Computer Aided Solution for Automatic Segmenting and Measurements of Blood Leucocytes Using Static Microscope Images

Area of Science:

• Medical Imaging
• Computer Vision
• Hematology

Background:

• Blood leucocyte segmentation in medical images is challenging due to cell variability and overlapping.
• Manual analysis of blood smears is time-consuming, error-prone, and requires expertise.
• Existing automated methods struggle with image complexity, cell morphology variations, and noise.

Purpose of the Study:

• To develop and validate an automated strategy for segmenting and identifying blood leucocytes from static microscope images.
• To improve the accuracy and efficiency of leucocyte analysis compared to manual methods.

Main Methods:

• A three-stage computer vision approach combining image enhancement, Support Vector Machine (SVM) for segmentation, and filtering of non-regions of interest (ROI) using Local Binary Patterns (LBP) and texture features.
• Training an SVM classifier to identify ROI and employing histogram analysis to exclude non-ROI.
• Utilizing LBP features for final blood leucocyte type identification.

Main Results:

• The proposed automated method achieved a high identification accuracy of 95.3% for blood leucocytes.
• The system demonstrated 100% sensitivity and 91.66% specificity in comparison to manual segmentation by a gynaecologist.
• Evaluation on 100 microscope images confirmed the method's robustness and viability.

Conclusions:

• The developed automated segmentation and identification technique is a reliable and accurate alternative to manual analysis of blood leucocytes in microscope images.
• This approach offers significant potential for streamlining hematological diagnostics and research.
• Further validation across diverse datasets can enhance its clinical applicability.