Automatic Detection of Breast Cancer through Mammogram Images

Breast Cancer (BC) is a primary cause of cancer death among women in which breast cells expand out of control. The chances of survival increase by urging patients to seek treatment as soon as possible. Several researches have been proposed to detect breast cancer. However, they possess a few drawbacks and are inefficient in detecting the disease accurately. In order to save the patient's life, a transfer learning (TL) based model is proposed for the diagnoses and classification of the sus-picious breast area. In the presented model, learning parameters from pre-trained models VGG-19, VGG-16, and Inception-V3 networks are fine-tuned to enhance the performance of malignant lesions classification. The key objectives of the research are to employ segmentation to automatically locate the impacted breast tumor region, reduce training time, and enhance classification accuracy. The Mammographic Image Analysis Society (MIAS) dataset extracts breast tumor features in the proposed model. Three evaluation metrics are used to analyze the performance of the proposed model: accuracy, sensitivity, and specificity. The trials revealed that transferring parameters from the VGG-16 model is more powerful than VGG-19 and Inception V3 for BC classification, with overall specificity, accuracy, and sensitivity of 97.12%, 97.80%, and 96.43%, respectively.