Classification of acute myeloid leukemia M1 and M2 subtypes using machine learning

Area of Science:

• Hematology
• Medical Imaging
• Machine Learning

Background:

• Acute myeloid leukemia (AML) classification traditionally relies on manual analysis of bone marrow or peripheral blood smear images.
• Accurate subtyping is crucial for effective treatment strategies.

Purpose of the Study:

• To develop an automated machine learning model for classifying AML-M1 and AML-M2 subtypes.
• To improve the efficiency and accuracy of AML subtyping.

Main Methods:

• Utilized bone marrow smear images from the Cancer Imaging Archive (TCIA).
• Employed random forest and broad learning system (BLS) methods for model development.
• Extracted and selected morphological, radiomics, and clinical features.

Main Results:

• A random forest model incorporating 9 selected variables (2 morphological, 6 radiomics, 1 clinical) achieved high performance.
• The model demonstrated an average accuracy of 0.998, AUC of 0.998, F1-score of 0.998, recall of 0.996, and precision of 1.

Conclusions:

• The random forest model shows excellent performance for classifying AML-M1 and AML-M2 subtypes.
• This automated approach may serve as a valuable tool for clinicians in AML diagnosis.