Feature selection and classification of leukocytes using random forest.

Area of Science:

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

Background:

• Automatic segmentation of leukocytes in tissue images generates large datasets with noise and artifacts.
• High-dimensional data from segmentation degrades classifier performance in distinguishing leukocytes from non-cellular elements.
• Effective feature selection is crucial for reducing computational load and enhancing classification accuracy.

Purpose of the Study:

• To introduce a novel feature selection method for improving leukocyte segmentation.
• To reduce computational complexity and enhance classifier performance by selecting prominent features.
• To accurately classify segmented objects into artifacts, mononuclear cells, and polymorphonuclear cells.

Main Methods:

• A Gini importance-based binary random forest method was developed for feature selection.
• The random forest classifier was employed for object classification.
• Feature reduction techniques were compared against the proposed method.

Main Results:

• The proposed method effectively identifies and eliminates irrelevant features from the segmented data.
• High classification accuracy was maintained, outperforming other feature reduction methods.
• The approach successfully discriminates between leukocytes and artifacts/noise.

Conclusions:

• The Gini importance-based random forest feature selection method significantly enhances leukocyte segmentation accuracy.
• This method offers an effective strategy for managing high-dimensional data in biomedical image analysis.
• The approach provides a robust solution for classifying cellular and non-cellular objects in tissue images.