Pathology

Routine tumor–normal whole-genome sequencing delivers high biomarker yield and survival gains in solid tumors

In 888 solid tumor patients, paired tumor–normal WGS succeeded in 89% with a median 6‑working‑day turnaround. Actionable biomarkers were found in 73%, including markers for reimbursed therapies in 27% and experimental options in 63%. Within one year, 40% and 19% started reimbursed or experimental biomarker-informed therapies, respectively. Biomarker-informed treatment correlated with 31% longer median overall survival than non–biomarker-informed management. Among treatment‑naïve patients, biomarker-informed therapy produced significantly longer overall survival than either non–biomarker-based systemic therapy or no systemic therapy. In cancers of unknown primary, WGS aided diagnosis or uncovered reimbursed targeted options in 67%, and clinically relevant germline variants appeared in 6.5% of all patients. 1

Integrative genomics reclassifies B‑ALL and sharpens identification of very-high-risk subtypes

Reclassification of 1,015 B‑ALL cases using WHO‑HAEM5, ICC, and integrative genomics markedly reduced unclassifiable disease compared with WHO‑HAEM4R. Unclassifiable cases decreased from 41.9% to 15.9% with WHO‑HAEM5 and 11.9% with ICC when WTS, fusion, mutation, and cytogenetics were integrated. Genomic classification independently predicted survival after adjusting for age, MRD status, and transplant, confirming true prognostic value. HLF‑rearranged and MEF2D‑rearranged B‑ALL showed persistently poor outcomes across ages despite allogeneic transplantation. Expression profiling clarified cryptic subtypes, resolved concurrent lesions, and highlighted emerging high‑risk groups such as IDH1/2‑ and ZEB2‑mutated B‑ALL. 2

Uncertainty-aware pan-cancer AI achieves workload-efficient, near-perfect sensitivity for nodal metastasis

UPATHLN combined a pathology foundation model with a dedicated uncertainty module to assess 26,229 lymph nodes from 14 primary cancers. Internal validation showed an AUC of 0.986 for detecting lymph node metastases. The uncertainty mechanism flagged potentially false‑negative outputs for mandatory review and intercepted all missed metastases in development and test cohorts. This strategy achieved 100% conditional sensitivity, including nodes from seven previously unseen primary tumor origins. Simultaneously, review of negative lymph nodes fell by 73.2%, indicating substantial workload reduction without sacrificing safety. 3

Misalignment-resistant generative AI produces virtual stains visually equivalent to chemical staining

A generative AI framework generated virtual stains that remained accurate despite significant misalignment between source and target images. Cascaded registration decoupled stain synthesis from exact pixel-level alignment, enabling training on imperfectly paired data. Across five datasets, the approach outperformed state-of-the-art models, including a 23.8% image quality gain in highly misaligned samples. In blinded review, experienced pathologists correctly distinguished virtual from chemical stains only 52% of the time, indicating effective visual equivalence. This misalignment-tolerant strategy lowers data curation demands and supports scalable clinical deployment of virtual staining. 4