Entropy Measurements for Leukocytes' Surrounding Informativeness Evaluation for Acute Lymphoblastic Leukemia...

Area of Science:

• Medical image analysis
• Computational pathology
• Machine learning in hematology

Background:

• Deep learning models show promise in leukemia classification from microscopic images.
• Lack of interpretability in deep learning hinders clinical adoption.
• Uncertainty exists regarding whether lymphocytes or their surroundings hold key diagnostic information.

Purpose of the Study:

• To investigate the diagnostic informativeness of lymphocyte surroundings for leukemia classification.
• To develop human-interpretable features for sickness probability assessment.
• To compare the utility of whole images versus lymphocyte-only regions.

Main Methods:

• Application of entropy measures and machine learning models.
• Analysis of both whole microscopic images and lymphocyte-surrounding regions.
• Utilized the Acute Lymphoblastic Leukemia Image Database forదల (ALL-IDB2).

Main Results:

• The hue distribution of lymphocyte-surrounding regions alone was sufficient for classification.
• Achieved 93.0% accuracy in discriminating between healthy and sick patients.
• Demonstrated the diagnostic value of leukocyte microenvironment features.

Conclusions:

• Leukocyte microenvironment features are highly informative for leukemia classification.
• Machine learning models can provide interpretable insights into diagnostic processes.
• This approach enhances the potential clinical utility of AI in hematopathology.