Automated Mammogram-Based Breast Cancer Detection with Deep Learning and Advanced Image Enhancement

Abstract

Breast cancer is a significant worldwide health issue, and early detection is key to enhancing survival. Although mammography is the accepted screening method, human interpretation is susceptible to errors, resulting in misdiagnosis. Convolutional neural networks (CNNs), in particular, have shown promise in deep learning for automating breast cancer detection, increasing accuracy, and reducing human variability. In this research, a deep learning model for automatically classifying breast cancer from mammograms is proposed and evaluated. The suggested model's performance on the CBIS-DDSM dataset is compared to transfer learning using pre-trained models like MobileNetV2, DenseNet121, and EfficientNetV2L in terms of classification accuracy and generalizability. Moreover, the study investigates how data augmentation and preprocessing affect the models. Accuracy, sensitivity, specificity and computational efficiency were used to measure performance. The proposed technique achieved the highest performance, with 98.21% accuracy, 99.04% Sensitivity and 97.33% Specificity in the 80%-10%-10% data split and 99.02% accuracy in the 85%-5%-10% split.  Affirming the effectiveness of deep learning in enhancing the accuracy, 99.24% Sensitivity and 98.80% Specificity of breast cancer detection. This study systematically contrasts CNN models for mammogram classification, optimizes preprocessing methods, and evaluates computational efficiency. It fills a research gap by balancing accuracy against computational tractability and proves proposed model higher diagnostic potential for AI-augmented mammography. The research verifies that deep learning models drastically enhance breast cancer detection based on mammograms, with proposed model being the most balanced between accuracy and efficiency.