Deep learning-driven prediction in healthcare systems: Applying advanced CNNs for enhanced breast cancer detection

Abstract

The mortality risk associated with breast cancer is experiencing an exponential rise, underscoring the critical importance of early detection. It is the primary cause of mortality among women under 50 and ranks as the second deadliest disease globally. Timely identification is crucial, as heightened public awareness and accurate diagnosis can significantly reduce mortality rates. Patients with a positive prognosis and timely diagnosis have a far greater chance of full recovery. A comprehensive study was conducted to develop a robust breast cancer detection system using Convolutional Neural Networks (CNNs). This study details the processes of data collection, preprocessing, model building, and performance evaluation. The Mini-DDSM dataset was utilized, which includes 1952 scanned film mammograms from a diverse population. Data preprocessing involved normalization, denoising, illumination correction, and augmentation techniques to enhance data quality and diversity. During the model-building stage, several CNN architectures were explored, including Basic CNN, FT-VGG19, FT-ResNet152, and FT-ResNet50. The FT-ResNet50 model, fine-tuned with transfer learning, emerged as the top performer, achieving an accuracy of 97.54%. The integrated system leverages the strengths of each model to deliver accurate and reliable results, significantly advancing early detection and treatment methods for breast cancer. The comparative analysis demonstrated that the developed models outperformed existing state-of-the-art models. By leveraging the capabilities of deep learning and meticulous design, the objective is to significantly advance early detection and treatment methods for breast cancer, leading to better patient outcomes and ultimately, saving lives.