The use of deep learning in interventional radiotherapy (brachytherapy): A review with a focus on open source and open data

Deep learning advanced to one of the most important technologies in almost all medical fields. Especially in areas, related to medical imaging it plays a big role. However, in interventional radiotherapy (brachytherapy) deep learning is still in an early phase. In this review, first, we investigated and scrutinised the role of deep learning in all processes of interventional radiotherapy and directly related fields. Additionally, we summarised the most recent developments. For better understanding, we provide explanations of key terms and approaches to solving common deep learning problems. To reproduce results of deep learning algorithms both source code and training data must be available. Therefore, a second focus of this work is on the analysis of the availability of open source, open data and open models. In our analysis, we were able to show that deep learning plays already a major role in some areas of interventional radiotherapy, but is still hardly present in others. Nevertheless, its impact is increasing with the years, partly self-propelled but also influenced by closely related fields. Open source, data and models are growing in number but are still scarce and unevenly distributed among different research groups. The reluctance in publishing code, data and models limits reproducibility and restricts evaluation to mono-institutional datasets. The conclusion of our analysis is that deep learning can positively change the workflow of interventional radiotherapy but there is still room for improvements when it comes to reproducible results and standardised evaluation methods.

Development and evaluation of two open-source nnU-Net models for automatic segmentation of lung tumors on PET and CT images with and without respiratory motion compensation

OBJECTIVES

In lung cancer, one of the main limitations for the optimal integration of the biological and anatomical information derived from Positron Emission Tomography (PET) and Computed Tomography (CT) is the time and expertise required for the evaluation of the different respiratory phases. In this study, we present two open-source models able to automatically segment lung tumors on PET and CT, with and without motion compensation.

MATERIALS AND METHODS

This study involved time-bin gated (4D) and non-gated (3D) PET/CT images from two prospective lung cancer cohorts (Trials 108237 and 108472) and one retrospective. For model construction, the ground truth (GT) was defined by consensus of two experts, and the nnU-Net with 5-fold cross-validation was applied to 560 4D-images for PET and 100 3D-images for CT. The test sets included 270 4D- images and 19 3D-images for PET and 80 4D-images and 27 3D-images for CT, recruited at 10 different centres.

RESULTS

In the performance evaluation with the multicentre test sets, the Dice Similarity Coefficients (DSC) obtained for our PET model were DSC(4D-PET) = 0.74 ± 0.06, improving 19% relative to the DSC between experts and DSC(3D-PET) = 0.82 ± 0.11. The performance for CT was DSC(4D-CT) = 0.61 ± 0.28 and DSC(3D-CT) = 0.63 ± 0.34, improving 4% and 15% relative to DSC between experts.

CONCLUSIONS

Performance evaluation demonstrated that the automatic segmentation models have the potential to achieve accuracy comparable to manual segmentation and thus hold promise for clinical application. The resulting models can be freely downloaded and employed to support the integration of 3D- or 4D- PET/CT and to facilitate the evaluation of its impact on lung cancer clinical practice.

CLINICAL RELEVANCE STATEMENT

We provide two open-source nnU-Net models for the automatic segmentation of lung tumors on PET/CT to facilitate the optimal integration of biological and anatomical information in clinical practice. The models have superior performance compared to the variability observed in manual segmentations by the different experts for images with and without motion compensation, allowing to take advantage in the clinical practice of the more accurate and robust 4D-quantification.

KEY POINTS

Lung tumor segmentation on PET/CT imaging is limited by respiratory motion and manual delineation is time consuming and suffer from inter- and intra-variability. Our segmentation models had superior performance compared to the manual segmentations by different experts. Automating PET image segmentation allows for easier clinical implementation of biological information.

Deep learning based automated delineation of the intraprostatic gross tumour volume in PSMA-PET for patients with primary prostate cancer

PURPOSE

With the increased use of focal radiation dose escalation for primary prostate cancer (PCa), accurate delineation of gross tumor volume (GTV) in prostate-specific membrane antigen PET (PSMA-PET) becomes crucial. Manual approaches are time-consuming and observer dependent. The purpose of this study was to create a deep learning model for the accurate delineation of the intraprostatic GTV in PSMA-PET.

METHODS

A 3D U-Net was trained on 128 different 18F-PSMA-1007 PET images from three different institutions. Testing was done on 52 patients including one independent internal cohort (Freiburg: n = 19) and three independent external cohorts (Dresden: n = 14 18F-PSMA-1007, Boston: Massachusetts General Hospital (MGH): n = 9 18F-DCFPyL-PSMA and Dana-Farber Cancer Institute (DFCI): n = 10 68Ga-PSMA-11). Expert contours were generated in consensus using a validated technique. CNN predictions were compared to expert contours using Dice similarity coefficient (DSC). Co-registered whole-mount histology was used for the internal testing cohort to assess sensitivity/specificity.

