Research2026-10-29
Researchers built a deep learning model to predict IDH1 mutation status in acute myeloid leukemia. It uses bone marrow smear images from 436 patients across four centers in Germany and the US. The model achieved a mean ROC-AUC of 0.81. Predictions for IDH1 were linked to immature leukemia phenotypes. IDH1 mutations affect about 8% of AML patients. Targeted therapies exist for IDH1 mutations in AML. The study provides proof-of-concept for image-based IDH1 prediction.
What it unlocks
Predicting IDH1 mutation status in AML from bone marrow smear images without genetic sequencing.
- 436 AML patients across four centers
- mean single pooled ROC-AUC of 0.81
- 8% of AML patients have IDH1 mutations
What you need to act on it
- bone marrow smear whole-slide images
- attention-based multiple-instance learning model
Sources