Auto-delineation of the hippocampus using Mask R-CNN for radiation oncology: A cross-sectional observational study

ABSTRACT Objective: Radiotherapy is an effective treatment for brain tumors, but it can lead to adverse side effects, particularly neurocognitive dysfunction. The hippocampus, a crucial structure for memory formation, is vulnerable to radiation-induced damage. To reduce the risk of adverse effects, the precise delineation of the hippocampus is necessary for radiation therapy planning. Automated hippocampal delineation using deep learning and mask regions with convolution neural networks (R-CNN) algorithm reduces the manual delineation time, ensures accuracy, and supports its implementation in radiation therapy. This study aimed to develop an automated method for hippocampal delineation and assess its feasibility for brain radiation treatment using deep learning and Mask R-CNN algorithm. Materials and Methods: The study applied several pre-processing techniques, including thresholding, Canny edge detection, and keypoint detection, to enhance the input images of patient hippocampal scans. The processed images were then analyzed using the Mask R-CNN algorithm for automated hippocampal delineation. The Mask R-CNN algorithm was assessed using the following metrics: accuracy, recall, intersection over union (IoU), and mean average precision (mAP). Results: The proposed network achieved a mAP of 96.2% and average sensitivity of 0.93, indicating high segmentation accuracy. Moreover, the recall was 0.87 and specificity was 0.96, indicating the high specificity of the method to segment the hippocampus. Conclusions: The results of this study suggest that the developed automated method is highly sensitive and specific for delineating the hippocampus. This automated segmentation method may be used to complement the manual review of hippocampal scans during radiotherapy planning for brain tumors.