Automatic Contours Research Articles

Minimizing human interference in an online fully automated daily adaptive radiotherapy workflow for bladder cancer

PurposeThe aim was to study the potential for an online fully automated daily adaptive radiotherapy (RT) workflow for bladder cancer, employing a focal boost and fiducial markers. The study focused on comparing the geometric and dosimetric aspects between the simulated automated online adaptive RT (oART) workflow and the clinically performed workflow.MethodsSeventeen patients with muscle-invasive bladder cancer were treated with daily Cone Beam CT (CBCT)-guided oART. The bladder and pelvic lymph nodes (CTVelective) received a total dose of 40 Gy in 20 fractions and the tumor bed received an additional simultaneously integrated boost (SIB) of 15 Gy (CTVboost). During the online sessions a CBCT was acquired and used as input for the AI-network to automatically delineate the bladder and rectum, i.e. influencers. These influencers were employed to guide the algorithm utilized in the delineation process of the target. Manual adjustments to the generated contours are common during this clinical workflow prior to plan reoptimization and RT delivery. To study the potential for an online fully automated workflow, the oART workflow was repeated in a simulation environment without manual adjustments. A comparison was made between the clinical and automatic contours and between the treatment plans optimized on these clinical (Dclin) and automatic contours (Dauto).ResultsThe bladder and rectum delineated by the AI-network differed from the clinical contours with a median Dice Similarity Coefficient of 0.99 and 0.92, a Mean Distance to Agreement of 1.9 mm and 1.3 mm and a relative volume of 100% and 95%, respectively. For the CTVboost these differences were larger, namely 0.71, 7 mm and 78%. For the CTVboost the median target coverage was 0.42% lower for Dauto compared to Dclin. For CTVelective this difference was 0.03%. The target coverage of Dauto met the clinical requirement of the CTV-coverage in 65% of the sessions for CTVboost and 95% of the sessions for the CTVelective.ConclusionsWhile an online fully automated daily adaptive RT workflow shows promise for bladder treatment, its complexity becomes apparent when incorporating a focal boost, necessitating manual checks to prevent potential underdosage of the target.

Integrated framework for quantitative T2-weighted MRI analysis following prostate cancer radiotherapy

PurposeThe aim of this study is to develop a framework for quantitative analysis of longitudinal T2-weighted MRIs (T2w) following radiotherapy (RT) for prostate cancer. Materials and methodsThe developed methodology includes: (i) deformable image registration of longitudinal series to pre-RT T2w for automated detection of prostate, peripheral zone (PZ), and gross tumor volume (GTV); and (ii) T2w signal-intensity harmonization based on three reference tissues. The REgistration and HARMonization (REHARM) framework was applied on T2w acquired in a clinical trial consisting of two pre-RT and three post-RT MRI exams. Image registration was assessed by the DICE coefficient between automatic and manual contours, and intensity normalization via inter-patient histogram intersection. Longitudinal consistency was evaluated by the repeatability coefficient and Pearson correlation (r) between the two T2w exams before RT. ResultsT2w from 107 MRI exams (23 patients) were utilized. Following REHARM, the histogram intersections for prostate, PZ and GTV increased from median = 0.43/0.16/0.13 to 0.66/0.44/0.46. The repeatability in T2w intensity estimation was better for the automatic than the manual contours for all three ROIs (r = 0.9, p < 0.0001, for GTV). The changes in the tissues’ T2w values pre- and post-RT became significant, indicating the measurable quantitative signal related to radiation. ConclusionsThe developed methodology allows to automate longitudinal analysis reducing data acquisition-related variation and improving consistency. The quantitative characterization of RT-induced changes in T2w will lead to new understanding of radiation effects enabling prediction modeling of RT response.

Impact of Manual Contour Editing on Plan Quality for Online Adaptive Radiation Therapy for Head and Neck Cancer

