Reinforcement learning-guided segment anything model for MRI prostate and dominant intraprostatic lesions auto-segmentation.
Authors
Affiliations (9)
Affiliations (9)
- Department of Radiation Oncology and Winship Cancer Institute, Emory University, 36 Linden Ave NE, Atlanta, Georgia, 30308, UNITED STATES.
- Department of Radiation Oncology, Emory University Atlanta, 1365 Clifton Road NE, Atlanta, Georgia, 30322, UNITED STATES.
- Department of Radiation Oncology and Winship Cancer Institute, Emory University, 1365 Clifton Rd NE, Atlanta, Georgia, 30322, UNITED STATES.
- Department of Radiology Oncology, Emory University, 1365 Clifton Road NE, Atlanta, Georgia, 30322, UNITED STATES.
- Emory University, 1365-C Clifton Road NE, Atlanta, Georgia, 30322, UNITED STATES.
- Department of Radiology and Sciences Imaging Department of Radiology Oncology, Emory University, 36 Linden Ave NE, Atlanta, Georgia, 30308, UNITED STATES.
- Department of Radiation Oncology, Emory University School of Medicine, 2015 Uppergate Dr, Atlanta, Georgia, 30303-3073, UNITED STATES.
- Department of Radiology and Sciences Imaging Department of Radiology Oncology, Emory University, 1365 Clifton Rd NE, Atlanta, Georgia, 30322, UNITED STATES.
- Department of Radiology Oncology, Emory University, Clifton Rd, Atlanta, Georgia, 30322-1007, UNITED STATES.
Abstract
Objective
Accurate segmentation of the prostate and dominant intraprostatic lesions (DILs) on magnetic resonance imaging (MRI) is important for prostate cancer radiation therapy treatment planning and targeted dose escalation. However, DIL segmentation remains challenging due to small datasets, institutional bias, and variable imaging protocols. Although the Segment Anything Model (SAM) has shown promise in medical image segmentation, most prior work depends on manual prompts. This study developed a fully automated pipeline that combines localization with a fine-tuned SAM model to segment the prostate and DIL.

Approach
Two datasets were utilized: the PI-CAI dataset, comprising 1,476 patients, and the Cancer Imaging Archive dataset, comprising 803 patients. The pipeline consisted of two stages: (1) a reinforcement learning-based localization network predicted bounding boxes as segmentation inputs, and (2) a fine-tuned SAM model performed segmentation. Model performance was evaluated using the Dice Similarity Coefficient (DSC), Intersection over Union (IoU), and detection rates, with additional analysis based on lesion volumes.

Main Results
The proposed method achieved a mean and median DSC of 0.896±0.070 and 0.915, and an IoU of 0.818±0.100 and 0.844 for prostate segmentation. For DIL segmentation, the mean and median DSC were 0.592±0.192 and 0.636, IoU of 0.446±0.190 and 0.466, with a detection rate of 89%. Four DIL groups were created based on lesion volume percentile. The mean/median DSC and IoU for each volume group is as follows: 0.5-1.0 cubic centimeters (cc): 0.555±0.201/0.562 & 0.414±0.205/0.391; 1.0-1.8 cc: 0.603±0.185/0.660 & 0.454±0.180/0.492; 1.8-4.0 cc: 0.588±0.183/0.627 & 0.439±0.174/0.456; >4.0 cc: 0.621±0.197/0.669 & 0.477±0.197/0.503. 

Significance
This study presented a fully automated prostate and DIL segmentation framework on MRI by integrating a localization network with fine-tuned SAM. The method achieved robust performance across large multi-institutional datasets and diverse lesion shapes. It shows strong potential for application to clinical workflows for prostate cancer radiation therapy planning and treatment.