COMPARING SEGMENTATION TECHNIQUES FOR SEGMENTING ENDOCERVIX IN COLPOSCOPY IMAGES
Loading...
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Malaysia Sarawak
Abstract
Description
Cervical cancer is a significant global health concern, and early detection through colposcopy plays a vital role in reducing mortality rates. Accurate segmentation of the endocervix in colposcopy images is essential for diagnosis and treatment. This study compares segmentation techniques to determine the most effective method for segmenting the endocervical region. A publicly available, medium-sized dataset of annotated colposcopy images from Roboflow was used. A standardised preprocessing pipeline was applied to resize images and generate binary segmentation masks. Four segmentation models: K-Means clustering, Random Forest, Support Vector Machine (SVM), and U-Net were implemented and evaluated using Intersection over Union (IoU), Dice Coefficient, and F1 Score. Among the models tested, U-Net achieved the highest performance, demonstrating superior ability to capture spatial and contextual features. In contrast, traditional machine learning models showed moderate results but were limited in handling complex anatomical structures. These findings highlight the advantage of deep learning, particularly the U-Net architecture, for accurate endocervix segmentation. This research supports the application of AI-based methods in cervical cancer screening by providing a comparative evaluation of segmentation approaches in medical imaging.