PurposeOnline adaptive radiotherapy (oART) has high resource costs especially for head-and-neck (H&N) cancer, which requires recontouring complex targets and numerous organs-at-risk (OARs). ART systems provide auto-contours to help, we aim to explore the optimal level of editing automatic contours to maintain plan quality in a cone-beam-computed-tomography (CBCT)-based oART system for H&N. In this system influencer OAR contours are generated and reviewed first, which then drives the auto-contouring of the remaining OARs and targets. Methods and MaterialsThree-hundred-and-forty-nine adapted fractions of forty-four H&N patients were retrospectively analyzed, with physician-edited OARs and targets. These contours and associated online adapted plans served as the gold standard for comparison. We simulated three contour editing workflows: (1) no editing of contours, (2) only editing the influencers, (3) editing the influencers and targets. The geometric difference was quantified with Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD). The dosimetric differences in target coverage and OAR doses were calculated between the gold standard and these three simulated workflows. ResultsWorkflow 1 resulted in significantly inferior contour quality for all OARs (mean DSC 0.85±0.17 and HD95 3.10±5.80mm), dosimetric data was hence not calculated for workflow 1. In workflow (2), the frequency of physician editing targets and remaining OARs were 80.8%-95.7% and 2.3% (brachial plexus)-67.7% (oral cavity) respectively, where the OAR differences were geometrically minor (mean DSC>0.95 with std≤0.09). However, due to the unedited target contours of workflow 2 (mean DSC 0.86-0.92 and mean HD95 2.56-3.30mm versus the ground-truth targets), plans were inadequate with insufficient coverage. In workflow (3) when both targets and influencers were edited (non-influencer OARs were unedited), over 95.5% of the adapted plans achieved the patient-specific dosimetry goals. ConclusionThe CBCT-based H&N oART workflow can be meaningfully accelerated by only editing the influencers and targets while omitting the remaining OARs without compromising the quality of the adaptive plans.

Deep learning with uncertainty estimation for automatic tumor segmentation in PET/CT of head and neck cancers: impact of model complexity, image processing and augmentation

Objective. Target volumes for radiotherapy are usually contoured manually, which can be time-consuming and prone to inter- and intra-observer variability. Automatic contouring by convolutional neural networks (CNN) can be fast and consistent but may produce unrealistic contours or miss relevant structures. We evaluate approaches for increasing the quality and assessing the uncertainty of CNN-generated contours of head and neck cancers with PET/CT as input. Approach. Two patient cohorts with head and neck squamous cell carcinoma and baseline 18F-fluorodeoxyglucose positron emission tomography and computed tomography images (FDG-PET/CT) were collected retrospectively from two centers. The union of manual contours of the gross primary tumor and involved nodes was used to train CNN models for generating automatic contours. The impact of image preprocessing, image augmentation, transfer learning and CNN complexity, architecture, and dimension (2D or 3D) on model performance and generalizability across centers was evaluated. A Monte Carlo dropout technique was used to quantify and visualize the uncertainty of the automatic contours. Main results. CNN models provided contours with good overlap with the manually contoured ground truth (median Dice Similarity Coefficient: 0.75–0.77), consistent with reported inter-observer variations and previous auto-contouring studies. Image augmentation and model dimension, rather than model complexity, architecture, or advanced image preprocessing, had the largest impact on model performance and cross-center generalizability. Transfer learning on a limited number of patients from a separate center increased model generalizability without decreasing model performance on the original training cohort. High model uncertainty was associated with false positive and false negative voxels as well as low Dice coefficients. Significance. High quality automatic contours can be obtained using deep learning architectures that are not overly complex. Uncertainty estimation of the predicted contours shows potential for highlighting regions of the contour requiring manual revision or flagging segmentations requiring manual inspection and intervention.

2859: Deep Learning reconstructed MR images: impact on manual and automatic contours

2859: Deep Learning reconstructed MR images: impact on manual and automatic contours

Comparison of commercial atlas-based automatic segmentation software for prostate radiotherapy treatment planning.

This study aims to assess the accuracy of automatic atlas-based contours for various key anatomical structures in prostate radiotherapy treatment planning. The evaluated structures include the bladder, rectum, prostate, seminal vesicles, femoral heads and penile bulb. CT images from 20 patients who underwent intensity-modulated radiotherapy were randomly chosen to create an atlas library. Atlas contours of the seven anatomical structures were generated using four software packages: ABAS, Eclipse, MIM, and RayStation. These contours were then compared to manual delineations performed by oncologists, which served as the ground truth. Evaluation metrics such as dice similarity coefficient (DSC), mean distance to agreement (MDA), and volume ratio (VR) were calculated to assess the accuracy of the contours. Additionally, the time taken by each software to generate the atlas contour was recorded. The mean DSC values for the bladder exhibited strong agreement (>0.8) with manual delineations for all software except for Eclipse and RayStation. Similarly, the femoral heads showed significant similarity between the atlas contours and ground truth across all software, with mean DSC values exceeding 0.9 and MDA values close to zero. On the other hand, the penile bulb displayed only moderate agreement with the ground truth, with mean DSC values ranging from 0.5 to 0.7 for all software. A similar trend was observed in the prostate atlas contours, except for MIM, which achieved a mean DSC of over 0.8. For the rectum, both ABAS and MIM atlases demonstrated strong agreement with the ground truth, resulting in mean DSC values of more than 0.8. Overall, MIM and ABAS outperformed Eclipse and RayStation in both DSC and MDA. These results indicate that the atlas-based segmentation employed in this study produces acceptable contours for the anatomical structures of interest in prostate radiotherapy treatment planning.

