CMJS-Net: Joint Segmentation and Classification via Cross-Modal Learning for Breast Cancer Diagnosis in CEUS

School of Biological Science and Medical Engineering, Southeast University, Nanjing, China

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

As an essential complement to B-mode ultrasound, contrast-enhanced ultrasound (CEUS) can effectively reduce the false-positive rate in breast cancer diagnosis by providing valuable dynamic perfusion information. However, the high spatiotemporal complexity in CEUS poses challenges for reliable computer- aided diagnosis. This paper proposes CMJS-Net, a dual-modality spatiotemporal network that jointly performs breast lesion segmentation and malignancy classification by integrating CEUS and B-mode ultrasound data. CMJS-Net employs an attention-based temporal aggregation strategy to model perfusion dynamics across time. Experiments conducted on a dual-modality breast tumor dataset demonstrate that the proposed method achieves a Dice similarity coefficient of 0.7813 for segmentation and an accuracy of 88.59% for classification, outperforming competing methods. These results indicate that CMJS-Net effectively exploits spatiotemporal perfusion cues for multi-task breast lesion analysis.