Manual prostate MRI segmentation by readers with different experience: a study of the learning progress

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence

Background:

• Prostate MRI segmentation is crucial for diagnosis and treatment planning.
• Evaluating the learning curve of less experienced readers is important for efficient workflow and AI development.

Purpose of the Study:

• To assess the learning progress and segmentation accuracy of readers with varying experience levels in prostate MRI.
• To determine the suitability of less experienced readers for creating datasets for AI development.

Main Methods:

• Retrospective analysis of 100 bi-parametric prostate MRI scans.
• Nine readers (expert, residents, novices) performed prostate segmentation.
• Iterative feedback sessions and performance evaluation using Dice Similarity Coefficient (DSC).

Main Results:

• Radiology residents achieved 99.2% accuracy (DSC > 0.8).
• Novices improved from 47.5% in round 1 to 84.0% by round 3.
• Novices demonstrated significant interindividual variation in segmentation accuracy.

Conclusions:

• Radiology residents, regardless of experience, provide high accuracy for prostate MRI segmentation.
• Novices show lower accuracy and higher variability, making them less ideal for AI dataset preparation.
• Employing residents offers a cost-effective balance for AI dataset creation, while novices require individual consideration.