Enhancing Breast Mass Cancer Detection Through Hybrid ViT-Based Image Segmentation Model

Breast cancer, primarily affecting women, is characterized by uncontrolled cell growth in breast tissue, leading to the formation of tumors. Although its exact causes remain elusive, factors such as age, genetics, and hormonal influences increase susceptibility. Early diagnosis plays a pivotal role in reducing mortality rates, emphasizing the importance of intelligent screening, diagnosis, and treatment methods. In recent years, artificial intelligence, particularly deep learning, has gained prominence in various industries, including medicine, due to its remarkable success in solving complex problems. This study focuses on leveraging the power of deep learning to accurately segment abnormal breast regions, a critical step in breast cancer diagnosis. We explore and compare the performance of two distinct deep learning architectures: A-UNet and Trans-UNet. These architectures have demonstrated impressive potential in improving breast cancer detection. A-UNet achieved a Dice coefficient of 90.21% and 82.11% of IoU. Notably, Trans-UNet outperformed the A-UNet with a Dice coefficient of 94.13% and an IoU of 88.92%. These results showcase the promising capabilities of deep learning models in enhancing breast cancer detection, underscoring their potential to contribute significantly to early diagnosis and improved patient outcomes in the battle against this challenging disease.