Breast Cancer Identification using Convolutional Neural Network

Breast cancer remains a major cause of cancer-induced mortality, with approximately 2.6 million new invasive and in-situ cases among women, 2,550 among men, and 40,920 deaths reported annually. Clinical screening followed by histopathological examination constitutes the standard diagnostic workflow; however, automated identification of tumor subtypes from histopathological and mammographic images remains challenging due to high intra-class variability. Deep learning methods provide a more robust and cost-effective alternative to conventional classification approaches. In this study, fine-tuned pre-trained deep neural networks are employed, and an Inception V3 based model is retrained for breast cancer classification. The proposed approach is evaluated on 10,239 mammographic images in DICOM format obtained from the Cancer Imaging Archive. The framework achieves an average classification accuracy of 75% across three diagnostic categories: benign, malignant, and benign without callback. Transfer learning with a four-layer neural network configuration is used to adapt the Inception V3 model to the breast cancer domain.