Evaluating automatically generated normal tissue contours for safe use in head and neck and cervical cancer treatment planning.

Volumetric-modulated arc therapy (VMAT) is a widely accepted treatment method for head and neck (HN) and cervical cancers; however, creating contours and plan optimization for VMAT plans is a time-consuming process. Our group has created an automated treatment planning tool, the Radiation Planning Assistant (RPA), that uses deep learning models to generate organs at risk (OARs), planning structures and automates plan optimization. This study quantitatively evaluates the quality of contours generated by the RPA tool. For patients with HN (54) and cervical (39) cancers, we retrospectively generated autoplans using the RPA. Autoplans were generated using deep-learning and RapidPlan models developed in-house. The autoplans were, then, applied to the original, physician-drawn contours, which were used as a ground truth (GT) to compare with the autocontours (RPA). Using a "two one-sided tests" (TOST) procedure, we evaluated whether the autocontour normal tissue dose was equivalent to that of the ground truth by a margin, δ, that we determined based on clinical judgement. We also calculated the number of plans that met established clinically accepted dosimetric criteria. For HN plans, 91.8% and 91.7% of structures met dosimetric criteria for automatic and manual contours, respectively; for cervical plans, 95.6% and 95.7% of structures met dosimetric criteria for automatic and manual contours, respectively. Autocontours were equivalent to the ground truth for 71% and 75% of common DVH metrics for the HN and cervix, respectively. This study shows that dosimetrically equivalent normal tissue contours can be created for HN and cervical cancers using deep learning techniques. In general, differences between the contours did not affect the passing or failing of clinical dose tolerances.

Deep learning and atlas-based models to streamline the segmentation workflow of total marrow and lymphoid irradiation.

To improve the workflow of total marrow and lymphoid irradiation (TMLI) by enhancing the delineation of organs at risk (OARs) and clinical target volume (CTV) using deep learning (DL) and atlas-based (AB) segmentation models. Ninety-five TMLI plans optimized in our institute were analyzed. Two commercial DL software were tested for segmenting 18 OARs. An AB model for lymph node CTV (CTV_LN) delineation was built using 20 TMLI patients. The AB model was evaluated on 20 independent patients, and a semiautomatic approach was tested by correcting the automatic contours. The generated OARs and CTV_LN contours were compared to manual contours in terms of topological agreement, dose statistics, and time workload. A clinical decision tree was developed to define a specific contouring strategy for each OAR. The two DL models achieved a median [interquartile range] dice similarity coefficient (DSC) of 0.84 [0.71;0.93] and 0.85 [0.70;0.93] across the OARs. The absolute median Dmean difference between manual and the two DL models was 2.0 [0.7;6.6]% and 2.4 [0.9;7.1]%. The AB model achieved a median DSC of 0.70 [0.66;0.74] for CTV_LN delineation, increasing to 0.94 [0.94;0.95] after manual revision, with minimal Dmean differences. Since September 2022, our institution has implemented DL and AB models for all TMLI patients, reducing from 5 to 2h the time required to complete the entire segmentation process. DL models can streamline the TMLI contouring process of OARs. Manual revision is still necessary for lymph node delineation using AB models.

Prospects for daily online adaptive radiotherapy for cervical cancer: Auto-contouring evaluation and dosimetric outcomes

