癫痫意识障碍的多模态脑影像模式识别关键方法研究

This project is mainly aimed at the clinical scientific problems of brain imaging characterization of consciousness impairment in epilepsy. It is proposed to develop key methods for multimodal brain imaging pattern recognition using information technology, to investigate the functional and structural changes of the core brain regions related to loss of consciousness in epilepsy, and to further explore the dynamic network evolution mechanism of the cortical generalization leading to impaired consciousness in patients. Based on this, a pattern recognition prediction model is constructed to realize data-driven classification of focal epilepsy subtypes and risk prediction of consciousness impairment. The main research contents including the following aspects: .(1) Propose large-scale rich club organization analysis method based on frequency dependent brain spontaneous oscillation mode and sparse optimization, to detect the similarities and differences of brain characterizations of temporal lobe epilepsy with consciousness impairment. We further calibrate the rich club regions of the consciousness network and excavate the brain imaging biomarkers specific to patients; .(2) Develop dynamic directed brain network theory methods to describe the possible propagation pathway of the consciousness network, and to further reveal the network mechanism of epileptic origin, transmission and dynamic evolution of consciousness-impairing seizures, which ultimately affects the extensive cortex; .(3) Construct pattern recognition prediction model based on long short term memory network, to achieve the identification of focal epilepsy subtypes and provide neuromarkers for clinical diagnosis, with an expected classification accuracy of 85%. Aiming at providing an objective brain imaging basis for the classification of focal epilepsy subtypes and the risk assessment of consciousness impairment in epilepsy.

本项目主要针对癫痫意识障碍脑影像学表征的科学问题，发展利用信息技术识别癫痫影像学特征的关键方法，研究癫痫患者与意识相关核心脑区的功能和结构改变，挖掘其皮层泛化导致意识受损的网络演化机制，并以此构建癫痫意识障碍的模式预测模型, 实现基于数据驱动的局灶性癫痫亚型分类和意识障碍的风险预测。主要研究内容包括：（1）提出基于分频的脑自发振荡模式和稀疏优化大尺度关键脑环路分析方法，探测伴有/无意识障碍的颞叶癫痫脑功能结构表征异同，标定意识网络的核心脑区，挖掘癫痫意识障碍特异性的脑影像生物学标记；（2）发展动态有向脑网络理论方法，刻画意识网络可能的传播路径，揭示伴有意识损伤癫痫痫性起源、传播及动态演化并最终影响广泛皮层的网络机制；（3）建立癫痫意识障碍患者具有多层特征学习能力的长短时记忆网络预测模型，以期分类准确率达85%以上，为完善局灶性癫痫亚型分类和癫痫伴意识损伤的风险评估提供客观的脑影像学依据。

本项目主要针对癫痫意识障碍脑影像学表征的科学问题，发展利用信息技术识别癫痫影像学特征的关键方法，研究癫痫患者与意识相关核心脑区的功能和结构改变，挖掘其皮层泛化导致意识受损的网络演化机制，并以此构建癫痫意识障碍的模式预测模型,实现基于数据驱动的局灶性癫痫亚型分类和意识障碍的风险预测。主要研究内容包括：（1）建立了包含174例伴意识受损的颞叶癫痫患者及正常对照的多模态数据库；提出基于分频的脑自发振荡模式和稀疏优化大尺度关键脑环路分析方法，探测伴有/无意识障碍的颞叶癫痫脑功能结构表征异同，标定意识网络的核心脑区，挖掘癫痫意识障碍特异性的脑影像生物学标记；（2）发展了多模态脑影像识别方法框架，从脑白质、脑灰质等不同模态探究癫痫等疾病的病理机制，从静态、动态等方向量化疾病的脑影像改变，并为临床症状的预测提供新的神经标志物，为癫痫影像机制的研究提供方法上的支持；（3）利用发展的多模态脑影像识别方法，标定意识网络的核心脑区，挖掘癫痫意识障碍特异性的脑影像生物学标记，为局部损伤可能影响癫痫各方面复杂临床症状和其他大脑功能网络提供了新证据；（4）建立了癫痫患者深度学习分类模型，实现对癫痫患者的智能识别和癫痫病灶的定位，癫痫分类平均准确率达80.3%。