Abstract Book

S1187

ESTRO 37

Conclusion Atlas patients containing contour probability maps can provide a measure of accuracy for RADAR CTV contours. Investigations using a 50% probability map revealed DSC = 0.55 to be predictive of local regression, but not overall survival or rectal bleeding. Due to DSCs inability to differentiate between contours that lie within or outside the true volume, further analysis using additional contouring metrics may reveal more significant correlations. EP-2151 Personalized Deep Learning-Based Auto- Segmentation V. Kearney 1 , S. Haaf 2 , J. Chan 1 , S. Wu 1 , M. Bogdanov 1 , S. Yom 1 , T. Solberg 1 1 University of California UCSF, department of radiation oncology, San Francisco CA, USA 2 Nimble Therapy- LLC, Department of Artificial Intelligence, San Francisco, USA Purpose or Objective To demonstrate the feasibility of a personalized deep learning-based auto-segmentation system. Material and Methods Deep learning-based auto-segmentation was evaluated for the spinal cord, brainstem, left parotid, right parotid and mandible in 102 head and neck cancer patients. Each patient contains two ground-truth structure sets, one previously contoured in the clinic under normal working conditions (Human-1), the other carefully contoured by a single physician (Human-2). 3 different training variations of a deep learning-based automatic contouring system, Nimble Contour TM (Nimble Therapy, San Francisco, CA) were investigated using 13 validation patients and 89 training patients. 3 physicians and 1 dosimetrist participated in this study. Prior to training, the clinicians submitted scores for each contour on every patient in Human-1. Variations were trained on Human-2 (AI-1) and Human-1 (AI-3), and a third was trained on Human-1 incorporating the aggregated scores into the training phase (AI-2). For validation, each clinicians graded all human and AI data sets side by side, without knowledge of which data set they were grading (Figure 1). Scores ranging from 0 to 100 were assigned based on individual clinician preferences. Results Averaging over all the contours for all the patients, 1 clinician ranked AI-1 the highest and 3 clinicians ranked AI-1 the second highest (Table 1). 2 clinicians ranked AI-2 the 3 rd highest. On average AI-3 had the lowest performance but was only slightly lower than Human-2. Conclusion Clinician styles and preferences tend to vary considerably within a clinic. However, this study has demonstrated that by using clinician scoring of retrospective data, or carefully contoured prospective data, a deep learning auto-contouring model can be created that outperforms the original clinical data.

Physics, Wollongong, Australia 6 Sir Charles Gairdner Hospital, Radiation Oncology, Nedlands, Australia 7 Liverpool and Macarthur Cancer Therapy Centres, Department of Radiation Oncology, Liverpool, Australia 8 Calvary Mater Newcastle Hospital, Radiation Oncology, Newcastle, Australia 9 University of Western Australia, School of Physics and Astrophysics- Faculty of Science, Crawley, Australia Purpose or Objective Development of a probabilistic prostate atlas can allow for secondary analysis of the large RADAR patient dataset. This study investigated whether clinical target volume (CTV) contouring variations for prostate cancer could be a predictor for local disease progression, overall survival and rectal toxicity. Material and Methods Five prostate cancer patient datasets each had CTV contoured on planning CT by ten observers, with probability maps constructed using observer contours. Patient scans were used to construct a pelvic atlas for retrospective analysis of the larger RADAR dataset, containing 711 patients that underwent external-beam radiotherapy for prostate cancer treatment. RADAR patients without prostate only CTV contours were excluded from the study. Remaining RADAR patients each had a single atlas patient, selected based off prior clustering work, undergo rigid and deformable (Demons non-rigid) registration to allow propagation of probability maps onto the RADAR patient. Probability maps were thresholded at 50% observer agreement, and spatial overlap of probability maps and RADAR CTVs were assessed using Dice similarity coefficient (DSC). Correlations with time to local progression and overall survival were calculated using Cox proportional hazards regression. Correlations with grade 2 and higher rectal bleeding were assessed using Wilcoxon signed-rank test. Results A total of 461 patients from the RADAR dataset contained prostate only CTV contours. Mean, median, and standard deviation DSC across these patients were 0.595, 0.621, and 0.158 respectively. Mean time to local progression for these patients was 76.48 months. Following Cox regression, a statistically significant difference (p = 0.0444) for local progression was found for RADAR patients with DSC > 0.55, with b = 3.521 and hazard ratio of 33.819. Figure 1 illustrates hazard rate estimates for local progression between patients stratified by DSC = 0.55. No statistically significant relationships between DSC and overall survival were found. Within the RADAR subset 214 patients experienced no rectal bleeding following radiotherapy treatment. 122, 80, 35, and 6 patients experienced grade 1, 2, 3, and 4 rectal bleeding respectively. Wilcoxon signed-rank analysis of DSC between patients with and without rectal bleeding (grade >= 2; grade < 2) revealed no statistically significant difference between DSC medians (p = 0.2531).

Figure 1: Hazard function for local progression for RADAR patients containing prostate CTV contours.

Figure 1. The contour prediction map for AI-1 is presented for the left parotid, right parotid, and spinal cord (A). The Human-2 (H-2) and AI-1 contours are