RESULTS

Median DSCs were Freiburg: 0.82 (IQR: 0.73-0.88), Dresden: 0.71 (IQR: 0.53-0.75), MGH: 0.80 (IQR: 0.64-0.83) and DFCI: 0.80 (IQR: 0.67-0.84), respectively. Median sensitivity for CNN and expert contours were 0.88 (IQR: 0.68-0.97) and 0.85 (IQR: 0.75-0.88) (p = 0.40), respectively. GTV volumes did not differ significantly (p > 0.1 for all comparisons). Median specificity of 0.83 (IQR: 0.57-0.97) and 0.88 (IQR: 0.69-0.98) were observed for CNN and expert contours (p = 0.014), respectively. CNN prediction took 3.81 seconds on average per patient.

CONCLUSION

The CNN was trained and tested on internal and external datasets as well as histopathology reference, achieving a fast GTV segmentation for three PSMA-PET tracers with high diagnostic accuracy comparable to manual experts.

Interobserver agreement on definition of the target volume in stereotactic radiotherapy for pancreatic adenocarcinoma using different imaging modalities

PURPOSE

The aim of this study was to evaluate interobserver agreement (IOA) on target volume definition for pancreatic cancer (PACA) within the Radiosurgery and Stereotactic Radiotherapy Working Group of the German Society of Radiation Oncology (DEGRO) and to identify the influence of imaging modalities on the definition of the target volumes.

METHODS

Two cases of locally advanced PACA and one local recurrence were selected from a large SBRT database. Delineation was based on either a planning 4D CT with or without (w/wo) IV contrast, w/wo PET/CT, and w/wo diagnostic MRI. Novel compared to other studies, a combination of four metrics was used to integrate several aspects of target volume segmentation: the Dice coefficient (DSC), the Hausdorff distance (HD), the probabilistic distance (PBD), and the volumetric similarity (VS).

RESULTS

For all three GTVs, the median DSC was 0.75 (range 0.17-0.95), the median HD 15 (range 3.22-67.11) mm, the median PBD 0.33 (range 0.06-4.86), and the median VS was 0.88 (range 0.31-1). For ITVs and PTVs the results were similar. When comparing the imaging modalities for delineation, the best agreement for the GTV was achieved using PET/CT, and for the ITV and PTV using 4D PET/CT, in treatment position with abdominal compression.

CONCLUSION

Overall, there was good GTV agreement (DSC). Combined metrics appeared to allow a more valid detection of interobserver variation. For SBRT, either 4D PET/CT or 3D PET/CT in treatment position with abdominal compression leads to better agreement and should be considered as a very useful imaging modality for the definition of treatment volumes in pancreatic SBRT. Contouring does not appear to be the weakest link in the treatment planning chain of SBRT for PACA.

Investigation and benchmarking of U-Nets on prostate segmentation tasks

In healthcare, a growing number of physicians and support staff are striving to facilitate personalized radiotherapy regimens for patients with prostate cancer. This is because individual patient biology is unique, and employing a single approach for all is inefficient. A crucial step for customizing radiotherapy planning and gaining fundamental information about the disease, is the identification and delineation of targeted structures. However, accurate biomedical image segmentation is time-consuming, requires considerable experience and is prone to observer variability. In the past decade, the use of deep learning models has significantly increased in the field of medical image segmentation. At present, a vast number of anatomical structures can be demarcated on a clinician's level with deep learning models. These models would not only unload work, but they can offer unbiased characterization of the disease. The main architectures used in segmentation are the U-Net and its variants, that exhibit outstanding performances. However, reproducing results or directly comparing methods is often limited by closed source of data and the large heterogeneity among medical images. With this in mind, our intention is to provide a reliable source for assessing deep learning models. As an example, we chose the challenging task of delineating the prostate gland in multi-modal images. First, this paper provides a comprehensive review of current state-of-the-art convolutional neural networks for 3D prostate segmentation. Second, utilizing public and in-house CT and MR datasets of varying properties, we created a framework for an objective comparison of automatic prostate segmentation algorithms. The framework was used for rigorous evaluations of the models, highlighting their strengths and weaknesses.

18F-FMISO-PET Hypoxia Monitoring for Head-and-Neck Cancer Patients: Radiomics Analyses Predict the Outcome of Chemo-Radiotherapy

