Gene mutations in acute myeloid leukemia (AML) cells can guide treatment options, and machine learning can rapidly guess the existence of gene mutations based on images of leukemia cells.
Published in npj Precision Oncology, the study explores a new way to predict gene mutations in acute myeloid leukemia (AML) using artificial intelligence (AI). AML is a type of blood cancer that affects white blood cells and causes high mortality, and identifying specific gene mutations is vital for diagnosis and treatment of AML. Traditionally, doctors use genetic tests to find these mutations, but these tests take time and could be expensive.
To solve this problem, a research team created an AI model that can identify gene mutations by analyzing images of bone marrow samples under a microscope. These images, called whole-slide images (WSIs), contain a lot of information, but it's difficult for doctors to find patterns that connect to gene mutations manually. The AI model, however, may detect these patterns.
The study used an AI strategy called multiple instance learning (MIL), which allows the AI to make predictions even when only part of the cell images contain useful information. To improve accuracy, the researchers combined several AI models into an "ensemble," meaning they used multiple models together to get better results.
The AI was trained and tested using real medical images, and the results showed that it can predict mutations in genes like NPM1 and FLT3-ITD with promising accuracy. This means that in the future, AI may help doctors quickly identify important genetic details about a patient's cancer without needing expensive genetic tests.
"Gene mutation predictions can be made as soon as bone marrow images become available, enabling doctors to promptly prepare potential treatments or anticipate treatment outcomes immediately after diagnosis," says Prof. Chien-Yu Chen.
More information: Bo-Han Wei et al, Annotation-free deep learning for predicting gene mutations from whole slide images of acute myeloid leukemia, npj Precision Oncology (2025). DOI: 10.1038/s41698-025-00804-0
Journal information: npj Precision Oncology