Accurate automatic detection of acute lymphatic leukemia using a refined simple classification.

Area of Science:

• Medical image analysis
• Hematology
• Computational biology

Background:

• Accurate subtyping of acute lymphatic leukemia (ALL) is crucial for effective treatment.
• Traditional methods for ALL subtype classification can be time-consuming and require specialized expertise.
• Automated classification techniques can potentially improve diagnostic efficiency and accuracy.

Purpose of the Study:

• To develop and validate an improved automated classification technique for acute lymphatic leukemia (ALL) subtypes.
• To identify the most significant features for accurate ALL subtype classification from peripheral blood smear images.
• To assess the diagnostic accuracy of a reduced feature set using robust machine learning classifiers.

Main Methods:

• Adaptive image segmentation of peripheral blood smear images to extract 10 geometric features from white blood cells (WBC), nucleus, and cytoplasm.
• Comprehensive feature importance analysis using permutation studies with K-nearest neighbor (KNN), support vector machine (SVM), and artificial neural network (ANN) classifiers.
• Evaluation of classification performance using receiver operating characteristic (ROC) curves and F1-score measures.

Main Results:

• A feature map was constructed, identifying the minimal set of features yielding the highest diagnostic accuracy.
• Vacuoles in the cytoplasm and the regularity of the nucleus membrane were identified as the most effective features for ALL subtype classification.
• The automated classification based on these two features demonstrated high accuracy, confirmed by ROC curve and F1-score analysis.

Conclusions:

• The proposed automated classification technique significantly improves the accuracy of ALL subtype identification.
• Utilizing only cytoplasm vacuoles and nucleus membrane regularity features streamlines the classification process.
• This method offers a highly accurate, efficient, and time-saving approach for diagnosing ALL subtypes.