BackgroundTraining senior radiation therapists as “adapters” to manage influencers and target editing is critical in daily online adaptive radiotherapy (oART) for cervical cancer. The purpose of this study was to evaluate the accuracy and dosimetric outcomes of automatic contouring and identify the key areas for modification.MethodsA total of 125 oART fractions from five postoperative cervical cancer patients and 140 oART fractions from five uterine cervical cancer patients treated with daily iCBCT-guided oART were enrolled in this prospective study. The same adaptive treatments were replanned using the Ethos automatic contours workflow without manual contouring edits. The clinical target volume (CTV) was subdivided into several separate regions, and the average surface distance dice (ASD), centroid deviation, dice similarity coefficient (DSC), and 95% Hausdorff distance (95% HD) were used to evaluate contouring for the above portions. Dosimetric results from automatic oART plans were compared to supervised oART plans to evaluate target volumes and organs at risk (OARs) dose changes.ResultsOverall, the paired CTV had high overlap rates, with an average DSC value greater than 0.75. The uterus had the largest consistency differences, with ASD, centroid deviation, and 95% HD being 2.67 ± 1.79 mm, 17.17 ± 12 mm, and 10.45 ± 5.68 mm, respectively. The consistency differences of the lower nodal CTVleft and nodal CTVright were relatively large, with ASD, centroid deviation, and 95% HD being 0.59 ± 0.53 mm, 3.6 ± 2.67 mm, and 5.41 ± 4.08 mm, and 0.59 ± 0.51 mm, 3.6 ± 2.54 mm, and 4.7 ± 1.57 mm, respectively. The automatic online-adapted plan met the clinical requirements of dosimetric coverage for the target volume and improved the OAR dosimetry.ConclusionsThe accuracy of automatic contouring from the Ethos adaptive platform is considered clinically acceptable for cervical cancer, and the uterus, upper vaginal cuff, and lower nodal CTV are the areas that need to be focused on in training.

Clinical validation of commercial deep-learning based auto-segmentation models for organs at risk in the head and neck region: a single institution study.

To evaluate organ at risk (OAR) auto-segmentation in the head and neck region of computed tomography images using two different commercially available deep-learning-based auto-segmentation (DLAS) tools in a single institutional clinical applications. Twenty-two OARs were manually contoured by clinicians according to published guidelines on planning computed tomography (pCT) images for 40 clinical head and neck cancer (HNC) cases. Automatic contours were generated for each patient using two deep-learning-based auto-segmentation models-Manteia AccuContour and MIM ProtégéAI. The accuracy and integrity of autocontours (ACs) were then compared to expert contours (ECs) using the Sørensen-Dice similarity coefficient (DSC) and Mean Distance (MD) metrics. ACs were generated for 22 OARs using AccuContour and 17 OARs using ProtégéAI with average contour generation time of 1 min/patient and 5 min/patient respectively. EC and AC agreement was highest for the mandible (DSC 0.90 ± 0.16) and (DSC 0.91 ± 0.03), and lowest for the chiasm (DSC 0.28 ± 0.14) and (DSC 0.30 ± 0.14) for AccuContour and ProtégéAI respectively. Using AccuContour, the average MD was<1mm for 10 of the 22 OARs contoured, 1-2mm for 6 OARs, and 2-3mm for 6 OARs. For ProtégéAI, the average mean distance was<1mm for 8 out of 17 OARs, 1-2mm for 6 OARs, and 2-3mm for 3 OARs. Both DLAS programs were proven to be valuable tools to significantly reduce the time required to generate large amounts of OAR contours in the head and neck region, even though manual editing of ACs is likely needed prior to implementation into treatment planning. The DSCs and MDs achieved were similar to those reported in other studies that evaluated various other DLAS solutions. Still, small volume structures with nonideal contrast in CT images, such as nerves, are very challenging and will require additional solutions to achieve sufficient results.

Real-world validation of Artificial Intelligence-based Computed Tomography auto-contouring for prostate cancer radiotherapy planning

Background and purposeArtificial Intelligence (AI)-based auto-contouring for treatment planning in radiotherapy needs extensive clinical validation, including the impact of editing after automatic segmentation. The aims of this study were to assess the performance of a commercial system for Clinical Target Volumes (CTVs) (prostate/seminal vesicles) and selected Organs at Risk (OARs) (rectum/bladder/femoral heads + femurs), evaluating also inter-observer variability (manual vs automatic + editing) and the reduction of contouring time. Materials and methodsTwo expert observers contoured CTVs/OARs of 20 patients in our Treatment Planning System (TPS). Computed Tomography (CT) images were sent to the automatic contouring workstation: automatic contours were generated and sent back to TPS, where observers could edit them if necessary. Inter- and intra-observer consistency was estimated using Dice Similarity Coefficients (DSC). Radiation oncologists were also asked to score the quality of automatic contours, ranging from 1 (complete re-contouring) to 5 (no editing). Contouring times (manual vs automatic + edit) were compared. ResultsDSCs (manual vs automatic only) were consistent with inter-observer variability (between 0.65 for seminal vesicles and 0.94 for bladder); editing further improved performances (range: 0.76–0.94). The median clinical score was 4 (little editing) and it was <4 in 3/2 patients for the two observers respectively. Inter-observer variability of automatic + editing contours improved significantly, being lower than manual contouring (e.g.: seminal vesicles: 0.83vs0.73; prostate: 0.86vs0.83; rectum: 0.96vs0.81). Oncologist contouring time reduced from 17 to 24 min of manual contouring time to 3–7 min of editing time for the two observers (p < 0.01). ConclusionAutomatic contouring with a commercial AI-based system followed by editing can replace manual contouring, resulting in significantly reduced time for segmentation and better consistency between operators.

