Enhancing VIA screening for cervical cancer: a comprehensive system integrating image processing, risk factors and follow-up facilitation

Abstract

Visual Inspection of cervix with Acetic Acid (VIA) is an inexpensive and effective screening test which is being conducted in many under-developped and developing regions. In medical image processing applications, the performance of Computer Vision has been promising. This research aims to establish a systematic process for VIA (Visual Inspection with Acetic Acid) screening of the cervix by incorporating Computer Vision and Machine Learning techniques. The paper analyzes the performance of VGG16, ResNet-50, YOLOv9, YOLO-NAS(Medium) on cervix images with Acetic Acid (VIA), with VGG-16 achieving 96% accuracy, ResNet-50 achieving 95%, YOLOv9 achieving 93%, and YOLO-NAS achieving 91%. Furthermore, we use feature importance scores from the Random Forest model to extract key features associated with cervical cancer from demographic, behavioral, and clinical factors. Training an ensemble model on these features yields an accuracy of 94%. The goal is to analyze the cervical images during VIA inspection and predict its outcome using Computer Vision and integrate patient risk factors so that our proposed system identifies high-risk individuals, even among VIA-negative cases and improves overall screening accuracy, surpassing the existing VIA screening method.