Machine Learning Techniques for Cervigram Image Analysis - CORE

Abstract

Machine learning is a popular technology widely used to solve a lot of problems in various areas in recent decades. In this work, we applied machine learning techniques to the problems of medical image analysis, especially cervigram image analysis. Combined with techniques developed in computer vision, we represent cervigram image data in the form of a combination of texture feature vector and color feature vector. We treat the task of detecting Cervical Intraepithelial Neoplasia (CIN) level as a classification problem in the view of machine learning and apply several popular machine learning classifiers to predict the categories. Furthermore, under receiver operating characteristic (ROC) curve as our performance measure, we do a comprehensive comparison among seven machine learning classification algorithms to see which ones might be suitable models for this kind of problems. From our experiments, we conjecture that the machine learning techniques can be a useful tool and ensemble-tree based models like Random Forest, Gradient Boosting Decision Tree and Adaboost outperform other algorithms for this task

Topics: Cervigram, Machine Learning, Medical Image Analysis, ROC analysis, Supervised Learning, Computer Sciences, Physical Sciences and Mathematics