Balancing Quality and Efficiency of Head and Neck Contour Revision for Online Adaptive Radiotherapy

The number of target and organs-at-risk (OAR) structures requiring contours is a barrier to online adaptive radiotherapy (oART) implementation for head and neck (H&N) cancer. In sharing our experience utilizing a cone beam CT-based oART system featuring automatic contours, we analyzed the system's clinical performance and the dosimetric benefits of human supervision to investigate the optimal balance between contour editing and plan quality. We analyzed 44 H&N patients and 349 adaptive fractions treated on the oART system. The unsupervised (automatically generated) and supervised (edited by clinicians) target and OAR contours as well as the corresponding adaptive plans were acquired. Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD) used to assess the geometric extent of contour supervision. We mapped the clinically treated adaptive plan to the unsupervised contours to quantify the dosimetric effect of the contour edits, and the adaptive plan optimized from the unsupervised contours to the supervised contours to simulate an unsupervised workflow. The targets were edited in >80.7% of the sessions, with the most changes seen in the primary GTV (DSC = 0.86 ± 0.09 and HD = 2.77 ± 1.58 mm) and the least in the nodal CTV (DSC = 0.92 ± 0.06, HD = 2.29 ± 1.49 mm). Among OARs, oral cavity was the most frequently edited (49.0%) and brainstem the least (2.2%). In general, OAR edits were minor (mean DSC > 0.95 with std Dosimetric quantification of the edits (Table 1) showed that while the unsupervised targets resulted in significant coverage loss, the change in unsupervised OAR dose was not as pronounced, with 81.5-96.8% still meeting the preplan constraints. Edits corresponding to major dose changes were mostly adjacent to or inside targets. The unsupervised workflow underdosed the targets and overdosed the OARs, and therefore is not recommended. While target contours needed physician supervision in H&N oART, edits to automatic OAR contours were minor. An alternative contour workflow could be adopted to focus on the targets and reduce time spent on the OARs. For OAR close to or overlapping with the targets, strategic cropping could potentially minimize time for editing and improve the robustness of the adapted plan quality.

Evaluation of Observer Variation as Natural Experiment to Detect Sensitive Heart Subregions

There is growing evidence associating dose to the base of the heart to reduced survival of lung cancer patients (McWilliam EJC 2017). However, randomized evidence on the benefits of sparing the base of the heart is missing. In this study we investigate variability in the shape between patients of heart contours used in planning as a natural experiment to evaluate selective heart sparing. The core assumption is that regions that are not included in the heart delineation will not be spared. Data was collected for 1705 lung cancer patients treated in one center between 2010 and 2016 with IMRT/VMAT (55-66Gy in 20-33#) or SABR (54-60Gy in 3-8#), planned using manual heart contours, called delineation 1. Consistent reference heart contours were obtained using a commercial autocontouring software, called delineation 2. Heart shapes were mapped to spherical coordinates (φ, Φ, r) and the difference in radius Δr = r1 - r2 for each set of angles φ, Φ was calculated for all patients. A large Δr means that the manual delineation use for planning is relatively large in the direction. Cox-regression was performed for each set of angles using Δr, r2 and its interaction using overall survival as endpoint. Permutation testing is used to avoid multiple testing issues. The aim is to locate a region of the heart where bigger delineations lead to better sparing and hence better survival. On average the heart base in our manual contours extends 34 mm more superior than automatic contours, because our protocol stipulates the inclusion of the full pericardial sac. Δr was not correlated with any clinical variables and is therefore a good candidate as instrumental variable in causal inference. Univariable Cox regression of Δr showed a uniform worsening of survival with larger manual delineations, likely due to reduced overall sparing given the use of volumetric dose constraints. After including the interaction with r2, no significant heart regions were found. However, analysis using the overall volume of the manual and auto-delineated heart did show a small but significant interaction effect where larger manual delineations improved outcome for smaller hearts. Our interpretation is that delineation variability relative to autocontouring (e.g., 1.9 mm SD at right atrium, up to 15 mm SD at apex) is not big enough to impact significantly on the heart dose and therewith survival because volumetric costs functions are used. In the future we will extend this analysis to include planned dose. Variability in contouring in our cohort is not large enough to be used as a natural experiment to test the impact of selective heart sparing. However, larger volume delineations of small hearts are associated with reduced mortality, suggesting the importance of sparing the base of the heart where most contouring variability occurs.

