Digital pathological classification of peripheral neuroblastic tumors across multi-continent cohorts

NB-PathAssist enables automated INPC-based classification of 2,030 peripheral neuroblastic tumors from multi-continent whole-slide images, achieving balanced accuracies up to 0.96 for prognostic groups.

• Why it matters: INPC stratification depends on age-linked histology and MKI, but manual assessment is variable and limits scalable prognostic prediction across diverse clinical settings.
• What they did: The team built NB-PathAssist using foundation-model feature extraction plus SummaryMixing with archetype-guided tumor-level classification, and trained MKI automation from 14,049 manually annotated nuclei.
• The result: Across cohorts spanning the United States, Canada, Australia, New Zealand, Germany, and Japan, the system reproduced reference INPC features and produced prognostic classifications concordant with annotations, with balanced accuracies of 0.84, 0.86, 0.89, and 0.96.

AI summary of the abstract · may contain errors · not medical advice · read the paper.