Application of Gray-Level Co-Occurrence Matrix (GLCM) Feature Selection Method in Precancerous Cervical Lesion Prediction of Women Living with HIV in Nigeria

Women living with HIV face a heightened risk of developing cervical cancer, particularly in low-resource settings such as Nigeria.Early detection of precancerous cervical lesions is critical to improving outcomes, yet conventional screening methods like VIA/VILI are highly dependent on clinician expertise and subject to variability.These limitations often result in inconsistent diagnoses and missed lesions, especially in settings with limited specialist availability.This study proposes an automated lesion detection system using Gray-Level Co-Occurrence Matrix (GLCM) feature extraction combined with machine learning classifiers to improve diagnostic accuracy.A dataset of 2,516 VIA/VILI cervical images from HIV-positive women was pre-processed using non-local means denoising and grayscale conversion, and key texture features including contrast, energy, entropy, and homogeneity were extracted.Four classifiers, such as Random Forest, Decision Tree, Gradient Boosting, and Logistic Regression, were trained and evaluated using an 80:20 train-test split, with performance assessed through accuracy, AUC, precision, recall, and F1-score.The Random Forest model exhibited superior predictive performance, achieving 97% accuracy and an AUC of 0.99, surpassing existing clinical and image-based approaches.The Decision Tree demonstrated strong performance, while Gradient Boosting showed moderate effectiveness, and Logistic Regression performed least effectively.Comparative analysis with prior studies confirms the superiority of ensemble-based, featureengineered models for automated cervical lesion detection.The findings highlight the potential of integrating such models into mobile screening platforms like AVIVA to enhance real-time diagnosis, reduce missed cases, and support cervical cancer prevention in low-resource settings.Future work will focus on external validation and scalability to ensure robustness and broad clinical applicability.