AI Algorithm for Lung Adenocarcinoma Pattern Quantification (PATQUANT): International Validation and Advanced Risk...

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Morphological patterns in lung adenocarcinoma (LUAD) are crucial for prognosis, yet optimal grading remains debated.
• Current grading systems for LUAD require refinement for improved accuracy and prognostic value.

Purpose of the Study:

• To develop and validate a fully automated, quantitative AI tool (PATQUANT) for LUAD pattern classification.
• To evaluate existing LUAD grading strategies and identify the most effective system.
• To propose and validate novel, explainable grading principles for enhanced patient risk stratification.

Main Methods:

• Training PATQUANT on a pathologist-annotated dataset for LUAD pattern classification.
• Validating PATQUANT using independent test datasets and comparing its performance against 13 expert pathologists.
• Analyzing five multinational cohorts (n=1120) of resectable LUAD to assess the prognostic value of identified patterns and grading systems.

Main Results:

• PATQUANT achieved high accuracy in LUAD pattern segmentation and classification, surpassing 8 out of 13 pathologists.
• The complex glandular pattern in LUAD showed a distinct prognostic profile.
• Predominant pattern-based and simplified IASLC grading systems demonstrated superior prognostic value compared to others.
• Two new, explainable grading principles were validated, offering fine-grained, independent risk stratification.

Conclusions:

• The developed AI tool, PATQUANT, offers a robust and automated solution for LUAD pattern analysis, exceeding expert pathologist performance.
• Novel grading approaches, informed by AI-driven pattern quantification, provide superior prognostic capabilities over traditional methods.
• Publicly releasing the agreement dataset will foster further advancements in LUAD grading and analysis.