Random forest modeling of acute toxicity in anal cancer: Effects of peritoneal cavity contouring approaches on model performance.
Researchers
Journal
Modalities
Models
Abstract
To analyze the impact on gastrointestinal (GI) toxicity models when their dose-volume metrics predictors are derived from segmentations of the peritoneal cavity following different contouring approaches.A random forest machine learning approach was utilized to predict acute grade 3+ GI toxicity from dose-volume metrics and clinicopathological factors for 246 patients (toxicity incidence = 9.5%) treated with definitive chemoradiation for squamous cell carcinoma of the anus (SCCA). Three types of random forest models were constructed based on different bowel bag segmentation approaches: 1) physician-delineated following Radiation Therapy Oncology Group (RTOG) guidelines 2) auto-segmented by a deep learning model (nnU-Net) following RTOG guidelines, and 3) auto-segmented, but spanning the entire bowel space. Each model type was evaluated using repeated cross-validation (100 iterations; 50%/50% training/test split). The performance of the models was assessed using area under the precision-recall curve (AUPRC) and the receiver operating characteristic curve (AUROCC), as well as optimal F1-Score.When following RTOG guidelines, the models based on the nnU-Net auto-segmentations (mean values: AUROCC = 0.71±0.07 AUPRC = 0.42±0.09; F1-Score = 0.46±0.08) significantly outperformed (p<0.001) those based on the physician-delineated contours (mean values: AUROCC = 0.67±0.07; AUPRC = 0.34±0.08; F1-Score = 0.36±0.07). When spanning the entire bowel space, the performance of the auto-segmentation models improved considerably (mean values: AUROCC = 0.87±0.05 AUPRC = 0.70±0.09; F1-Score = 0.68±0.09).Random forest models were superior at predicting acute G3+ GI toxicity when based on RTOG defined bowel bag auto-segmentations rather than physician-delineated contours. Models based on auto-segmentations spanning the entire bowel space show further considerable improvement in model performance. The results of this study should be further validated using an external dataset.Copyright © 2023. Published by Elsevier Inc.