Objective:

To explore how AI can extract structured data from pathology reports, focusing on improving consistency and accuracy in laboratory workflows by addressing specific challenges.

Key Findings:

• No single model outperformed across all variables; model diversity enhances performance, indicating the need for a multi-faceted approach.
• High extraction accuracy was achieved on TCGA reports across multiple cancer types, demonstrating the framework's robustness.
• Real-world reports from Moffitt Cancer Center showed comparable or higher accuracy than TCGA, despite added complexity, highlighting the framework's adaptability.

Interpretation:

The study indicates that AI can effectively structure complex pathology reports, but challenges remain, particularly with biomarker extraction due to variability in documentation practices and reporting styles.

Limitations:

• Biomarker extraction accuracy was lower due to non-standardized reporting and information distribution, which complicates AI interpretation.
• Performance may vary based on specific variables and organ contexts, necessitating tailored approaches for different scenarios.

Conclusion:

AI-driven structured extraction can support precision medicine by ensuring timely access to critical diagnostic and molecular information, ultimately improving patient outcomes.