Large language models demonstrate promising performance in automating PI-RADS classification from structured prostate MRI reports, with some limitations in intermediate-risk lesions.
Key Details
- 1Study included 146 structured prostate MRI reports from October 2023 to October 2024.
- 2Four LLMs compared: GPT-4o, GPT-o1, Google Gemini 1.5 Pro, Google Gemini 2.0 Experimental Advanced.
- 3Radiologist consensus used as ground truth; Cohen's kappa measured agreement.
- 4GPT-o1 achieved the highest agreement (kappa = 0.87) and perfect F1 score (1.00) for high-risk PI-RADS category.
- 5All LLMs struggled with PI-RADS 3 (equivocal risk) category (F1 scores 0.53–0.75).
- 6Authors recommend further multicenter validation and larger datasets before clinical adoption.
Why It Matters
Automating PI-RADS classification could streamline prostate MRI reporting, reduce variability, and enhance workflow efficiency for radiologists. However, challenges in intermediate-risk lesion classification underscore the need for more robust validation before real-world clinical use.
Source
AuntMinnie
Related News
•AuntMinnie
Radiologists Struggle to Spot AI-Generated Radiology Images
Radiologists correctly identify AI-generated images 75% of the time, with CT and MRI images being particularly challenging to spot.
•AuntMinnie
Radiology Leads FDA AI Device Approvals Over Three Decades
Radiology accounts for 76% of all FDA-cleared AI/ML-enabled medical devices as of the end of 2025.
•Radiology Business
Automation Bias: How AI Can Compromise Radiologist Accuracy
AI decision support can induce automation bias, leading radiologists to accept incorrect interpretations and potentially reduce their diagnostic accuracy.