Thermal imaging based breast cancer analysis using BEMD and uniform RLBP

Abstract

Breast cancer is the foremost cause of morbidity and mortality among womenfolk. India has 17% of world's population suffering from breast cancer. World Health Organization's International agency for Research on Cancer (IARC) estimates that more than 4,00,000 women die every year due to breast cancer. Thus early identification of breast cancer plays a vital role in reducing the mortality rate. Medical infrared thermography is gaining popularity as it can detect abnormality at an early stage when compared to mammogram. In this work an attempt has been made to distinguish healthy and pathological subjects using Bidirectional Empirical Mode Decomposition (BEMD) and uniform Rotated Local Binary Pattern (RLBP). Frontal thermogram images are pre-processed using anisotropic diffusion filtering which preserves sharp edges, fine details and effectively removes noise. Level set segmentation technique is applied on pre-processed image to obtain the required Region Of Interest (ROI). The delineated ROI's are subjected to BEMD where the image is decomposed into Intrinsic Mode Functions (IMFs) and a residue. IMF1 is considered for further processing as it contains maximum edge information about the image. Texture analysis is performed on IMF1 using uniform RLBP technique to obtain feature vector comprising of 59 unique elements. The dimensionality of features is reduced to 20 using Kernel Principal Component Analysis (KPCA). The subjects are classified using Least Square Support Vector Machine (LSSVM) classifier with different kernel functions and the classifier results are evaluated based on classifier performance measures. Analysis is performed on 43 healthy and 24 cancerous subjects chosen from DMR (Database of Mastology Research) database. The experimental results obtained by LSSVM classifier using Radial Basis Function (RBF) kernel with σ2=3D0.5 and γ=3D1 outstands by achieving 89% accuracy. Hence the proposed system effectively distinguishes healthy and pathological subjects.