Improved segmentation and classification of breast cancer using mammogram images with residual network based deep learning
Saruchi Saruchi · Computational Methods in Science and Technology · 2024
The use of mammogram images for breast cancer detection focusses on image acquisition, preprocessing, segmentation, and feature extraction. Digital mammograms are used to classify images into benign, malignant, or micro calcification stages, enabling further diagnosis and treatment. The BI-RADS score from the mammogram determines the classification of abnormalities. Micro calcifications, small white specks in dense breast tissue, are early indicators of cancer. The research study uses the K-means algorithm for segmentation and CNN for classification, comparing results with thermal and MRI images. The proposed work includes an unsupervised hard clustering method called K-means clustering for segmentation and three deep learning models: CNN architecture, Residual network 50, and Inception V3. In this research mammogram images, employing K-means clustering and Convolutional Neural Network (CNN) methods. The K-means algorithm, a hard clustering method, effectively partitions the images into normal and abnormal categories. Subsequently, these segmented images undergo classification using advanced CNN models, which are enhanced through transfer learning techniques. A significant finding of this research is the superior classification accuracy achieved using thermal images in breast cancer diagnosis, surpassing the results obtained with mammogram and MRI images. The dataset for this study comprised 1300 mammogram images, of which 300 were utilized for testing. The proposed methodology demonstrated impressive classification accuracies of 91.66% for benign cases, 89.33% for malignant cases, and 91.66% for micro- calcification