Tumor hypoxia is associated with radiation resistance and can be longitudinally monitored by 18F-fluoromisonidazole (18F-FMISO)-PET/CT. Our study aimed at evaluating radiomics dynamics of 18F-FMISO-hypoxia imaging during chemo-radiotherapy (CRT) as predictors for treatment outcome in head-and-neck squamous cell carcinoma (HNSCC) patients. We prospectively recruited 35 HNSCC patients undergoing definitive CRT and longitudinal 18F-FMISO-PET/CT scans at weeks 0, 2 and 5 (W0/W2/W5). Patients were classified based on peritherapeutic variations of the hypoxic sub-volume (HSV) size (increasing/stable/decreasing) and location (geographically-static/geographically-dynamic) by a new objective classification parameter (CP) accounting for spatial overlap. Additionally, 130 radiomic features (RF) were extracted from HSV at W0, and their variations during CRT were quantified by relative deviations (∆RF). Prediction of treatment outcome was considered statistically relevant after being corrected for multiple testing and confirmed for the two 18F-FMISO-PET/CT time-points and for a validation cohort. HSV decreased in 64% of patients at W2 and in 80% at W5. CP distinguished earlier disease progression (geographically-dynamic) from later disease progression (geographically-static) in both time-points and cohorts. The texture feature low grey-level zone emphasis predicted local recurrence with AUCW2 = 0.82 and AUCW5 = 0.81 in initial cohort (N = 25) and AUCW2 = 0.79 and AUCW5 = 0.80 in validation cohort. Radiomics analysis of 18F-FMISO-derived hypoxia dynamics was able to predict outcome of HNSCC patients after CRT.

Radiomics in prostate cancer imaging for a personalized treatment approach - current aspects of methodology and a systematic review on validated studies

Prostate cancer (PCa) is one of the most frequently diagnosed malignancies of men in the world. Due to a variety of treatment options in different risk groups, proper diagnostic and risk stratification is pivotal in treatment of PCa. The development of precise medical imaging procedures simultaneously to improvements in big data analysis has led to the establishment of radiomics - a computer-based method of extracting and analyzing image features quantitatively. This approach bears the potential to assess and improve PCa detection, tissue characterization and clinical outcome prediction. This article gives an overview on the current aspects of methodology and systematically reviews available literature on radiomics in PCa patients, showing its potential for personalized therapy approaches. The qualitative synthesis includes all imaging modalities and focuses on validated studies, putting forward future directions.

Measuring breathing induced oesophageal motion and its dosimetric impact

PURPOSE

Stereotactic body radiation therapy allows for a precise dose delivery. Organ motion bears the risk of undetected high dose healthy tissue exposure. An organ very susceptible to high dose is the oesophagus. Its low contrast on CT and the oblong shape render motion estimation difficult. We tackle this issue by modern algorithms to measure oesophageal motion voxel-wise and estimate motion related dosimetric impacts.

METHODS

Oesophageal motion was measured using deformable image registration and 4DCT of 11 internal and 5 public datasets. Current clinical practice of contouring the organ on 3DCT was compared to timely resolved 4DCT contours. Dosimetric impacts of the motion were estimated by analysing the trajectory of each voxel in the 4D dose distribution. Finally an organ motion model for patient-wise comparisons was built.

RESULTS

Motion analysis showed mean absolute maximal motion amplitudes of 4.55 ± 1.81 mm left-right, 5.29 ± 2.67 mm anterior-posterior and 10.78 ± 5.30 mm superior-inferior. Motion between cohorts differed significantly. In around 50% of the cases the dosimetric passing criteria was violated. Contours created on 3DCT did not cover 14% of the organ for 50% of the respiratory cycle and were around 38% smaller than the union of all 4D contours. The motion model revealed that the maximal motion is not limited to the lower part of the organ. Our results showed motion amplitudes higher than most reported values in the literature and that motion is very heterogeneous across patients.

CONCLUSIONS

Individual motion information should be considered in contouring and planning.

Experimental phantom evaluation to identify robust positron emission tomography (PET) radiomic features

Background

Radiomics analysis usually involves, especially in multicenter and large hospital studies, different imaging protocols for acquisition, reconstruction, and processing of data. Differences in protocols can lead to differences in the quantification of the biomarker distribution, leading to radiomic feature variability. The aim of our study was to identify those radiomic features robust to the different degrading factors in positron emission tomography (PET) studies. We proposed the use of the standardized measurements of the European Association Research Ltd. (EARL) accreditation to retrospectively identify the radiomic features having low variability to the different systems and reconstruction protocols. In addition, we presented a reproducible procedure to identify PET radiomic features robust to PET/CT imaging metal artifacts. In 27 heterogeneous homemade phantoms for which ground truth was accurately defined by CT segmentation, we evaluated the segmentation accuracy and radiomic feature reliability given by the contrast-oriented algorithm (COA) and the 40% threshold PET segmentation. In the comparison of two data sets, robustness was defined by Wilcoxon rank tests, bias was quantified by Bland–Altman (BA) plot analysis, and strong correlations were identified by Spearman correlation test (r > 0.8 and p satisfied multiple test Bonferroni correction).

Results

Forty-eight radiomic features were robust to system, 22 to resolution, 102 to metal artifacts, and 42 to different PET segmentation tools. Overall, only 4 radiomic features were simultaneously robust to all degrading factors. Although both segmentation approaches significantly underestimated the volume with respect to the ground truth, with relative deviations of −62 ± 36% for COA and −50 ± 44% for 40%, radiomic features derived from the ground truth were strongly correlated and/or robust to 98 radiomic features derived from COA and to 102 from 40%.

