A Deep Learning Based Approach to Breast Cancer Detection

Abstract

Breast cancer is the most frequent malignancy in women worldwide. Lack of understanding and late tumor diagnosis increase women's mortality. Early discovery, therapy, and periodic check-ups may reduce mortality. Breast tissue cells proliferate too quickly, causing cancer. Numerous researchers calculate breast cancer tumor accuracy and prediction using different machine learning models on different datasets. Researchers employ CNN and other deep learning algorithms. This study classifies benign and malignant cancers using inceptionv3 deep learning model and a convolutional neural network using convolution layers. The DDSM Mammography dataset comprises 11170 pictures, while the BreakHis dataset has 7909 images. This study trains the CNN model inception V3. A PCA-based logistic regression classifier outperformed other machine learning algorithms. This study uses transfer learning on a pre-trained proposed model, Inception V3, which has a testing accuracy of 96% on the DDSM dataset and on the WCBD dataset the accuracy of 96.24%. The third dataset, the Breast Cancer Histopathological Database (BreakHis), has 7909 microscopic pictures, and the CNN model had 98.8% accuracy, the best of all datasets in this study. Cross-validation of accuracy, precision, recall, and f1 score on deep learning approaches improves results.