Performance of ChatGPT-4o in Determining Radiology-Pathology Concordance and Management Recommendations Following...

Area of Science:

• Artificial Intelligence in Medicine
• Radiology and Pathology
• Breast Imaging and Diagnostics

Background:

• Accurate radiology-pathology concordance is crucial for breast cancer patient management.
• Limited access to expert multidisciplinary input poses a challenge in some settings.

Purpose of the Study:

• To assess the capability of ChatGPT-4o in determining radiology-pathology concordance for breast biopsies.
• To evaluate ChatGPT-4o's suggested management strategies based on biopsy results.

Main Methods:

• A retrospective analysis of 244 image-guided breast biopsy cases was conducted.
• ChatGPT-4o evaluated de-identified radiology and pathology reports for concordance and management recommendations.
• Radiologist assessments were used as a reference standard, with surgical pathology and imaging follow-up as gold standards.

Main Results:

• ChatGPT-4o achieved a high concordance rate (98.8%), comparable to radiologists (98.0%).
• The AI demonstrated strong diagnostic agreement with the gold standard (Cohen's kappa = 0.947).
• ChatGPT-4o suggested imaging follow-up more frequently and surgical management less frequently than radiologists.

Conclusions:

• ChatGPT-4o exhibits diagnostic performance on par with subspecialist radiologists for breast biopsy concordance.
• The AI's more conservative management approach may support shared decision-making, particularly in resource-limited environments.