving the same level of agreement as human labellers.

Conclusions

Removing radiomic features or CT slices containing DAs could reduce the unwanted correlations between the location of DAs and radiomic features. Automated DA detection can be used to improve the reproducibility of radiomic studies; an important step towards creating effective radiomic models for use in clinical radiation oncology.

Development and dosimetric assessment of an automatic dental artifact classification tool to guide artifact management techniques in a fully automated treatment planning workflow

Purpose:

We conducted our study to develop a tool capable of automatically detecting dental artifacts in a CT scan on a slice-by-slice basis and to assess the dosimetric impact of implementing the tool into the Radiation Planning Assistant (RPA), a web-based platform designed to fully automate the radiation therapy treatment planning process.

Methods:

We developed an automatic dental artifact identification tool and assessed the dosimetric impact of its use in the RPA. Three users manually annotated 83,676 head-and-neck (HN) CT slices (549 patients). Majority-voting was applied to the individual annotations to determine the presence or absence of dental artifacts. The patients were divided into train, cross-validation, and test data sets (ratio: 3:1:1, respectively). A random subset of images without dental artifacts was used to balance classes (1:1) in the training data set. The Inception-V3 deep learning model was trained with the binary cross-entropy loss function. With use of this model, we automatically identified artifacts on 15 RPA HN plans on a slice-by-slice basis and investigated three dental artifact management methods applied before and after volumetric modulated arc therapy (VMAT) plan optimization. The resulting dose distributions and target coverage were quantified.

Results:

Per-slice accuracy, sensitivity, and specificity were 99 %, 91 %, and 99 %, respectively. The model identified all patients with artifacts. Small dosimetric differences in total plan dose were observed between the various density-override methods (±1 Gy). For the pre- and post-optimized plans, 90 % and 99 %, respectively, of dose comparisons resulted in normal structure dose differences of ±1 Gy. Differences in the volume of structures receiving 95 % of the prescribed dose (V95[%]) were ≤0.25 % for 100 % of plans.

Conclusion:

The dosimetric impact of applying dental artifact management before and after artifact plan optimization was minor. Our results suggest that not accounting for dental artifacts in the current RPA workflow (where only post-optimization dental artifact management is possible) may result in minor dosimetric differences. If RPA users choose to override CT densities as a solution to managing dental artifacts, our results suggest segmenting the volume of the artifact and overriding its density to water is a safe option.

Diagnostic Accuracy and Failure Mode Analysis of a Deep Learning Algorithm for the Detection of Intracranial Hemorrhage

OBJECTIVE

To determine the institutional diagnostic accuracy of an artificial intelligence (AI) decision support systems (DSS), Aidoc, in diagnosing intracranial hemorrhage (ICH) on noncontrast head CTs and to assess the potential generalizability of an AI DSS.

METHODS

This retrospective study included 3,605 consecutive, emergent, adult noncontrast head CT scans performed between July 1, 2019, and December 30, 2019, at our institution (51% female subjects, mean age of 61 ± 21 years). Each scan was evaluated for ICH by both a certificate of added qualification certified neuroradiologist and Aidoc. We determined the diagnostic accuracy of the AI model and performed a failure mode analysis with quantitative CT radiomic image characterization.

RESULTS

Of the 3,605 scans, 349 cases of ICH (9.7% of studies) were identified. The neuroradiologist and Aidoc interpretations were concordant in 96.9% of cases and the overall sensitivity, specificity, positive predictive value, and negative predictive value were 92.3%, 97.7%, 81.3%, and 99.2%, respectively, with positive predictive values unexpectedly lower than in previously reported studies. Prior neurosurgery, type of ICH, and number of ICHs were significantly associated with decreased model performance. Quantitative image characterization with CT radiomics failed to reveal significant differences between concordant and discordant studies.

DISCUSSION

This study revealed decreased diagnostic accuracy of an AI DSS at our institution. Despite extensive evaluation, we were unable to identify the source of this discrepancy, raising concerns about the generalizability of these tools with indeterminate failure modes. These results further highlight the need for standardized study design to allow for rigorous and reproducible site-to-site comparison of emerging deep learning technologies.

