Thinking like a pathologist: Morphologic approach to hepatobiliary tumors by ChatGPT

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Hepatobiliary pathology

Background:

• Accurate diagnosis of hepatobiliary tumors is crucial for patient management.
• Histopathologic image analysis is a cornerstone of liver tumor diagnosis.
• Artificial intelligence (AI) offers potential to augment diagnostic capabilities.

Purpose of the Study:

• To evaluate the diagnostic performance of ChatGPT (GPT-4) in identifying hepatobiliary tumors using histopathologic images.
• To compare the effectiveness of different input prompts for AI-driven histopathologic diagnosis.

Main Methods:

• The study utilized 120 histopathologic images of hepatobiliary tumors and nonneoplastic liver tissue.
• Two input prompts were tested with the GPT-4 model: one mimicking pathologist's morphologic analysis and another without.
• Diagnostic accuracy and consistency were statistically analyzed.

Main Results:

• The AI model using a morphologic approach significantly outperformed the model without it in diagnosing various hepatobiliary tumors and nonneoplastic tissues.
• Specific improvements were noted for hepatocellular carcinoma, bile duct adenoma, and cholangiocarcinoma.
• Diagnostic consistency was substantially higher with the morphologic approach (κ=0.46 vs 0.27).

Conclusions:

• Integrating pathologist-like diagnostic strategies into AI enhances accuracy and consistency in histopathologic interpretation.
• AI, particularly when trained on expert diagnostic processes, demonstrates significant potential in medical diagnostics.
• This study highlights the value of human-AI collaboration in pathology.