Multinomial Logistic Regression For Breast Thermogram Classification

The use of thermal imaging in breast cancer diagnosis is a state of art in many oncology diagnosis centers. This research is fueled by the growing popularity of Thermograms. Breast Cancer is the most common and lethal disease these days. In order to help radiologist examine different stages of Breast Cancer, Logistic Regression Classifier helps them to interpret images. The dataset of 400 images have been collected from HCG Hospital Bangalore. First images are de noised using median filters. Removal of noise will help in increasing the accuracy of classifier, hence detecting cancer at the early stage. Secondly, Histogram equalization and color segmentation is done to differentiate the pixels of same intensity and further determine the region of interest (ROI). Thirdly, Texture Analysis Using the Gray-Level Co-Occurrence Matrix is done where GLCM features like entropy, energy, auto correlation etc are extracted. At last these Texture Features serve as inputs to proposed classifier Multinomial Logistic Regression to detect the normal, early and late stages of breast cancer. The proposed Classifier is also compared with other classifiers like SVM, ANN and KNN for accuracy, precision, reliability and Compression ratio.