Conclusion

In multicenter studies, we recommend the analysis of EARL accreditation measurements in order to retrospectively identify the robust PET radiomic features. Furthermore, 4 radiomic features (area under the curve of the cumulative SUV volume histogram, skewness, kurtosis, and gray-level variance derived from GLRLM after application of an equal probability quantization algorithm on the voxels within lesion) were robust to all degrading factors. In addition, the feasibility of 40% and COA segmentations for their use in radiomics analysis has been demonstrated.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40658-021-00390-7.

FDG-PET Radiomics for Response Monitoring in Non-Small-Cell Lung Cancer Treated with Radiation Therapy

Simple Summary

In this study, we strive to identify clinically relevant image feature (IF) changes during chemoradiation in patients with non-small-cell lung cancer (NSCLC) to be able to predict tumor responses in an early stage of treatment. All patients underwent static (3D) and respiratory-gated 4D PET/CT scans before treatment and a 3D scan during or after treatment. Our proposed method rejects IF changes due to intrinsic variability such as noise, resolution and movement through breathing. The IF variability observed across 4D PET is employed as a patient individualized normalization factor to emphasize statistically relevant IF changes during treatment.

Abstract

The aim of this study is to identify clinically relevant image feature (IF) changes during chemoradiation and evaluate their efficacy in predicting treatment response. Patients with non-small-cell lung cancer (NSCLC) were enrolled in two prospective trials (STRIPE, PET-Plan). We evaluated 48 patients who underwent static (3D) and retrospectively-respiratory-gated 4D PET/CT scans before treatment and a 3D scan during or after treatment. Our proposed method rejects IF changes due to intrinsic variability. The IF variability observed across 4D PET is employed as a patient individualized normalization factor to emphasize statistically relevant IF changes during treatment. Predictions of overall survival (OS), local recurrence (LR) and distant metastasis (DM) were evaluated. From 135 IFs, only 17 satisfied the required criteria of being normally distributed across 4D PET and robust between 3D and 4D images. Changes during treatment in the area-under-the-curve of the cumulative standard-uptake-value histogram (δ_AUCCSH) within primary tumor discriminated (AUC = 0.87, Specificity = 0.78) patients with and without LR. The resulted prognostic model was validated with a different segmentation method (AUC = 0.83) and in a different patient cohort (AUC = 0.63). The quantification of tumor FDG heterogeneity by δ_AUCCSH during chemoradiation correlated with the incidence of local recurrence and might be recommended for monitoring treatment response in patients with NSCLC.

Dosimetric Impact of the Positional Imaging Frequency for Hypofractionated Prostate Radiotherapy - A Voxel-by-Voxel Analysis

Background

To investigate deviations between planned and applied treatment doses for hypofractionated prostate radiotherapy and to quantify dosimetric accuracy in dependence of the image guidance frequency.

Methods

Daily diagnostic in-room CTs were carried out in 10 patients in treatment position as image guidance for hypofractionated prostate radiotherapy. Fraction doses were mapped to the planning CTs and recalculated, and applied doses were accumulated voxel-wise using deformable registration. Non-daily imaging schedules were simulated by deriving position correction vectors from individual scans and used to rigidly register the following scans until the next repositioning before dose recalculation and accumulation. Planned and applied doses were compared regarding dose-volume indices and TCP and NTCP values in dependence of the imaging and repositioning frequency.

Results

Daily image-guided repositioning was associated with only negligible deviations of analyzed dose-volume parameters and conformity/homogeneity indices for the prostate, bladder and rectum. Average CTV T did not significantly deviate from the plan values, and rectum NTCPs were highly comparable, while bladder NTCPs were reduced. For non-daily image-guided repositioning, there were significant deviations in the high-dose range from the planned values. Similarly, CTV dose conformity and homogeneity were reduced. While TCPs and rectal NTCPs did not significantly deteriorate for non-daily repositioning, bladder NTCPs appeared falsely diminished in dependence of the imaging frequency.

Conclusion

Using voxel-by-voxel dose accumulation, we showed for the first time that daily image-guided repositioning resulted in only negligible dosimetric deviations for hypofractionated prostate radiotherapy. Regarding dosimetric aberrations for non-daily imaging, daily imaging is required to adequately deliver treatment.

