As a global health challenge, the intervention process of diabetes has fallen into the dilemma of "triple contradiction" in clinical practice. Research has shown that traditional improvement methods are difficult to achieve long-term compliance (Lee et al. 2019). In recent years, studies have attempted to improve patients’ dietary behavior through multisensory integration. However, existing approaches lack the ability for dynamic closed-loop regulation (Zhang et al. 2023). First, through questionnaires and interviews, we investigated the reaction of diabetes patients to food with different sugar levels. Secondly, we used this behavior pattern as the result of neural feedback for dynamic detection. In order to better study and improve the sugar management of diabetes patients, based on the theory of multimodal sensory remapping, this study constructed a system that combines the biological perception layer, intelligent decision-making layer and multimodal executive layer. The system includes hardware and APP interface design. Recruit 30 patients of different ages for a two-week test. According to the System Usability Scale (SUS) evaluation, the average score was 82.4 (SD = 7.6), and it can effectively reduce sugar intake. The research verified the regulatory effect of the plasticity of the system, and provided a new paradigm of human-computer collaborative intervention for the treatment improvement of diabetes patients. Long term follow-up studies will be conducted to verify the sustained effects.