Performance of an Artificial Intelligence Foundation Model for Prostate Radiotherapy Segmentation.

Area of Science:

• Medical Physics
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate target segmentation is crucial for effective prostate cancer radiation therapy.
• Artificial intelligence (AI) offers potential for automating and improving segmentation tasks.

Purpose of the Study:

• To evaluate the performance of a general-purpose AI foundation model, Segment Anything Model 2 (SAM 2), for prostate radiotherapy target segmentation.
• To assess the impact of varying levels of human intervention on AI segmentation accuracy.

Main Methods:

• AI segmentation using SAM 2 was performed on computed tomography (CT) images.
• Segmentation accuracy was evaluated using Dice Similarity Coefficient (DSC) and 95th percentile Hausdorff Distance (HD95).
• Performance was assessed across different intervals of ground truth slices (every 2nd to 10th slice) for both intact and postoperative prostate cases.

Main Results:

• SAM 2 performance was comparable to or worse than interpolation methods for both intact and postoperative prostate segmentation.
• AI segmentation accuracy was significantly better for intact preoperative cases (P < .01) compared to postoperative cases.
• Increasing the interval between ground truth slices reduced DSC and increased HD95, particularly in postoperative cases.

Conclusions:

• Current general-purpose AI foundation models are inadequate for prostate radiotherapy segmentation.
• Further research is needed on fine-tuning and task-specific AI models for clinical application in radiotherapy planning.