Utility of CT texture analysis to differentiate olfactory neuroblastoma from sinonasal squamous cell carcinoma

The purpose of this study was to examine differences in texture features between olfactory neuroblastoma (ONB) and sinonasal squamous cell carcinoma (SCC) on contrast-enhanced CT (CECT) images, and to evaluate the predictive accuracy of texture analysis compared to radiologists’ interpretations. Forty-three patients with pathologically-diagnosed primary nasal and paranasal tumor (17 ONB and 26 SCC) were included. We extracted 42 texture features from tumor regions on CECT images obtained before treatment. In univariate analysis, each texture features were compared, with adjustment for multiple comparisons. In multivariate analysis, the elastic net was used to select useful texture features and to construct a texture-based prediction model with leave-one-out cross-validation. The prediction accuracy was compared with two radiologists’ visual interpretations. In univariate analysis, significant differences were observed for 28 of 42 texture features between ONB and SCC, with areas under the receiver operating characteristic curve between 0.68 and 0.91 (median: 0.80). In multivariate analysis, the elastic net model selected 18 texture features that contributed to differentiation. It tended to show slightly higher predictive accuracy than radiologists’ interpretations (86% and 74%, respectively; P = 0.096). In conclusion, several texture features contributed to differentiation of ONB from SCC, and the texture-based prediction model was considered useful.

Interpretable Machine Learning Model for Locoregional Relapse Prediction in Oropharyngeal Cancers

BACKGROUND

There is no evidence to support surgery or radiotherapy as the best treatment for resectable oropharyngeal cancers with a negative HPV status. Predictive algorithms may help to decide which strategy to choose, but they will only be accepted by caregivers and European authorities if they are interpretable. As a proof of concept, we developed a predictive and interpretable algorithm to predict locoregional relapse at 18 months for oropharyngeal cancers as a first step towards that goal.

METHODS

The model was based on clinical and Pyradiomics features extracted from the dosimetric CT scan. Intraclass correlation was used to filter out features dependant on delineation. Correlated redundant features were also removed. An XGBoost model was cross-validated and optimised on the HN1 cohort (79 patients), and performances were assessed on the ART ORL cohort (45 patients). The Shapley Values were used to provide an overall and local explanation of the model.

RESULTS

On the ART ORL cohort, the model trained on HN1 yielded a precision-or predictive positive value-of 0.92, a recall of 0.42, an area under the curve of the receiver operating characteristic of 0.68 and an accuracy of 0.64. The most contributory features were shape Voxel Volume, grey level size zone matrix Small Area Emphasis (glszmSAE), gldm Dependence Non Uniformity Normalized (gldmDNUN), Sex and Age.

CONCLUSIONS

We developed an interpretable and generalizable model that could yield a good precision-positive predictive value-for relapse at 18 months on a different test cohort.

Prediction of post-radiotherapy locoregional progression in HPV-associated oropharyngeal squamous cell carcinoma using machine-learning analysis of baseline PET/CT radiomics

•

Radiomics quantitatively captures visually inappreciable imaging features.

•

PET/CT radiomics provides wholistic metabolic and structural tumor characterization.

•

Machine-learning algorithms can generate radiomics-based biomarkers for OPSCC.

•

PET/CT radiomics can predict post-radiotherapy locoregional progression in HPV-associated OPSCC.

•

Such biomarkers may improve patient selection for treatment de-intensification trials.

Locoregional failure remains a therapeutic challenge in oropharyngeal squamous cell carcinoma (OPSCC). We aimed to devise novel objective imaging biomarkers for prediction of locoregional progression in HPV-associated OPSCC. Following manual lesion delineation, 1037 PET and 1037 CT radiomic features were extracted from each primary tumor and metastatic cervical lymph node on baseline PET/CT scans. Applying random forest machine-learning algorithms, we generated radiomic models for censoring-aware locoregional progression prognostication (evaluated by Harrell's C-index) and risk stratification (evaluated in Kaplan-Meier analysis). A total of 190 patients were included; an optimized model yielded a median (interquartile range) C-index of 0.76 (0.66-0.81; p = 0.01) in prognostication of locoregional progression, using combined PET/CT radiomic features from primary tumors. Radiomics-based risk stratification reliably identified patients at risk for locoregional progression within 2-, 3-, 4-, and 5-year follow-up intervals, with log-rank p-values of p = 0.003, p = 0.001, p = 0.02, p = 0.006 in Kaplan-Meier analysis, respectively. Our results suggest PET/CT radiomic biomarkers can predict post-radiotherapy locoregional progression in HPV-associated OPSCC. Pending validation in large, independent cohorts, such objective biomarkers may improve patient selection for treatment de-intensification trials in this prognostically favorable OPSCC entity, and eventually facilitate personalized therapy.