Intraprostatic Tumor Segmentation on PSMA PET Images in Patients with Primary Prostate Cancer with a Convolutional Neural Network

Accurate delineation of the intraprostatic gross tumor volume (GTV) is a prerequisite for treatment approaches in patients with primary prostate cancer (PCa). Prostate-specific membrane antigen PET (PSMA PET) may outperform MRI in GTV detection. However, visual GTV delineation underlies interobserver heterogeneity and is time consuming. The aim of this study was to develop a convolutional neural network (CNN) for automated segmentation of intraprostatic tumor (GTV-CNN) in PSMA PET. Methods: The CNN (3D U-Net) was trained on the ⁶⁸Ga-PSMA PET images of 152 patients from 2 different institutions, and the training labels were generated manually using a validated technique. The CNN was tested on 2 independent internal (cohort 1: ⁶⁸Ga-PSMA PET, n = 18 and cohort 2: ¹⁸F-PSMA PET, n = 19) and 1 external (cohort 3: ⁶⁸Ga-PSMA PET, n = 20) test datasets. Accordance between manual contours and GTV-CNN was assessed with the Dice-Sørensen coefficient (DSC). Sensitivity and specificity were calculated for the 2 internal test datasets (cohort 1: n = 18, cohort 2: n = 11) using whole-mount histology. Results: The median DSCs for cohorts 1-3 were 0.84 (range: 0.32-0.95), 0.81 (range: 0.28-0.93), and 0.83 (range: 0.32-0.93), respectively. Sensitivities and specificities for the GTV-CNN were comparable with manual expert contours: 0.98 and 0.76 (cohort 1) and 1 and 0.57 (cohort 2), respectively. Computation time was around 6 s for a standard dataset. Conclusion: The application of a CNN for automated contouring of intraprostatic GTV in ⁶⁸Ga-PSMA and ¹⁸F-PSMA PET images resulted in a high concordance with expert contours and in high sensitivities and specificities in comparison with histology as a reference. This robust, accurate and fast technique may be implemented for treatment concepts in primary prostate cancer. The trained model and the study's source code are available in an open source repository.

One-Shot Learning for Deformable Medical Image Registration and Periodic Motion Tracking

Deformable image registration is a very important field of research in medical imaging. Recently multiple deep learning approaches were published in this area showing promising results. However, drawbacks of deep learning methods are the need for a large amount of training datasets and their inability to register unseen images different from the training datasets. One shot learning comes without the need of large training datasets and has already been proven to be applicable to 3D data. In this work we present a one shot registration approach for periodic motion tracking in 3D and 4D datasets. When applied to a 3D dataset the algorithm calculates the inverse of the registration vector field simultaneously. For registration we employed a U-Net combined with a coarse to fine approach and a differential spatial transformer module. The algorithm was thoroughly tested with multiple 4D and 3D datasets publicly available. The results show that the presented approach is able to track periodic motion and to yield a competitive registration accuracy. Possible applications are the use as a stand-alone algorithm for 3D and 4D motion tracking or in the beginning of studies until enough datasets for a separate training phase are available.

Dosimetric Impact of Interfractional Variations for Post-prostatectomy Radiotherapy to the Prostatic Fossa-Relevance for the Frequency of Position Verification Imaging and Treatment Adaptation

Background and purpose: To analyze divergences between the planned and applied treatment doses for post-prostatectomy radiotherapy to the prostatic fossa on a voxel-by-voxel basis based on interfractional anatomic variations and imaging frequency. Materials and methods: For 10 patients receiving intensity-modulated postoperative radiotherapy to the prostatic fossa, position verification was carried out by daily in-room CT imaging in treatment position (340 fraction CTs). Applied fraction doses were recalculated on daily CT scans, and treatment doses were accumulated on a voxel-by-voxel basis after deformable image registration. To simulate weekly imaging, derived weekly position correction vectors were used to rigidly register all daily scans of the respective treatment week onto the planning CT before dose accumulation. Detailed dose statistics of the prescribed and applied treatment doses were compared in relation to the frequency of position verification imaging. Derived NTCP and P_injury values were calculated for the rectum and bladder. Results: Despite a large variability in the pelvic anatomy, daily CT-based patient repositioning resulted in largely negligible deviations of the analyzed dose-volume, conformity, and uniformity parameters from the planned doses for post-prostatectomy radiotherapy, and only the bladder exhibited significant increases in the accumulated mean and median doses. Derived NTCP for the applied doses to the rectum and bladder and P_injury values did not significantly deviate from the treatment plan. In contrast, weekly CT-based repositioning resulted in significant decreases of the PTV coverage and dose conformity as well as large deviations of the applied doses to the rectum and bladder from the planned doses. Consecutively, NTCP for the rectum and P_injury were found falsely reduced for weekly patient repositioning. Conclusions: Our data indicate for the first time in a voxel-by-voxel analysis that daily imaging is required for reliable adaptive delivery of intensity-modulated radiotherapy to the prostatic fossa. This work will help guiding adaptive treatment strategies for post-prostatectomy radiotherapy.

Dosimetric Impact of Interfractional Variations in Prostate Cancer Radiotherapy-Implications for Imaging Frequency and Treatment Adaptation

Background and purpose: To analyze deviations of the applied from the planned doses on a voxel-by-voxel basis for definitive prostate cancer radiotherapy depending on anatomic variations and imaging frequency.

