Adaptive diagnostic reasoning framework for pathology with multimodal large language models

Why this matters for cancer diagnosis

When a pathologist looks at a tissue sample under the microscope, they are making life‑shaping decisions: is there cancer, what type is it, and how aggressive might it be? Computer programs already help scan thousands of such images, but most act like mysterious black boxes, offering a yes‑or‑no answer without clearly showing their work. This study introduces a new way to pair image‑understanding AI with language‑based reasoning so that the computer not only says what it thinks, but also explains why, in terms that line up with how real pathologists reason.

Turning black boxes into explainers

Most existing systems for reading pathology slides focus on raw accuracy. They learn patterns from millions of pixels and then output a label such as \

Citation: Hong, Y., Kao, KC., Edwards, L. et al. Adaptive diagnostic reasoning framework for pathology with multimodal large language models. Commun Med 6, 236 (2026). https://doi.org/10.1038/s43856-026-01491-z

Keywords: computational pathology, explainable AI, cancer diagnosis, multimodal language models, medical imaging