Comparing AI and Manual Segmentation of Prostate MRI: Towards AI-Driven 3D-Model-Guided Prostatectomy

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Surgical Technology

Background:

• Robot-assisted radical prostatectomy (RARP) is a standard treatment for prostate cancer.
• Three-dimensional (3D) models from MRI aid surgical guidance, but manual segmentation is time-consuming and variable.
• Automated segmentation methods are needed to improve efficiency and consistency.

Purpose of the Study:

• To evaluate an AI tool for automated and AI-assisted prostate and tumor segmentation in MRI for RARP.
• To compare AI performance against manual segmentation using Dice Coefficient, Hausdorff distance, recall, and precision.

Main Methods:

• A commercially available AI tool was used for fully automated and AI-assisted segmentation on 120 patient MRIs.
• Performance was compared to manual segmentation by expert radiologists.
• Segmentation accuracy and tumor detection rates were quantified.

Main Results:

• AI-assisted segmentation yielded improved Dice scores (0.62) and lower Hausdorff distance (6.62 mm) for tumors compared to fully automated (0.53 Dice, 9.53 mm Hausdorff).
• AI-assisted tumor detection showed high recall (0.95) and precision (0.94).
• AI-assisted segmentation required only slightly more time than fully automated, significantly less than manual segmentation.

Conclusions:

• Fully automated AI segmentation offers promising tumor detection and acceptable metrics.
• AI-assisted segmentation significantly enhances accuracy and detection rates with minimal time increase.
• AI-assisted segmentation is a viable approach for 3D-model-guided RARP, improving surgical planning and outcomes.