Materials and methods: Daily in-room CT imaging was performed in treatment position for 10 patients with prostate cancer undergoing intensity-modulated radiotherapy (340 fraction CTs). Applied fraction doses were recalculated on daily images, and voxel-wise dose accumulation was performed using a deformable registration algorithm. For weekly imaging, weekly position correction vectors were derived and used to rigidly register daily scans of that week to the planning CT scan prior to dose accumulation. Applied and prescribed doses were compared in dependence of the imaging frequency, and derived TCP and NTCP values were calculated.

Results: Daily CT-based repositioning resulted in non-significant deviations of all analyzed dose-volume, conformity and uniformity parameters to the CTV, bladder and rectum irrespective of anatomic changes. Derived average TCP values were comparable, and NTCP values for the applied doses to the bladder and rectum did not significantly deviate from the planned values. For weekly imaging, the applied D₂ to the CTV, rectum and bladder significantly varied from the planned doses, and the CTV conformity index and D₉₈ decreased. While TCP values were comparable, the NTCP for the bladder erroneously appeared reduced for weekly repositioning.

Conclusions: Based on daily diagnostic quality CT imaging and voxel-wise dose accumulation, we demonstrated for the first time that daily, but not weekly imaging resulted in only negligible deviations of the applied from the planned doses for prostate intensity-modulated radiotherapy. Therefore, weekly imaging may not be adequately reliable for adaptive treatment delivery techniques for prostate. This work will contribute to devising adaptive re-planning strategies for prostate radiotherapy.

Validation of different PSMA-PET/CT-based contouring techniques for intraprostatic tumor definition using histopathology as standard of reference

PURPOSE

Accurate definition of the intraprostatic gross tumor volume (GTV) is crucial for diagnostic and therapeutic approaches in patients with primary prostate cancer (PCa). The optimal methodology for contouring of GTV using Prostate specific membrane antigen positron emission tomography (PSMA-PET) information has not yet been defined.

METHODS AND MATERIALS

PCa patients who underwent a [68Ga]PSMA-11-PET/CT followed by radical prostatectomy were prospectively enrolled (n = 20). Six observer teams with different levels of experience and using different PET image scaling techniques performed manual contouring of GTV. Additionally, semi-automatic segmentation of GTVs was performed using SUVmax thresholds of 20-50%. Coregistered histopathological gross tumor volume (GTV-Histo) served as reference. Inter-observer agreement was assessed by calculating the Dice similarity coefficient (DSC).

RESULTS

Most contouring methods provided high sensitivity and specificity. For manual delineation, scaling the PET images from SUVmin-max: 0-5 resulted in high sensitivity (>86%). The highest specificity (100%) was obtained by scaling the PET images from SUVmin-max: 0-SUVmax. High interobserver agreement (median DSC 0.8) was observed when using the same PET image scaling technique (PET images SUVmin-max: 0-5). For semi-automatic segmentation, a low SUVmax threshold of 20% optimized sensitivity (SUVmax threshold 20%, 100% sensitivity, 32% of prostatic volume), whereas a higher threshold optimized specificity (SUVmax threshold 40%-50%, 100% specificity).

CONCLUSIONS

Contouring of regions with high tracer-uptake resulted in very high specificities and should be used for biopsy guidance. Both manual and semi-automatic approaches using validated SUV scaling (SUVmin-max: 0-5) or thresholding (20%) may provide high sensitivity, and should be considered for PSMA-PET-based focal therapy approaches.

Radiomic features from PSMA PET for non-invasive intraprostatic tumor discrimination and characterization in patients with intermediate- and high-risk prostate cancer - a comparison study with histology reference

Purpose: To evaluate the performance of radiomic features (RF) derived from PSMA PET for intraprostatic tumor discrimination and non-invasive characterization of Gleason score (GS) and pelvic lymph node status. Patients and methods: Patients with prostate cancer (PCa) who underwent [⁶⁸Ga]-PSMA-11 PET/CT followed by radical prostatectomy and pelvic lymph node dissection were prospectively enrolled (n=20). Coregistered histopathological gross tumor volume (GTV-Histo) in the prostate served as reference. 133 RF were derived from GTV-Histo and from manually created segmentations of the intraprostatic tumor volume (GTV-Exp). Spearman´s correlation coefficients (ρ) were assessed between RF derived from the different GTVs. We additionally analyzed the differences in RF values for PCa and non-PCa tissues. Furthermore, areas under receiver-operating characteristics curves (AUC) were calculated and uni- and multivariate analyses were performed to evaluate the RF based discrimination of GS 7 and ≥8 disease and of patients with nodal spread (pN1) and non-nodal spread (pN0) in surgical specimen. The results found in the latter analyses were validated by a retrospective cohort of 40 patients. Results: Most RF from GTV-Exp showed strong correlations with RF from GTV-Histo (86% with ρ>0.7). 81% and 76% of RF from GTV-Exp and GTV-Histo significantly discriminated between PCa and non-PCa tissue. The texture feature QSZHGE discriminated between GS 7 and ≥8 considering GTV-Histo (AUC=0.93) and GTV-Exp (prospective cohort: AUC=0.91 / validation cohort: AUC=0.84). QSZHGE also discriminated between pN1 and pN0 disease considering GTV-Histo (AUC=0.85) and GTV-Exp (prospective cohort: AUC=0.87 / validation cohort: AUC=0.85). In uni- and multivariate analyses including patients of both cohorts QSZHGE was a statistically significant (p<0.01) predictor for PCa patients with GS ≥8 tumors and pN1 status. Conclusion: RF derived from PSMA PET discriminated between PCa and non-PCa tissue within the prostate. Additionally, the texture feature QSZHGE discriminated between GS 7 and GS ≥8 tumors and between patients with pN1 and pN0 disease. Our results support the role of RF in PSMA PET as a new tool for non-invasive PCa discrimination and characterization of its biological properties.

