A novel diffusion-based deep learning model to suppress acoustic interference for real-time monitoring in ultrasound-guided HIFU surgery

In ultrasound-guided HIFU surgery, simultaneous transmission of sound waves for therapy and imaging faces challenges due to strong acoustic interference contaminating the guiding images. This paper introduces HIFU-Diff, a deep learning model based on diffusion to suppress HIFU interference by treating contaminated images as low-resolution and reconstructing them into high-resolution images. Ex-vivo and in-vivo experiments were conducted to validate its performance. HIFU-Diff outperformed notch filtering and the recent FUS-Net DL network, particularly under strong interference and complex in-vivo conditions, preserving acoustic characteristics and enabling quantitative imaging. The proposed method allows simultaneous HIFU therapy and real-time imaging, improving safety and efficacy in clinical applications. This is the first diffusion-based model for HIFU interference suppression.