Enhanced Breast Tumor Area Segmentation with Berkeley Wavelet Transformations

Breast cancer, a prevalent cancer type among women, underscores the significance of early detection and treatment to improve survival rates. Timely diagnosis is pivotal for enhancing treatment outcomes and reducing mortality rates. This research presents an approach to accurately identify tumor-affected areas in a limited set of BUS images. The proposed work utilizes a learning model based on an architecture named U-Net, coupled with the Berkeley Wavelet transform, for precise segmentation of affected regions. The developed model demonstrated promising outcomes, achieving a dice similarity coefficient of approximately 0.95 on BUS images and an accuracy of 93%. Notably, this accomplishment is remarkable given the successful segmentation of affected regions with a constrained dataset.