Breast Cancer Detection from Mammogram with Deep Features Fusion
Abin Antony, Khaled Tawfiq Al-Assaf, Thangamuthu Kumaran Vaiyapuri, K. Vijayakumar, B. S. Harish, Alok Jain, Md. Tabil Ahammed
Abstract
Breast cancer (BC) is a severe disease in women and causes a large diagnostic burden globally. Early diagnosis and treatment of the BC is necessary for suitable treatment. Hospital level detection of the BC is generally performed using the medical imaging approaches and digital mammogram is one of the common procedures in early screening of the BC. This work plans to propose deep-learning (DL) based approach to detect the benign/malignant BC from the mammogram. The different phases of the proposed scheme include; (i) image collection and resizing, (ii) feature extraction with DL-model and best model selection using SoftMax classification, (iii) implementing 50% dropout-based feature reduction and serial features integration to get the fused-feature-vector (FFV), and (iv) classification and 3-fold cross-validation to confirm the performance. In this work, VGG-variants, and DenseNet-varisnts are considered for the evaluation and the achieved result of this study confirms accuracy of >96%.