Abstract
High-Intensity Focused Ultrasound (HIFU) is a well-established non-invasive tumor ablation technique. However, accurately and real-time monitoring the temperature distribution within the HIFU focal region remains a significant challenge. Although Magnetic Resonance Imaging (MRI) temperature measurement is considered the gold standard for controlling focal temperatures during HIFU treatment, it suffers from limitations in spatiotemporal resolution. In this context, the ultrasound technique, with its advantages of real-time and non-invasiveness, has emerged as a promising supplementary tool for HIFU temperature field monitoring. A potential solution to address the spatiotemporal resolution issue in HIFU temperature monitoring is the combination of MRI and ultrasound technologies. Given the significant potential of Artificial Intelligence (AI) in biomedical applications, a novel ultrasound-based temperature measurement algorithm was proposed, utilizing an end-to-end deep neural network architecture called the "Multimodal Attention Multiscale Teacher-Student Knowledge Distillation Model" (MMAM-TSD). Experimental results demonstrate that MMAM-TSD not only clarifies the interaction between HIFU and complex heterogeneous biological media, but also achieves real-time, precise temperature measurement of the HIFU focal zone, marking a breakthrough in the field of temperature monitoring for HIFU treatments.