Innovative breast cancer classification through semantic segmentation of ultrasound images: Advancing diagnostic precision and clinical utility

Abstract

There is still a very important public health concern about breast cancer worldwide, which has led to the development of more sophisticated diagnostic techniques. The current study focused on the development of a novel classification approach to breast cancer by incorporating advanced semantic segmentation into ultrasound image analysis. The methodology described here has contributed to enhancing diagnostic accuracy in breast cancer by obtaining correct outlines of anatomical structures within ultrasound images. The proposed method integrates a hybrid architecture that combines DeepLabV3 and U-Net for semantic segmentation, embedded within a GAN framework. A fine-tuned ResNet50 model is employed for final classification based on segmented outputs. This approach enhances diagnostic precision by leveraging the synergy between segmentation and classification. As can be seen from our research results, a significant step was made beyond the conventional methods of classification by showing the strength of semantic segmentation in the refinement process of diagnostics. Our model reaches an accuracy rate of 99.04%, which is really a large step from the state of the art.

In general, the current study enhances the state-of-the-art in the diagnosis of breast cancer by proposing much better and more efficient methods to analyze ultrasound images. The increased accuracy and reliability of our method are of crucial importance to early detection and intervention in meeting the urgent needs of global health.