An AI Score for a Blood Cancer That Often Leaves Pathologists Guessing

A Weill Cornell Medicine team has built an AI scoring tool for myelodysplastic neoplasms (MDS), a chronic, progressive blood cancer that mostly strikes older adults and forces patients through repeated biopsies to track its course. About one in three MDS patients progresses to acute myeloid leukemia, a more aggressive cancer, so the read on each slide has real consequences.

The new tool, MDS-MAPS (MDS-Microarchitectural Perturbation Score), scores a bone marrow sample against healthy tissue using 82 features tied to normal marrow and to genetic MDS subtypes, capturing the size and shape of hematopoietic (blood-forming) cells and how they are arranged in space. It runs on standard tissue stains and imaging that most hospital pathology labs already have, and can track a patient's score from diagnosis onward, the team reports in Leukemia.

"There are some clear-cut cases and a lot of gray area," said Sanjay Patel, clinical chief of hematopathology at Weill Cornell, framing the problem the tool targets. Co-author David Redmond, an assistant professor of computational biology, called it a step toward triaging patients for precision therapies.

The next concrete step is validation in larger patient cohorts, including other MDS subtypes, in collaboration with Pinkal Desai, MD's group at NewYork-Presbyterian/Weill Cornell Medical Center. Until that work lands, MDS-MAPS is a research tool, not a deployed diagnostic.