Leukocyte recognition in human fecal samples using texture features.

Area of Science:

• Biomedical imaging
• Computational pathology
• Medical diagnostics

Background:

• Human fecal samples present unique challenges for leukocyte analysis due to abundant food debris and cellular degradation.
• Standard segmentation techniques struggle with impurities and ambiguous cell morphology in fecal matter.
• Accurate leukocyte identification in fecal samples is crucial for diagnosing gastrointestinal conditions.

Purpose of the Study:

• To develop an effective method for segmenting leukocytes in human fecal samples.
• To identify robust texture features and classifiers for accurate leukocyte recognition.
• To improve the diagnostic potential of fecal sample analysis.

Main Methods:

• A novel threshold segmentation method utilizing inscribed and circumscribed circles was employed to remove impurities.
• Five texture features (LBP-uniform, Gabor, HOG, GLCM, Haar) were extracted from segmented cells.
• Leukocyte classification was performed using Support Vector Machine (SVM), Artificial Neural Network, AdaBoost, and Random Forest classifiers.

Main Results:

• The proposed segmentation method achieved a high accuracy of 97.6% in removing impurities.
• Histogram of Oriented Gradients (HOG) features combined with an SVM classifier yielded the best identification performance.
• The HOG+SVM approach demonstrated a precision of 88.46% and a recall of 88.72% for leukocyte identification.

Conclusions:

• The developed segmentation and identification pipeline significantly enhances the accuracy of leukocyte analysis in human fecal samples.
• The combination of HOG features and SVM classification offers a promising approach for automated fecal leukocyte detection.
• This methodology has the potential to improve non-invasive diagnostics for various gastrointestinal diseases.