Generative AI Outperforms in Lung Adenocarcinoma Grading and Prognosis
July 30, 2025
GenAI models demonstrate high accuracy and consistency in pathological grading and risk assessment for lung adenocarcinoma.
Key Details
- Study tested GPT-4o, Claude-3.5-Sonnet, and Gemini-1.5-Pro on 492 digitized slides from public and independent sources.
- Claude-3.5-Sonnet achieved 82.3% accuracy in lung adenocarcinoma grading, with strong repeatability.
- Models extracted 11 histological features and 4 clinical variables for comprehensive risk assessment.
- Diagnostic and prognostic tasks were completed in minutes, much faster than human experts.
- AI models reduced inter-observer variability and enabled objective, standardized analysis across multiple samples.
- GenAI models surfaced new significant prognostic factors including interstitial fibrosis and papillary patterns.
Why It Matters
This breakthrough underscores the transformative potential of generative AI in digital pathology, promising faster, standardized, and potentially more insightful cancer diagnosis and prognosis translation. Such advances could boost accuracy in resource-limited settings and improve reproducibility in pathological assessments, reshaping clinical workflows.