Detection of Breast Cancer in Mammography Using Pretrained Convolutional Neural Networks with Fine-Tuning

Breast cancer is a major health concern for women, especially in Latin America where the incidence and mortality rates are high. Mammography is an essential diagnostic tool in detecting breast cancer, but interpreting mammogram images can be challenging due to their complex nature. To assist radiologists in identifying abnormalities in mammogram images, deep learning algorithms, specifically deep convolutional neural networks (DCNNs), are being employed. This chapter explores the effectiveness of several pretrained DCNN models, such as ResNet-50, ResNet152, VGG19, and EfficientB7, in classifying mammogram images. To ensure reliable results, the Mini-MIAS and CBIS-DDSM datasets, consisting of 334 and 2620 scanned film mammography images, respectively, were selected for this study. The images were categorized into binary classification and multiclassification groups based on the severity of the lesion. For both datasets, the same preprocessing approach was used to enhance image quality. This involved normalizing the images and applying contrast limited adaptive histogram equalization (CLAHE). The efficacy of the preprocessing techniques was evaluated by comparing the performance of the models on the entire dataset and just the normalized images. Four different stages were tested using images from both datasets, and the performance of each model was evaluated using five metrics, namely, accuracy, precision, recall, F1-score, and area under the ROC curve (AUC).