HGMMEM Algorithm with Optimized DCNN Classification for Breast Cancer Image Screening and Detection

Abstract

AbstractMammography is a safe diagnostic and painless procedure that uses low-dose x-rays for the early detection of breast cancer. Quality of mammogram images can be enhanced using digital image processing tools so as to assist physicians. A lot of researchers had worked on enhancing the quality and classification of mammogram images. Choosing appropriate prediction algorithm for the classification of mammogram images and proper decision support systems remains a major task in the research field. Here, in this research article of experimental analysis work, mammography images are taken from the both Public Digital Database of Screening Mammography (DDSM), and in-house clinical datasets from Metro scans and laboratories. The proposed work was carried out in four phases in which the first phase was the selection of suitable algorithm for denoising and contrast enhancement of the mammogram image by using Trilateral Filter with Histogram Equalization (TFHE). Second stage involved segmenting the denoise image and applying the Hierarchical Gaussian Mixture Model and Expectation-Maximization technique to detect the breast tumor accurately. Third stage entails extracting the GLCM features from the segmented ROI image. Finally, benign and malignant breast cancer images are classified using four different classifiers. Using TCKNN classifier, accuracy of 72.1% is obtained, DCNN classifier an accuracy of 89.4% is obtained, IMPA with TCKNN classifier an accuracy of 80.1% is obtained and IMPA with DCNN classifier an accuracy of 97.6% is obtained.