Pathology

Pathology-derived micro-architectural RMELS stratifies colorectal tumours beyond stage

Using rule-based NLP on 1,978 TCGA colorectal pathology reports, authors derived barrier-disruption and invasion-access features linked to microbial activity. They combined features into the Report-based Microbial Ecology Likelihood Score (RMELS), a z-scored index that ordered tumours along a microbiome-permissiveness gradient independent of stage. RMELS discriminated early (T1) from advanced (T4) disease with AUC 0.66 and showed modest associations with overall and progression-free survival in multivariable models. Features were prevalent (infiltrative growth 59.4%, ulceration 41.1%) and varied non-monotonically across pathologic stages. 1

Systematic review: AI on H&E predicts neoadjuvant response but needs stronger validation

This systematic review synthesized 25 studies applying AI to H&E slides to predict neoadjuvant therapy response across solid tumours. Reported model AUCs clustered between 0.70–0.90, but only about 40% of studies included external validation cohorts. Quality assessment (QUADAS-2) identified heterogeneity in data acquisition, lack of patient-level validation, and limited transparency. Authors conclude AI is promising but needs standardized workflows, data sharing, and patient-level external validation before clinical adoption. 2

Expert guidance to maximise research value of small biopsy tissue

An expert panel summarized best practices to mitigate limited tissue and low analyte yields from small biopsies for molecular research. Recommendations emphasize prebiopsy scoring, clear cross-team communication, standardized collection and handling, and tumor enrichment techniques. The guidance draws on optimized workflows from multicenter trials to improve reproducible molecular data generation from scarce specimens. Implementing these steps can increase the likelihood of producing assay‑appropriate biopsy material for translational studies. 3