Deep Learning Approach for Breast Ultrasound Image Segmentation

Recent advances in computer-aided diagnosis (CAD) technology have brought about more possibilities to segment and classify breast tumors. In the past decades, breast related diseases have significantly grown among women and have become a leading cause of death worldwide. An effective way to diminish breast cancer is to offer a proper diagnosis in the early stages of the disease by using ultrasound images. Our work proposes a modified U-Net architecture equipped with pre-trained inception residual blocks as an encoder for breast ultrasound (BUS) image segmentation. To enhance the performance, we increased the depth of the network by adapting the inception blocks. Our proposed solution consists of a preprocessing stage, feature extraction based on inception layers and a basic U-Net decoder. We utilized two public datasets named BUSI and UDIAT. This model predicts a mask for regions of interest (ROI) in BUS images by utilizing residual connections to ensure a minimum error rate and to preserve dimensionality. Our results show improved performance over the existing U-Net architecture, as well as the more recent deep adversarial learning and Selective K-U-Net models.