Real-world analysis of manual editing of deep learning contouring in the thorax region

Clinical Implementation and Evaluation of Auto-Segmentation Tools for Multi-Site Contouring in Radiotherapy

“…However, in some organs at risk (OARs), the results were systematically inferior and did not meet the clinical requirements of radiation oncologists. This agrees with previous studies [6] , [10] , [12] , [15] , [16] .…”

“…Our overall positive conclusion regarding the clinical implementation of deep learning-based auto-segmentation is in line with recently published literature. Our study covers all clinically relevant sites, compared to more narrow studies for head-and-neck [6] , [10] , [12] , breast [13] , prostate [12] , [14] , thorax [15] , [16] , central nervous system [12] , or cervix [17] .…”

“…Therefore, performing an extensive evaluation of auto-segmentation prior to clinical implementation to understand their accuracy and limitations is crucial. Previous studies have focused on single sites [6] , [10] , [12] , [13] , [14] , [15] , [16] , [17] . This study aimed to develop a comprehensive procedure for evaluating AI-based auto-segmentation software for all sites before clinical implementation.…”

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

“…Very few studies have clearly compared performances of AI-based auto-contouring against manual delineation [25] , or against inter-institution variations [24] . Reasonable editing is expected especially for CTV, at least in a fraction of patients, as already shown in other scenarios [26] .…”

Challenges and chances for deep-learning based target and organ at risk segmentation in radiotherapy of head and neck cancer

“…Disentangling the problem of inter-observer variation (IOV) from the accuracy of DLS tools is challenging, as the manual contour is the only “truth” available, and it is dependent on who has done the segmentation. With more and more DLS tools for OARs being used in the clinics, it now becomes possible to assess auto-segmentation accuracies in clinical practice [10] , [11] . By assessing the adaptations of the auto-contours, insight is obtained into where observers disagree with the DLS tools.…”