Examining the feasibility of predicting drug resistance of lung tuberculosis using image data

Abstract

This work is dedicated to the problem of early diagnosis of tuberculosis drug resistance using X-ray and CT images of tuberculosis patients. Image features were extracted using extended co-occurrence matrix approach followed by Prin- cipal Component Analysis method. Classification was done with help of recent classifiers such as SVM, Naive Bayesian, Logistic Regression and Linear Discrim- inant Analysis. The maximum achieved accuracy of drug resistance prediction was 75% when using SVM classifier. Results of the present study suggest that the approach may potentially be employed for early predictions of drug resistance.