A radiomics platform for computing imaging features from µCT images of Thoroughbred racehorse proximal sesamoid bones: Benchmark performance and evaluation

BACKGROUND

Proximal sesamoid bone (PSB) fractures are the most common fatal musculoskeletal injury in North American racehorses. Computed tomography has the potential to detect morphological changes in bone structure but can be challenging to analyse reliably and quantitatively.

OBJECTIVES

To develop a radiomics platform that allows the comparison of features from micro-CTs (µCT) of PSBs in horses that sustained catastrophic fractures with horses that did not. To compare features calculated with a radiomics approach with features calculated from a previously published study that used quantitative µCT in the same specimens.

STUDY DESIGN

Retrospective study of cadaver specimens of µCT images of PSBs using prospectively applied radiomics.

METHODS

Radiomics features were computed on standardised CT datasets to benchmark the software. Features from µCT images of PSBs from eight horses that sustained PSB fracture and eight controls were computed using the contralateral, intact forelimb from horses sustaining PSB fracture (cases, n = 19) and all available forelimbs for controls (n = 30). Two-hundred and fifteen radiomic features were calculated, and similar or comparable features were compared with those reported in a previous study that used the same specimens.

RESULTS

Morphologic features computed with the radiomics approach, such as volume, minor axis dimensions and anisotropy were highly correlated with previously published data. A high number of imperceptible radiomic features, such as entropy, coarseness and histogram features were also found to be significantly different (P < .01). The extent of the differences in image features for the cases and controls PSBs depends on radiomic calculation settings.

MAIN LIMITATIONS

Only datasets obtained from cadaver specimens were included in the study.

CONCLUSIONS

A radiomics approach for analysing µCT images of PSBs was able to identify and reproduce differences in image features in cases and controls. Furthermore, radiomics revealed many more imperceptible image features between cases and control PSBs.

Potential Added Value of PET/CT Radiomics for Survival Prognostication beyond AJCC 8th Edition Staging in Oropharyngeal Squamous Cell Carcinoma

Accurate risk-stratification can facilitate precision therapy in oropharyngeal squamous cell carcinoma (OPSCC). We explored the potential added value of baseline positron emission tomography (PET)/computed tomography (CT) radiomic features for prognostication and risk stratification of OPSCC beyond the American Joint Committee on Cancer (AJCC) 8th edition staging scheme. Using institutional and publicly available datasets, we included OPSCC patients with known human papillomavirus (HPV) status, without baseline distant metastasis and treated with curative intent. We extracted 1037 PET and 1037 CT radiomic features quantifying lesion shape, imaging intensity, and texture patterns from primary tumors and metastatic cervical lymph nodes. Utilizing random forest algorithms, we devised novel machine-learning models for OPSCC progression-free survival (PFS) and overall survival (OS) using “radiomics” features, “AJCC” variables, and the “combined” set as input. We designed both single- (PET or CT) and combined-modality (PET/CT) models. Harrell’s C-index quantified survival model performance; risk stratification was evaluated in Kaplan–Meier analysis. A total of 311 patients were included. In HPV-associated OPSCC, the best “radiomics” model achieved an average C-index ± standard deviation of 0.62 ± 0.05 (p = 0.02) for PFS prediction, compared to 0.54 ± 0.06 (p = 0.32) utilizing “AJCC” variables. Radiomics-based risk-stratification of HPV-associated OPSCC was significant for PFS and OS. Similar trends were observed in HPV-negative OPSCC. In conclusion, radiomics imaging features extracted from pre-treatment PET/CT may provide complimentary information to the current AJCC staging scheme for survival prognostication and risk-stratification of HPV-associated OPSCC.