Visualization of 4D multimodal imaging data and its applications in radiotherapy planning

PURPOSE

To explore the benefit of using 4D multimodal visualization and interaction techniques for defined radiotherapy planning tasks over a treatment planning system used in clinical routine (C-TPS) without dedicated 4D visualization.

METHODS

We developed a 4D visualization system (4D-VS) with dedicated rendering and fusion of 4D multimodal imaging data based on a list of requirements developed in collaboration with radiation oncologists. We conducted a user evaluation in which the benefits of our approach were evaluated in comparison to C-TPS for three specific tasks: assessment of internal target volume (ITV) delineation, classification of tumor location in peripheral or central, and assessment of dose distribution. For all three tasks, we presented test cases for which we measured correctness, certainty, consistency followed by an additional survey regarding specific visualization features.

RESULTS

Lower quality of the test ITVs (ground truth quality was available) was more likely to be detected using 4D-VS. ITV ratings were more consistent in 4D-VS and the classification of tumor location had a higher accuracy. Overall evaluation of the survey indicates 4D-VS provides better spatial comprehensibility and simplifies the tasks which were performed during testing.

CONCLUSIONS

The use of 4D-VS has improved the assessment of ITV delineations and classification of tumor location. The visualization features of 4D-VS have been identified as helpful for the assessment of dose distribution during user testing.

Comparison of PET/CT and whole-mount histopathology sections of the human prostate: a new strategy for voxel-wise evaluation

Background

Implementation of PET/CT in diagnosis of primary prostate cancer (PCa) requires a profound knowledge about the tracer, preferably from a quantitative evaluation. Direct visual comparison of PET/CT slices to whole prostate sections is hampered by considerable uncertainties from imperfect coregistration and fundamentally different image modalities. In the current study, we present a novel method for advanced voxel-wise comparison of histopathology from excised prostates to pre-surgical PET. Resected prostates from eight patients who underwent PSMA-PET/CT were scanned (ex vivo CT) and thoroughly pathologically prepared. In vivo and ex vivo CT including histopathology were coregistered with three different methods (manual, semi−/automatic). Spatial overlap after CT-based registration was evaluated with dice similarity (DSC). Furthermore, we constructed 3D cancer distribution models from histopathologic information in various slices. Subsequent smoothing reflected the intrinsically limited spatial resolution of PSMA-PET. The resulting histoPET models were used for quantitative analysis of spatial histopathology-PET pattern agreement focusing on p values and coefficients of determination (R²). We examined additional rigid mutual information (MI) coregistration directly based on PSMA-PET and histoPET.

Results

Mean DSC for the three different methods (ManReg, ScalFactReg, and DefReg) were 0.79 ± 0.06, 0.82 ± 0.04, and 0.90 ± 0.02, respectively, while quantification of PET-histopathology pattern agreement after CT-based registration revealed R² 45.7, 43.2, and 41.3% on average with p < 10⁻⁵. Subsequent PET-based MI coregistration yielded R² 61.3, 55.9, and 55.6%, respectively, while implying anatomically plausible transformations.

Conclusions

Creating 3D histoPET models based on thorough histopathological preparation allowed sophisticated quantitative analyses showing highly significant correlations between histopathology and (PSMA-)PET. We recommend manual CT-based coregistration followed by a PET-based MI algorithm to overcome limitations of purely CT-based coregistrations for meaningful voxel-wise comparisons between PET and histopathology.

Visualization of Deformable Image Registration Quality Using Local Image Dissimilarity

Deformable image registration (DIR) has the potential to improve modern radiotherapy in many aspects, including volume definition, treatment planning and image-guided adaptive radiotherapy. Studies have shown its possible clinical benefits. However, measuring DIR accuracy is difficult without known ground truth, but necessary before integration in the radiotherapy workflow. Visual assessment is an important step towards clinical acceptance. We propose a visualization framework which supports the exploration and the assessment of DIR accuracy. It offers different interaction and visualization features for exploration of candidate regions to simplify the process of visual assessment. The visualization is based on voxel-wise comparison of local image patches for which dissimilarity measures are computed and visualized to indicate locally the registration results. We performed an evaluation with three radiation oncologists to demonstrate the viability of our approach. In the evaluation, lung regions were rated by the participants with regards to their visual accuracy and compared to the registration error measured with expert defined landmarks. Regions rated as "accepted" had an average registration error of 1.8 mm, with the highest single landmark error being 3.3 mm. Additionally, survey results show that the proposed visualizations support a fast and intuitive investigation of DIR accuracy, and are suitable for finding even small errors.

