基于多时间尺度癫痫网络模式学习的常规MRI阴性癫痫分型研究

Classification diagnosis is the basis for clinical treatment of epilepsy, while which confronts with challenge in the epilepsy patients with negative conventional MRI. Advanced imaging techniques combining multi-modal MRI and pattern learning provide feasible tool for solving the problem. However, there lack of neuroimaging features for epilepsy type classification. Based on the epilepsy network theory and our previous research, this project proposes a model of "Multi-temporal scale epilepsy network of cortical-thalamic-striatum-cerebellar structures". Three types of MRI-negative epilepsies, including idiopathic generalized epilepsy, temporal lobe epilepsy and frontal lobe epilepsy, all of 1000 patients are included in this project. Multi-modality MRI techniques of EEG-fMRI, diffusion MRI and morphometric MRI are used to construct the functional network associated with epileptic discharges at short timescale, anatomical white matter fiber network associated with seizure events at medium timescale and structure covariance network associated with epilepsy progression at long timescale, respectively. This research will shade lights on the epilepsy network theory. Furthermore, based on the existing epilepsy neuroimaging database platform, we use artificial intelligence algorithms, e.g., convolutional neural network for extracting imaging features, and pattern learning for classifying epilepsy types, to develop an imaging tool for epilepsy diagnosis. This project will promote clinical translation of advanced MRI techniques.

癫痫分型诊断是癫痫临床诊治的基础，常规MRI阴性癫痫是分型诊断的难点；多模态MRI结合模式分类是解决该问题的重要突破方向，但当前缺乏适宜的癫痫分类影像模型及特征标记。基于癫痫症状分类的网络理论及课题组前期结果，本项目提出“不同时间尺度皮层-丘脑-纹状体-小脑癫痫特征网络”模型假说，以1000例原发全面性癫痫、颞叶癫痫与额叶癫痫等3类常规MRI阴性的成人癫痫为研究对象，采用同步EEG-fMRI、DTI及形态学MRI数据，构建痫样发放（电活动）相关的短时间尺度功能网络、癫痫发作事件相关的中时间尺度白质纤维解剖网络，以及癫痫疾病进程相关的大时间尺度结构协变网络，研究不同癫痫类型的网络机制；采用卷积神经网络及模式分类等人工智能算法，提取影像特征指标、建立MRI阴性癫痫影像学分型模型；结合既有的癫痫数据库平台，开发辅助癫痫分类的影像诊断工具；促进高级MRI脑影像技术的临床转化。

常规MRI阴性癫痫是癫痫临床分型诊断的难点，结合定量分析的多模态MRI技术为该难点解决提供了关键指标，快速发展的模式识别、深度学习等人工智能技术，为髙维度的多影像指标在癫痫分类及预测诊断提供了工具。本项目以1000余例原发全面性癫痫、内侧颞叶癫痫与额叶癫痫为研究对象，并辅以间期癫痫活动发放频繁的儿童癫痫数据，以同步EEG-fMRI、DTI及形态学MRI多模态影像数据为基础，构建不同时间尺度的癫痫网络，开发和筛选适宜的网络指标和局域性质指标，建立癫痫影像学分型模型；结合既有的癫痫数据库平台，开发辅助癫痫分类的影像诊断工具；为“不同时间尺度皮层-丘脑-纹状体-小脑癫痫网络”模型假说提供证据，加深对不同癫痫类型网络机制的认识；并促进高级MRI脑影像技术的临床转化。