Radiomic analysis identifies tumor subtypes associated with distinct molecular and microenvironmental factors in head and neck squamous cell carcinoma

PURPOSE

To identify whether radiomic features from pre-treatment computed tomography (CT) scans can predict molecular differences between head and neck squamous cell carcinoma (HNSCC) using The Cancer Imaging Archive (TCIA) and The Cancer Genome Atlas (TCGA).

METHODS

77 patients from the TCIA with HNSCC had imaging suitable for analysis. Radiomic features were extracted and unsupervised consensus clustering was performed to identify subtypes. Genomic data was extracted from the matched patients in the TCGA database. We explored relationships between radiomic features and molecular profiles of tumors, including the tumor immune microenvironment. A machine learning method was used to build a model predictive of CD8 + T-cells. An independent cohort of 83 HNSCC patients was used to validate the radiomic clusters.

RESULTS

We initially extracted 104 two-dimensional radiomic features, and after feature stability tests and removal of volume dependent features, reduced this to 67 features for subsequent analysis. Consensus clustering based on these features resulted in two distinct clusters. The radiomic clusters differed by primary tumor subsite (p = 0.0096), HPV status (p = 0.0127), methylation-based clustering results (p = 0.0025), and tumor immune microenvironment. A random forest model using radiomic features predicted CD8 + T-cells independent of HPV status with R² = 0.30 (p < 0.0001) on cross validation. Consensus clustering on the validation cohort resulted in two distinct clusters that differ in tumor subsite (p = 1.3 × 10^-7) and HPV status (p = 4.0 × 10^-7).

CONCLUSION

Radiomic analysis can identify biologic features of tumors such as HPV status and T-cell infiltration and may be able to provide other information in the near future to help with patient stratification.

Radiomic biomarkers for head and neck squamous cell carcinoma

Tumor heterogeneity is a well-known prognostic factor in head and neck squamous cell carcinoma (HNSCC). A major limitation of tissue- and blood-derived tumor markers is the lack of spatial resolution to image tumor heterogeneity. Tissue markers derived from tumor biopsies usually represent only a small tumor subregion at a single timepoint and are therefore often not representative of the tumors' biology or the biological alterations during and after treatment. Similarly, liquid biopsies give an overall picture of the tumors' secreted factors but completely lack any spatial resolution. Radiomics has the potential to give complete three-dimensional information about the tumor. We conducted a comprehensive literature search to assess the correlation of radiomics to tumor biology and treatment outcome in HNSCC and to assess current limitations of the radiomic biomarkers. In total, 25 studies that explored the ability of radiomics to predict tumor biology and phenotype in HNSCC and 28 studies that explored radiomics to predict post-treatment events were identified. Out of these 53 studies, only three failed to show a significant correlation. The major technical challenges are currently artifacts due to metal implants, non-standardized contrast injection, and delineation uncertainties. All studies to date were retrospective and none of the above-mentioned radiomics signatures have been validated in an independent cohort using an independent software implementation, which shows that transferability due to the numerous technical challenges is currently a major limitation. However, radiomics is a very young field and these studies hopefully pave the way for clinical implementation of radiomics for HNSCC in the future.

External validation and transfer learning of convolutional neural networks for computed tomography dental artifact classification

