基于多模态数据和医学知识融合驱动的睡眠分期方法研究

睡眠分期是医生用来对患者整夜睡眠状态、质量以及相关疾病诊断的基础。现有的自动睡眠分期方法多数是以数据驱动建立的决策模型，忽视了模型的临床可解释性、鲁棒性和泛化性。.本项目拟探索基于多模态数据和医学知识融合驱动的自动睡眠分期新方法。通过对基于深度学习目标检测关键技术的研究，探索和求解生理信号中多尺度、多类别睡眠事件的准确分类和精准定位问题，实现完备睡眠事件信息的获取。通过对数据和知识融合驱动机制的研究，探索睡眠事件信息和医学知识融合方法，实现自动睡眠分期决策模型，提升模型的可解释性和准确性。同时，考虑到患者的个体差异，探讨决策模型中睡眠事件的个性化语义转换机制，从而进一步提升模型的鲁棒性和泛化能力。通过结合数据驱动算法强大的学习能力以及知识驱动算法丰富的先验知识，本项目可以为临床睡眠数据分析提供切实有效、临床解释性强、准确率高的解决方法，也有望为个性化的睡眠分析和临床诊断提供新的思路。

Sleep staging is the fundamental step for physicians to evaluate the sleep states, sleep quality and to diagnose sleep disorders. Most of the existing automatic sleep staging methods are data-driven-based models, which ignore the clinical interpretability, system robustness, and model generalization ability..In this project, we aim to explore a novel sleep staging method based on the integration of knowledge-driven and data-driven approaches. Firstly, to offer comprehensive information on the sleep events, we delve into the deep learning method to surmount multi-scale and multi-class sleep events classification and localization challenges. Meanwhile, to improve model interpretability and accuracy, an automatic sleep staging decision model is proposed by integrating knowledge-driven and data-driven models to fully fuse the detected sleep events and prior knowledge. Furthermore, considering the individual variability, a personalized semantic conversion mechanism of sleep events is investigated to further improve the robustness and generalization ability of the model. The proposed model is expected to provide a practical, effective, highly clinically interpretable and accurate solution for analyzing clinical sleep data, and offer new ideas for personalized sleep analysis and clinical diagnosis.