Urethra contours on MRI: Multidisciplinary consensus educational atlas and reference standard for artificial...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate urethra identification is crucial for prostate cancer treatment but challenging for physicians.
• Existing automated segmentation tools lack reliable validation due to insufficient ground truth and evaluation standards.

Purpose of the Study:

• Establish a reference-standard dataset for urethra segmentation using expert consensus.
• Define clinically relevant evaluation metrics for segmentation accuracy.
• Assess the performance and generalizability of a deep-learning-based segmentation model.

Main Methods:

• A multidisciplinary panel created consensus urethra contours on MRI data from 71 patients across 6 centers.
• A deep-learning AI model was developed using an independent training dataset (n=151).
• AI performance was evaluated against the reference standard and human performance using Dice, coverage, and Hausdorff Distance (HD).

Main Results:

• The AI model achieved a median Dice of 0.40 and 89% coverage, outperforming the average physician in a subset of cases.
• AI performance remained consistent across the full reference dataset, demonstrating generalizability.
• The AI model showed a Max 2D HD of 2.0 mm, indicating high accuracy.

Conclusions:

• A consensus benchmark for urethra segmentation was established.
• The deep-learning model demonstrates comparable performance to specialist physicians.
• The AI tool shows potential as a clinical decision-support system for precise urethra segmentation in prostate cancer radiotherapy.