CCDSNet: A Neural Network-Based Ultrasonic Thermometry Model for Microwave Thermotherapy

Microwave thermotherapy, as an emerging cancer treatment, precise temperature control is the key to its success. In this study, a novel ultrasound image classification network model, CCDSNet (ConvMixer-CoordAtt-DuaISimAM Network), is proposed for the temperature prediction task. The model combines the ConvMixer architecture, the CoordAtt and SimAM attention mechanisms, and the C1DeepSupdeep-supervision module to accurately extract and classify temperature features. Experimental results show that CCDSNet achieves an accuracy of 95.18% and 91.19% in the 1°C and 0.5°C interval classification tasks, respectively, which is significantly better than other common models. The ablation experiments also verified the synergistic effect of multiple modules of the CCDSNet model, which not only improves the accuracy of temperature classification, but also enhances the robustness and generalization ability of the model, providing a novel solution for ultrasound image thermometry with important clinical application value.