Quality assurance of data prior to use in automated pipelines and image analysis would assist in safeguarding against biases and incorrect interpretation of results. Automation of quality assurance steps would further improve robustness and efficiency of these methods, motivating widespread adoption of techniques. Previous work by our group demonstrated the ability of convolutional neural networks (CNN) to efficiently classify head and neck (H&N) computed-tomography (CT) images for the presence of dental artifacts (DA) that obscure visualization of structures and the accuracy of Hounsfield units. In this work we demonstrate the generalizability of our previous methodology by validating CNNs on six external datasets, and the potential benefits of transfer learning with fine-tuning on CNN performance. 2112 H&N CT images from seven institutions were scored as DA positive or negative. 1538 images from a single institution were used to train three CNNs with resampling grid sizes of 64³, 128³ and 256³. The remaining six external datasets were used in five-fold cross-validation with a data split of 20% training/fine-tuning and 80% validation. The three pre-trained models were each validated using the five-folds of the six external datasets. The pre-trained models also underwent transfer learning with fine-tuning using the 20% training/fine-tuning data, and validated using the corresponding validation datasets. The highest micro-averaged AUC for our pre-trained models across all external datasets occurred with a resampling grid of 256³ (AUC  =  0.91  ±  0.01). Transfer learning with fine-tuning improved generalizability when utilizing a resampling grid of 256³ to a micro-averaged AUC of 0.92  ±  0.01. Despite these promising results, transfer learning did not improve AUC when utilizing small resampling grids or small datasets. Our work demonstrates the potential of our previously developed automated quality assurance methods to generalize to external datasets. Additionally, we showed that transfer learning with fine-tuning using small portions of external datasets can be used to fine-tune models for improved performance when large variations in images are present.

Automatic classification of dental artifact status for efficient image veracity checks: effects of image resolution and convolutional neural network depth

Enabling automated pipelines, image analysis and big data methodology in cancer clinics requires thorough understanding of the data. Automated quality assurance steps could improve the efficiency and robustness of these methods by verifying possible data biases. In particular, in head and neck (H&N) computed-tomography (CT) images, dental artifacts (DA) obscure visualization of structures and the accuracy of Hounsfield units; a challenge for image analysis tasks, including radiomics, where poor image quality can lead to systemic biases. In this work we analyze the performance of three-dimensional convolutional neural networks (CNN) trained to classify DA statuses. 1538 patient images were scored by a single observer as DA positive or negative. Stratified five-fold cross validation was performed to train and test CNNs using various isotropic resampling grids (64³, 128³ and 256³), with CNN depths designed to produce 32³, 16³, and 8³ machine generated features. These parameters were selected to determine if more computationally efficient CNNs could be utilized to achieve the same performance. The area under the precision recall curve (PR-AUC) was used to assess CNN performance. The highest PR-AUC (0.92  ±  0.03) was achieved with a CNN depth  =  5, resampling grid  =  256. The CNN performance with 256³ resampling grid size is not significantly better than 64³ and 128³ after 20 epochs, which had PR-AUC  =  0.89  ±  0.03 (p -value  =  0.28) and 0.91  ±  0.02 (p -value  =  0.93) at depths of 3 and 4, respectively. Our experiments demonstrate the potential to automate specific quality assurance tasks required for unbiased and robust automated pipeline and image analysis research. Additionally, we determined that there is an opportunity to simplify CNNs with smaller resampling grids to make the process more amenable to very large datasets that will be available in the future.

Combined CT radiomics of primary tumor and metastatic lymph nodes improves prediction of loco-regional control in head and neck cancer

Loco-regional control (LRC) is a major clinical endpoint after definitive radiochemotherapy (RCT) of head and neck cancer (HNC). Radiomics has been shown a promising biomarker in cancer research, however closer related to primary tumor control than composite endpoints. Radiomics studies often focus on the analysis of primary tumor (PT). We hypothesize that the combination of PT and lymph nodes (LN) radiomics better predicts LRC in HNC treated with RCT. Radiomics analysis was performed in CT images of 128 patients using Z-Rad implementation (training n = 77, validation n = 51). 285 features were extracted from PT and involved LN. Features were preselected with the maximum relevance minimum redundancy method and the multivariate Cox model was trained using least absolute shrinkage and selection operator. The mixed model was based on the combination of PT and LN radiomics, whereas the PT model included only the PT features. The mixed model showed significantly higher performance than the PT model (p < 0.01), c-index of 0.67 and 0.63, respectively; and better risk group stratification. The clinical nodal status was not a significant predictor in the combination with PT radiomics. This study shows that the LRC can be better predicted by expansion of radiomics analysis with LN features.

