Classification of Cervical Precursor Lesions via Local Histogram and Cell Morphometric Features

Area of Science:

• Pathology
• Computer-Aided Diagnosis
• Biomedical Image Analysis

Background:

• Cervical squamous intra-epithelial lesions (SIL) are precancerous conditions requiring accurate diagnosis for effective treatment.
• Current diagnostic methods rely on pathologists' assessment of cell distribution within cervical tissues.
• Understanding cell distribution patterns across tissue layers is crucial for identifying Cervical Intraepithelial Neoplasia (CIN) grades.

Purpose of the Study:

• To propose and evaluate novel classification schemes for cervical tissues utilizing local histogram features.
• To demonstrate the utility of histogram information for explainable tissue classification in cervical cancer screening.
• To compare the performance of the proposed methods against existing morphology-based and Convolutional Neural Network (CNN) approaches.

Main Methods:

• Development of two distinct classification schemes based on local histogram analysis.
• Implementation of a Kullback Leibler divergence-based classifier.
• Integration of morphometric features with histogram data for a combined classification approach.

Main Results:

• The proposed method achieved a classification accuracy of 78.69% on a public dataset.
• This performance surpasses traditional morphology-based methods (69.07%) and CNN patch-based algorithms (75.77%).
• The statistical features derived from local histograms proved robust and independent of lesion characteristics.

Conclusions:

• Local histogram analysis is a valuable and effective feature for classifying cervical tissues.
• The proposed explainable AI approach offers improved diagnostic accuracy for cervical lesions.
• These findings support the potential of histogram-based statistical features in real-world clinical applications for cervical cancer prevention.