A novel white blood cells segmentation algorithm based on adaptive neutrosophic similarity score

Area of Science:

• Medical Imaging
• Computational Pathology
• Digital Hematology

Background:

• Automated analysis of blood smear images is crucial for disease diagnosis.
• Previous WBC segmentation methods struggled with healthy and non-healthy cells separately and adaptive color variations.
• Existing algorithms often relied on external enhancement methods that could alter cell morphology.

Purpose of the Study:

• To develop a novel, adaptive segmentation algorithm for white blood cells (WBCs) in digital blood smear images.
• To segment both the nucleus and cytoplasm of WBCs accurately.
• To create a unified algorithm applicable to both healthy and non-healthy WBCs, including those in mixed-cell images.

Main Methods:

• Proposed a multi-scale similarity measure based on the neutrosophic domain for WBC segmentation.
• Employed an adaptive neutrosophic similarity score to handle variations across different color components and spaces.
• Developed distinct segmentation frameworks for WBC nucleus and cytoplasm, applied to diverse datasets without external morphological enhancement.

Main Results:

• Achieved high precision rates in segmentation performance, with measurements A1 = 96.5% and A2 = 97.2%.
• The proposed method demonstrated an average segmentation performance of 97.6% across various WBC types.
• Successfully segmented both healthy and non-healthy WBCs, including complex cases with mixed cell populations.

Conclusions:

• Introduced an adaptive neutrosophic sets similarity score method for robust WBC detection and component segmentation (nucleus and cytoplasm).
• The algorithm is suitable for fully-automated classification systems.
• This method shows particular promise for classifying non-healthy cells, such as leukemia cells.