Pathology

WHO Breast Tumours 6th edition tightens entities and biomarker language

The 6th WHO Breast Tumours edition updates HER2 reporting categories in light of DESTINY-Breast 04 and 06, including therapeutically relevant low-expression disease. The term “variant” is now reserved for molecular or genetic alterations, clarifying terminology for invasive tumor subtypes. Invasive lobular carcinoma with extracellular mucin is recognized as a new diagnostic entity with prognostic implications distinct from classic ILC. Mucinous carcinoma is restricted to mucin-secreting, grade 1–2 tumors with a favorable biomarker profile, narrowing use of this diagnosis. Malignant phyllodes tumor now requires only four of the original five adverse histologic criteria, and breast neuroendocrine tumor classification is revised. New sections address diagnosis on small samples, B coding, multidisciplinary review, and the growing role and challenges of digital pathology and AI. 1

Methylation profiling defines 44 hematolymphoid classes with high diagnostic concordance

Genome-wide methylation profiling of 1,156 hematolymphoid neoplasms defined 44 reproducible classes aligning with WHO 5th edition and ICC entities. Several classes also captured clinically and biologically meaningful subgroups, and allowed inference of characteristic copy number alterations. A machine-learning classifier achieved 97% concordance with the original diagnosis in high-confidence cases, despite only modest rates of high-confidence calls. In discrepant high-confidence cases, additional review often favored the methylation-based prediction, highlighting its value as an adjunctive classifier. Low tumor purity was a major contributor to low-confidence scores, underscoring pre-analytic constraints on routine deployment. 2

QuANTUM standardizes tumor cell fraction for HGSC HRD testing

In 70 high-grade serous tubo-ovarian carcinomas, pathologist tumor cell fraction estimates showed only low-to-substantial agreement, with weighted κ values from 0.28 to 0.63. The QuANTUM computational pipeline showed substantial agreement with a manually derived ground truth and with the AmoyDx algorithm, with κ 0.73 and 0.63 respectively. QuANTUM, ground truth, and AmoyDx did not differ significantly in classifying samples around the 30% tumor cellularity cutoff used for HRD testing. These results support QuANTUM as a reliable tool to standardize tumor cell fraction estimation in HGSC molecular workflows. 3

Hybrid-supervised AI improves frozen diagnosis and surgical planning in lung adenocarcinoma

The HSFLA framework was developed to assist intraoperative frozen diagnosis of lung adenocarcinoma and optimize resection strategies. Using 1,161 whole-slide images from two centers and three subtypes, HSFLA achieved 95.6% accuracy, outperforming manual review and weakly supervised deep learning. The model automatically annotated invasive regions, registered serial sections, and reconstructed 3D invasive tumor volume with 86.6% concordance to manual annotations. HSFLA annotations improved pathologist diagnostic accuracy by 22.9% in a small reader study and showed concordance with spatial transcriptomics patterns. In prospective use, human–machine diagnosis changed surgical recommendations in 5 of 70 patients, demonstrating real-world clinical impact. 4