Clinical evaluation of deep learning-based automatic clinical target volume segmentation: a single-institution multi-site tumor experience.

The large variability in tumor appearance and shape makes manual delineation of the clinical target volume (CTV) time-consuming, and the results depend on the oncologists' experience. Whereas deep learning techniques have allowed oncologists to automate the CTV delineation, multi-site tumor analysis is often lacking in the literature. This study aimed to evaluate the deep learning models that automatically contour CTVs of tumors at various sites on computed tomography (CT) images from objective and subjective perspectives. 577 patients were selected for the present study, including nasopharyngeal (n = 34), esophageal (n = 40), breast-conserving surgery (BCS) (left-sided, n = 71; right-sided, n = 71), breast-radical mastectomy (BRM) (left-sided, n = 43; right-sided, n = 37), cervical (radical radiotherapy, n = 45; postoperative, n = 85), prostate (n = 42), and rectal (n = 109) carcinomas. Manually delineated CTV contours by radiation oncologists are served as ground truth. Four models were evaluated: Flexnet, Unet, Vnet, and Segresnet, which are commercially available in the medical product "AccuLearning AI model training platform". The data were divided into the training, validation, and testing set at a ratio of 5:1:4. The geometric metrics, including Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD), were calculated for objective evaluation. For subjective assessment, oncologists rated the segmentation contours of the testing set visually. High correlations were observed between automatic and manual contours. Based on the results of the independent test group, most of the patients achieved satisfactory quantitative results (DSC > 0.8), except for patients with esophageal carcinoma (DSC: 0.62-0.64). The subjective review indicated that 82.65% of predicted CTVs scored either as clinically accepting (8.68%) or requiring minor revision (73.97%), and no patients were scored as rejected. This experimental work demonstrated that auto-generated contours could serve as an initial template to help oncologists save time in CTV delineation. The deep learning-based auto-segmentations achieve acceptable accuracy and show the potential to improve clinical efficiency for radiotherapy of a variety of cancer.

Dosimetric comparison of autocontouring techniques for online adaptive proton therapy

Objective. Anatomical and daily set-up uncertainties impede high precision delivery of proton therapy. With online adaptation, the daily plan is reoptimized on an image taken shortly before the treatment, reducing these uncertainties and, hence, allowing a more accurate delivery. This reoptimization requires target and organs-at-risk (OAR) contours on the daily image, which need to be delineated automatically since manual contouring is too slow. Whereas multiple methods for autocontouring exist, none of them are fully accurate, which affects the daily dose. This work aims to quantify the magnitude of this dosimetric effect for four contouring techniques. Approach. Plans reoptimized on automatic contours are compared with plans reoptimized on manual contours. The methods include rigid and deformable registration (DIR), deep-learning based segmentation and patient-specific segmentation. Main results. It was found that independently of the contouring method, the dosimetric influence of using automatic OAR contours is small (<5% prescribed dose in most cases), with DIR yielding the best results. Contrarily, the dosimetric effect of using the automatic target contour was larger (>5% prescribed dose in most cases), indicating that manual verification of that contour remains necessary. However, when compared to non-adaptive therapy, the dose differences caused by automatically contouring the target were small and target coverage was improved, especially for DIR. Significance. The results show that manual adjustment of OARs is rarely necessary and that several autocontouring techniques are directly usable. Contrarily, manual adjustment of the target is important. This allows prioritizing tasks during time-critical online adaptive proton therapy and therefore supports its further clinical implementation.

