Impact of Preanalytical and Analytical Factors on AI - based Tumor Cell Content Quantification for Molecular Tumor Profiling: A Real -World Study

University Hospital Zurich(USZ) conducted a real-world evaluation of AI-based tumor cell content (TCC) quantification in routine molecular profiling, analyzing 283 digitized H&E samples across five cancer types. The study reported correlations of Rs=0.41 between pathologist and AI estimates, Rs=0.48 between pathologist and bioinformatics estimates, and Rs=0.61 between AI and bioinformatics estimates, across cancer origins and sample types. The authors conclude that high cellularity (tumor and non-tumor), technical artifacts, and necrosis contribute to discrepancies between pathologist vs AI and bioinformatics vs AI assessments and can reduce the overall performance of AI-based TCC estimation; they also emphasize that pre-analytical standardization and considerations such as specimen type and cellularity are central to implementing AI-based TCC assessment in molecular pathology workflows.

SGPath/SSPath 2025