Predictive and probabilistic model for cancer detection using computer tomography images

Abstract

This paper presents one predictive and two probabilistic models for detecting cancer in human liver using computed tomography image. Two probabilistic models are built using Logistic Regression (LR) and Linear Discriminant Analysis (LDA). Multilayer Perceptron (MLP) is used to make a predictive model. The proposed method consists of three basic steps. Initially, fuzzy c-means (FCM) clustering algorithm is used to segment the lesions from the human liver. Among all the segmented lesions, some of them are marked as abnormal (malignant) and others are marked as normal (benign) by the radiologist. It has been observed experimentally that the marked normal and abnormal lesions are distinguishable by their textures. Gray Level Co-occurrence Matrices (GLCMs) are one of the earliest methods for texture analysis. Thirteen Haralick features are extracted from the GLCMs of abnormal and normal lesions, which are further employed to build two probabilistic models using LR, LDA and a predictive model using MLP to determine the probability that the patient has cancer in his liver or not. A comparative study has been made based on the prediction accuracies of these three models. Moreover, LR and LDA are used to identify some of the features out of those thirteen features which play a statistically significant role in decision making by these probabilistic models. On the other hand, MLP doesn’t have the ability to select such significant features. It is proved that logistic regression (96.67%) gives better accuracy as compared to LDA (95%) and MLP (94.4%).