An investigation into the risk of population bias in deep learning autocontouring

Background and PurposeTo date, data used in the development of Deep Learning-based automatic contouring (DLC) algorithms have been largely sourced from single geographic populations. This study aimed to evaluate the risk of population-based bias by determining whether the performance of an autocontouring system is impacted by geographic population. Materials and methods80 Head Neck CT deidentified scans were collected from four clinics in Europe (n = 2) and Asia (n = 2). A single observer manually delineated 16 organs-at-risk in each. Subsequently, the data was contoured using a DLC solution, and trained using single institution (European) data. Autocontours were compared to manual delineations using quantitative measures. A Kruskal-Wallis test was used to test for any difference between populations. Clinical acceptability of automatic and manual contours to observers from each participating institution was assessed using a blinded subjective evaluation. ResultsSeven organs showed a significant difference in volume between groups. Four organs showed statistical differences in quantitative similarity measures. The qualitative test showed greater variation in acceptance of contouring between observers than between data from different origins, with greater acceptance by the South Korean observers. ConclusionMuch of the statistical difference in quantitative performance could be explained by the difference in organ volume impacting the contour similarity measures and the small sample size. However, the qualitative assessment suggests that observer perception bias has a greater impact on the apparent clinical acceptability than quantitatively observed differences. This investigation of potential geographic bias should extend to more patients, populations, and anatomical regions in the future.

Automatic AI-based contouring of prostate MRI for online adaptive radiotherapy

Background and purposeMR-guided radiotherapy (MRgRT) online plan adaptation accounts for tumor volume changes, interfraction motion and thus allows daily sparing of relevant organs at risk. Due to the high interfraction variability of bladder and rectum, patients with tumors in the pelvic region may strongly benefit from adaptive MRgRT. Currently, fast automatic annotation of anatomical structures is not available within the online MRgRT workflow. Therefore, the aim of this study was to train and validate a fast, accurate deep learning model for automatic MRI segmentation at the MR-Linac for future implementation in a clinical MRgRT workflow. Materials and methodsFor a total of 47 patients, T2w MRI data were acquired on a 1.5 T MR-Linac (Unity, Elekta) on five different days. Prostate, seminal vesicles, rectum, anal canal, bladder, penile bulb, body and bony structures were manually annotated. These training data consisting of 232 data sets in total was used for the generation of a deep learning based autocontouring model and validated on 20 unseen T2w-MRIs. For quantitative evaluation the validation set was contoured by a radiation oncologist as gold standard contours (GSC) and compared in MATLAB to the automatic contours (AIC). For the evaluation, dice similarity coefficients (DSC), and 95% Hausdorff distances (95% HD), added path length (APL) and surface DSC (sDSC) were calculated in a caudal-cranial window of ± 4 cm with respect to the prostate ends. For qualitative evaluation, five radiation oncologists scored the AIC on the possible usage within an online adaptive workflow as follows: (1) no modifications needed, (2) minor adjustments needed, (3) major adjustments/ multiple minor adjustments needed, (4) not usable. ResultsThe quantitative evaluation revealed a maximum median 95% HD of 6.9 mm for the rectum and minimum median 95% HD of 2.7 mm for the bladder. Maximal and minimal median DSC were detected for bladder with 0.97 and for penile bulb with 0.73, respectively. Using a tolerance level of 3 mm, the highest and lowest sDSC were determined for rectum (0.94) and anal canal (0.68), respectively.Qualitative evaluation resulted in a mean score of 1.2 for AICs over all organs and patients across all expert ratings. For the different autocontoured structures, the highest mean score of 1.0 was observed for anal canal, sacrum, femur left and right, and pelvis left, whereas for prostate the lowest mean score of 2.0 was detected. In total, 80% of the contours were rated be clinically acceptable, 16% to require minor and 4% major adjustments for online adaptive MRgRT. ConclusionIn this study, an AI-based autocontouring was successfully trained for online adaptive MR-guided radiotherapy on the 1.5 T MR-Linac system. The developed model can automatically generate contours accepted by physicians (80%) or only with the need of minor corrections (16%) for the irradiation of primary prostate on the clinically employed sequences.

