Automated Breast Cancer Detection Using Optimized CNN Models on Mammography Images: A Deep Learning Approach for Enhanced Diagnostic Accuracy

Abstract

Breast cancer is among the most common and increasing illnesses worldwide; this disease is primarily seen in women. Early identification is key to managing and controlling breast cancer. However, many previous deep learning models for breast cancer detection suffer from overfitting, poor generalization across diverse clinical cases, and reliance on limited public datasets. In this study, a deep learning model based on a custom Convolutional Neural Networks (CNNs) architecture is developed to automatically classify mammography images into normal or abnormal categories. The model was implemented using the Keras API with TensorFlow as the backend. To overcome existing limitations, key parameters such as learning rate and dropout were optimized to reduce overfitting and enhance classification performance. Various data augmentation techniques, including flipping, rotation, and contrast adjustments, were applied to improve generalization across different cases. The dataset, collected from the hospital, includes images from 430 patients across different age groups, ensuring clinical relevance. After training and evaluation, the model achieved a high accuracy of 98%, along with high sensitivity, specificity, and precision, confirming its reliability in distinguishing between normal and abnormal breast tissue. Compared to previous studies, this approach demonstrates competitive and, in many cases, improved classification performance. The promising outcomes indicate that the proposed CNN model has the potential to assist radiologists in identifying abnormal cases more accurately and could be a valuable tool in computer-aided diagnosis (CAD) systems for breast cancer screening.