AI and the World's Hard Problems · Medicine & drug discovery · Nature (computational biology) · 2026-10-22
Ontology-guided neuro-symbolic convNeXtV2–swin transformer for explainable cytological classification of papillary thyroid carcinoma
The researchers developed a hybrid model combining fine cellular and nuclear detail with broader spatial context to address difficulties in classifying papillary thyroid carcinoma from cytology. It achieved 96.19% accuracy on an independent test set and 91.4% on external testing. Attribution analysis and ontology-guided interpretation supported pathology-informed explanations, without providing independent biomarker detection or expert-equivalent pathological reasoning.