Radiomics features of the primary tumor fail to improve prediction of overall survival in large cohorts of CT- and PET-imaged head and neck cancer patients

Radiomics studies require many patients in order to power them, thus patients are often combined from different institutions and using different imaging protocols. Various studies have shown that imaging protocols affect radiomics feature values. We examined whether using data from cohorts with controlled imaging protocols improved patient outcome models. We retrospectively reviewed 726 CT and 686 PET images from head and neck cancer patients, who were divided into training or independent testing cohorts. For each patient, radiomics features with different preprocessing were calculated and two clinical variables—HPV status and tumor volume—were also included. A Cox proportional hazards model was built on the training data by using bootstrapped Lasso regression to predict overall survival. The effect of controlled imaging protocols on model performance was evaluated by subsetting the original training and independent testing cohorts to include only patients whose images were obtained using the same imaging protocol and vendor. Tumor volume, HPV status, and two radiomics covariates were selected for the CT model, resulting in an AUC of 0.72. However, volume alone produced a higher AUC, whereas adding radiomics features reduced the AUC. HPV status and one radiomics feature were selected as covariates for the PET model, resulting in an AUC of 0.59, but neither covariate was significantly associated with survival. Limiting the training and independent testing to patients with the same imaging protocol reduced the AUC for CT patients to 0.55, and no covariates were selected for PET patients. Radiomics features were not consistently associated with survival in CT or PET images of head and neck patients, even within patients with the same imaging protocol.

Predicting survival and local control after radiochemotherapy in locally advanced head and neck cancer by means of computed tomography based radiomics

PURPOSE

To appraise the ability of a radiomics signature to predict clinical outcome after definitive radiochemotherapy (RCT) of stage III-IV head and neck cancer.

METHODS

A cohort of 110 patients was included in a retrospective analysis. Radiomics texture features were extracted from the gross tumor volumes contoured on planning computed tomography (CT) images. The cohort of patients was randomly divided into a training (70 patients) and a validation (40 patients) cohorts. Textural features were correlated to survival and control data to build predictive models. All the significant predictors of the univariate analysis were included in a multivariate model. The quality of the models was appraised by means of the concordance index (CI).

RESULTS

A signature with 3 features was identified as predictive of overall survival (OS) with CI = 0.88 and 0.90 for the training and validation cohorts, respectively. A signature with 2 features was identified for progression-free survival (PFS; CI = 0.72 and 0.80); 2 features also characterized the signature for local control (LC; CI = 0.72 and 0.82). In all cases, the stratification in high- and low-risk groups for the training and validation cohorts led to significant differences in the actuarial curves. In the validation cohort the mean OS times (in months) were 78.9 ± 2.1 vs 67.4 ± 6.0 in the low- and high-risk groups, respectively, the PFS was 73.1 ± 3.7 and 50.7 ± 7.2, while the LC was 78.7 ± 2.1 and 63.9 ± 6.5.

CONCLUSION

CT-based radiomic signatures that correlate with survival and control after RCT were identified and allow low- and high-risk groups of patients to be identified.

Automatic recognition and analysis of metal streak artifacts in head and neck computed tomography for radiomics modeling

Background and purpose

Computed tomography (CT) radiomics of head and neck cancer (HNC) images is susceptible to dental implant artifacts. This work devised and validated an automated algorithm to detect CT metal artifacts and investigate their impact on subsequent radiomics analyses. A new method based on features from total variation, gradient directional distribution, and Hough transform was developed and evaluated.

Materials and methods

Two HNC datasets were analyzed: a training set of 131 patients for developing the detection algorithm and a testing set of 220 patients. Seven designated features were extracted from ROIs (regions of interest) and machine learning with random forests was used for building the artifact detection algorithm. Performance was assessed using the area under the receiver operating characteristics curve (AUC).

Results

The testing results of artifacts detection yielded a cross-validated AUC of 0.91 (95% CI: 0.89–0.94), and a test AUC of 0.89. External testing validation yielded an accuracy of 0.82. For radiomics model prediction, training with artifacts yielded an AUC of 0.64 (95% CI: 0.63–0.65), while training on images without artifacts improved the AUC to 0.75 (95% CI: 0.74–0.76). This was compared to visual inspection of artifacts (AUC = 0.71 [95% CI: 0.69–0.73]).

Conclusion

We developed a new method for automated and efficient detection of streak artifacts. We also showed that such streak artifacts in HNC CT images can worsen the performance of radiomics modeling.