Feature Analysis and Automatic Identification of Leukemic Lineage Blast Cells and Reactive Lymphoid Cells from...

Area of Science:

• Hematology
• Computational Biology
• Medical Image Analysis

Background:

• Automated peripheral blood analyzers often misclassify blast cells as lymphocytes.
• Current systems lack the ability to differentiate between myeloid and lymphoid blast cell lineages.
• Accurate blast cell identification is crucial for diagnosing and classifying hematologic malignancies.

Purpose of the Study:

• To develop an automated method for discriminating reactive lymphoid cells (RLC), lymphoid blast cells, and myeloid blast cells.
• To identify distinct morphologic patterns of these cell types using feature analysis.
• To improve the accuracy of automated blood cell analysis in identifying blast cells.

Main Methods:

• Utilized a training set of 696 blood cell images from patients with acute leukemia and infections.
• Employed support vector machines with various feature selection techniques for classification.
• Validated the selected features on a separate set of 220 images from new patients.

Main Results:

• Achieved 90.1% discrimination accuracy in the training set using 60 selected features.
• Nucleus-cytoplasm ratio and cytoplasm color-texture features were key discriminators.
• Validation stage yielded overall classification accuracy of 80%, with true-positive rates of 85% for RLC, 82% for myeloid blast cells, and 74% for lymphoid blast cells.

Conclusions:

• The developed methodology effectively recognizes reactive lymphocytes, particularly differentiating them from lymphoblasts.
• This automated approach shows promise for improving the accuracy of blast cell identification in clinical settings.
• Further refinement could enhance the system's utility in diagnosing hematologic disorders.