(68)Ga-HBED-CC-PSMA PET/CT Versus Histopathology in Primary Localized Prostate Cancer: A Voxel-Wise Comparison

Purpose: We performed a voxel-wise comparison of ⁶⁸Ga-HBED-CC-PSMA PET/CT with prostate histopathology to evaluate the performance of ⁶⁸Ga-HBED-CC-PSMA for the detection and delineation of primary prostate cancer (PCa).

Methodology: Nine patients with histopathological proven primary PCa underwent ⁶⁸Ga-HBED-CC-PSMA PET/CT followed by radical prostatectomy. Resected prostates were scanned by ex-vivo CT in a special localizer and histopathologically prepared. Histopathological information was matched to ex-vivo CT. PCa volume (PCa-histo) and non-PCa tissue in the prostate (NPCa-histo) were processed to obtain a PCa-model, which was adjusted to PET-resolution (histo-PET). Each histo-PET was coregistered to in-vivo PSMA-PET/CT data.

Results: Analysis of spatial overlap between histo-PET and PSMA PET revealed highly significant correlations (p < 10^-5) in nine patients and moderate to high coefficients of determination (R²) from 42 to 82 % with an average of 60 ± 14 % in eight patients (in one patient R² = 7 %). Mean SUVmean in PCa-histo and NPCa-histo was 5.6 ± 6.1 and 3.3 ± 2.5 (p = 0.012). Voxel-wise receiver-operating characteristic (ROC) analyses comparing the prediction by PSMA-PET with the non-smoothed tumor distribution from histopathology yielded an average area under the curve of 0.83 ± 0.12. Absolute and relative SUV (normalized to SUVmax) thresholds for achieving at least 90 % sensitivity were 3.19 ± 3.35 and 0.28 ± 0.09, respectively.

Conclusions: Voxel-wise analyses revealed good correlations of ⁶⁸Ga-HBED-CC-PSMA PET/CT and histopathology in eight out of nine patients. Thus, PSMA-PET allows a reliable detection and delineation of PCa as basis for PET-guided focal therapies.

3 Tesla multiparametric MRI for GTV-definition of Dominant Intraprostatic Lesions in patients with Prostate Cancer--an interobserver variability study

PURPOSE

To evaluate the interobserver variability of gross tumor volume (GTV) - delineation of Dominant Intraprostatic Lesions (DIPL) in patients with prostate cancer using published MRI criteria for multiparametric MRI at 3 Tesla by 6 different observers.

MATERIAL AND METHODS

90 GTV-datasets based on 15 multiparametric MRI sequences (T2w, diffusion weighted (DWI) and dynamic contrast enhanced (DCE)) of 5 patients with prostate cancer were generated for GTV-delineation of DIPL by 6 observers. The reference GTV-dataset was contoured by a radiologist with expertise in diagnostic imaging of prostate cancer using MRI. Subsequent GTV-delineation was performed by 5 radiation oncologists who received teaching of MRI-features of primary prostate cancer before starting contouring session. GTV-datasets were contoured using Oncentra Masterplan® and iplan® Net. For purposes of comparison GTV-datasets were imported to the Artiview® platform (Aquilab®), GTV-values and the similarity indices or Kappa indices (KI) were calculated with the postulation that a KI > 0.7 indicates excellent, a KI > 0.6 to < 0.7 substantial and KI > 0.5 to < 0.6 moderate agreement. Additionally all observers rated difficulties of contouring for each MRI-sequence using a 3 point rating scale (1 = easy to delineate, 2 = minor difficulties, 3 = major difficulties).

RESULTS

GTV contouring using T2w (KI-T2w = 0.61) and DCE images (KI-DCE = 0.63) resulted in substantial agreement. GTV contouring using DWI images resulted in moderate agreement (KI-DWI = 0.51). KI-T2w and KI-DCE was significantly higher than KI-DWI (p = 0.01 and p = 0.003). Degree of difficulty in contouring GTV was significantly lower using T2w and DCE compared to DWI-sequences (both p < 0.0001). Analysis of delineation differences revealed inadequate comparison of functional (DWI, DCE) to anatomical sequences (T2w) and lack of awareness of non-specific imaging findings as a source of erroneous delineation.

CONCLUSIONS

Using T2w and DCE sequences at 3 Tesla for GTV-definition of DIPL in prostate cancer patients by radiation oncologists with knowledge of MRI features results in substantial agreement compared to an experienced MRI-radiologist, but for radiotherapy purposes higher KI are desirable, strengthen the need for expert surveillance. DWI sequence for GTV delineation was considered as difficult in application.