PO-2116 Manual or automatic heart contours do not lead to difference in predicted survival in NSCLC patients

PO-2116 Manual or automatic heart contours do not lead to difference in predicted survival in NSCLC patients

Validation of a Fully Automated Hybrid Deep Learning Cardiac Substructure Segmentation Tool for Contouring and Dose Evaluation in Lung Cancer Radiotherapy

Background and purposeAccurate and consistent delineation of cardiac substructures is challenging. The aim of this work was to validate a novel segmentation tool for automatic delineation of cardiac structures and subsequent dose evaluation, with potential application in clinical settings and large-scale radiation-related cardiotoxicity studies. Materials and methodsA recently developed hybrid method for automatic segmentation of 18 cardiac structures, combining deep learning, multi-atlas mapping and geometric segmentation of small challenging substructures, was independently validated on 30 lung cancer cases. These included anatomical and imaging variations, such as tumour abutting heart, lung collapse and metal artefacts. Automatic segmentations were compared with manual contours of the 18 structures using quantitative metrics, including Dice similarity coefficient (DSC), mean distance to agreement (MDA) and dose comparisons. ResultsA comparison of manual and automatic contours across all cases showed a median DSC of 0.75–0.93 and a median MDA of 2.09–3.34 mm for whole heart and chambers. The median MDA for great vessels, coronary arteries, cardiac valves, sinoatrial and atrioventricular conduction nodes was 3.01–8.54 mm. For the 27 cases treated with curative intent (planned target volume dose ≥50 Gy), the median dose difference was –1.12 to 0.57 Gy (absolute difference of 1.13–3.25%) for the mean dose to heart and chambers; and –2.25 to 4.45 Gy (absolute difference of 0.94–6.79%) for the mean dose to substructures. ConclusionThe novel hybrid automatic segmentation tool reported high accuracy and consistency over a validation set with challenging anatomical and imaging variations. This has promising applications in substructure dose calculations of large-scale datasets and for future studies on long-term cardiac toxicity.

Abstract P1-10-12: AI-based cardiac sub-structures segmentation for safer radiotherapy planning

Abstract Purpose: Whilst radiotherapy increases cure rates in breast cancer, lung cancer, among others, it may also involve some cardiac exposure, which in turn may increase the risk of different heart diseases. The heart is a complex anatomical organ that involves many different structures making it difficult to contour cardiac sub-structures reproducibly. Contouring, especially for these cases, suffers from inter- and intra-expert variability while being time consuming. Cardiac atlases have been developed to aid in the delineation of cardiac substructures. However, these methods have many shortcomings, including the inability to overcome variations in patient anatomy. In this study, a deep learning based commercial solution for automatic OAR delineation was trained following international guidelines for heart substructures delineation and tested on an unseen cohort of lung and breast patients to evaluate its clinical acceptability. Methods: ART-Net, a CE-marked, FDA-cleared anatomically preserving deep-learning ensemble architecture for automatic annotation of OAR was evaluated using data of 20 breast/lung patients from 2 centers. Automatic annotation of 27 different structures (Ventricles (left and right), atria (left and right), left ventricle (anterior, apical, inferior, lateral, septal), LAD coronary (mid, proximal, distal, total), circumflex coronary (distal, proximal, total), RCA (distal, mid, proximal, total), coronary sinus, left main coronary artery, ascending aorta, pulmonary arteries, vena cava inferior, vena cava superior and the heart) was performed and submitted to 2 experts across 2 centers for qualitative evaluation. Contours were scored as A/acceptable, B/acceptable after minor corrections, and C/not acceptable for clinical use. To avoid any bias, experts were blind to whether the contour were manually, or AI delineated. The DSC between automatic and manual (ground truth) contours of the heart sub-structures were evaluated and compared with interobserver variability from the literature [1,2] using average and min DSC scores. Results: Automatic contours were generated in a mean time of 0.5s per scan slice. Out of the 27 structures, 20 were considered clinically acceptable in the qualitative study. In the inter-expert variability study, 12 structures passed the test successfully using initial acceptance criterion over an acceptable sample size and 9 other structures demonstrated performances above the minimal threshold of inter-expert variability, sometimes on smaller datasets due to lack of manual data. Overall, 16 structures were included in the final model. 13 structures were considered clinically acceptable in 100% of the cases with AI contours rated at the same level as manual contours. For the other 3 structures (coronary sinus, left main coronary artery and vena cava inferior), the performance of the AI contours was slightly below that of the manual contours (within 3.4% difference), with the least performing structure being the coronary sinus (84% for AI vs 87% manual). Conclusion: We show first results for the evaluation of AI-based auto-contouring tool for annotation of the substructures of the heart. The results show very good clinical acceptance, highlighting the high usability of the commercial tool for cardiac cases and its clinical implementation feasibility. The use of this AI tool can facilitate and accelerate future research studies investigating relationships between substructure doses and cardiac outcomes. Future work will include improvement of the sub-structures (mid, proximal, distal) and a retrospective meta-analysis to assess heart sub-structures degree of importance in terms of toxicity.