Pathology

H&E-based deep learning predicts actionable NSCLC biomarkers with high accuracy

This study validated deep learning classifiers for EGFR, ALK, BRAF, and MET alterations directly from H&E-stained NSCLC tissue using a foundation model. In 968 independent NSCLC samples, AUCs were 0.87 for EGFR, 0.96 for ALK, 0.88 for BRAF, and 0.83 for MET. The algorithms also accurately flagged tumors without these alterations, directly relevant for immunotherapy eligibility and test triage. These results support using H&E-based AI as a rapid pre-screen to prioritize confirmatory NGS or FISH in NSCLC workflows. 1

Interpretable H&E-based AI classifies endometrial cancer molecular subtypes

An end-to-end deep learning model on H&E whole-slide images predicted four clinically relevant endometrial cancer molecular subtypes. Trained on 364 cases with validation in two external cohorts, the network achieved a macro-average AUROC of 0.867 in cross-validation. Subtype AUROCs ranged from 0.835 for POLEmut to 0.910 for p53-abnormal tumors, indicating robust discrimination. Model-derived morphologic patterns aligned with known subtype features, enhancing interpretability and potential integration into diagnostic workflows. 2

NGS-based HER2 DNA and RNA assays partly align with IHC across solid tumors

Over 78,000 solid tumors were profiled for HER2 DNA copy number, mRNA expression, and protein overexpression using whole-exome sequencing, whole-transcriptome sequencing, and IHC. In the US cohort, HER2 DNA amplification, RNA overexpression, and IHC positivity occurred in 4.9%, 10.1%, and 4.7% of tumors, respectively. Only 3.9% of tumors were positive by all three modalities, while 10.7% were positive in at least one assay, underscoring incomplete concordance. Using IHC as reference, RNA testing showed higher sensitivity than DNA copy number but lower positive predictive value. Correlation between DNA, RNA, and protein varied by tumor type, and heterogeneous HER2 IHC distribution contributed to discordances, limiting direct substitution of IHC. 3

Tissue-free methylation ctDNA assay shows strong MRD performance in colorectal cancer

This GALAXY sub-study validated a tissue-free, methylation-based ctDNA assay for MRD in 195 colorectal cancer patients across 1230 timepoints. CtDNA positivity was strongly associated with worse disease-free survival in both MRD and surveillance windows, with large reported hazard ratios. In the MRD window, sensitivity was 58.5%, and specificity in patients without adjuvant chemotherapy reached 100%. Longitudinal surveillance achieved 84.4% relapse sensitivity with high patient- and sample-level specificity and a median 4.6-month lead time. High-risk stage II–III ctDNA-positive patients appeared to benefit from adjuvant chemotherapy, unlike ctDNA-negative patients. 4