基于深度神经网络多模态融合的颞叶内侧癫痫鉴别

Diagnosis of Medial Temporal Lobe Epilepsy Based on Deep Neural Network Multimodal Fusion

目前的颞叶内外侧癫痫的诊断分类主要基于电生理-临床诊断体系,该诊断体系的优势在于简单易得。脑电图的获取比较方便,同时基于临床症状的诊断也较容易得到推广,因而该体系成为目前主流的诊断手段。但是,由于不同颞叶内外侧癫痫综合征之间的症状以及脑电图特征存在互相交叉的部分,使得基于该体系的诊断可能面临主观因素的干扰,同时由于医生临床经验的个体差异,使得错误诊断的出现难以避免,更重要的是,基于有限信息的分类体系,可能并没有真正展现出不同颞叶内外侧癫痫之间的真正差异,进而影响对治疗方案的选择以及预后的判断。人工智能算法对数据的整合,使得我们获得新的手段来分析癫痫患者的多维度数据。依赖脑电图信号诊断观察周期长,而磁共振成像在时间上具有更大的优势。但由于磁共振影像可读性差,前人很少基于磁共振影像鉴别癫痫。区别于基于传统脑电图鉴别的方法,提出基于深度神经网络的多模态融合方法进行癫痫鉴别。

The current classification of the temporal and medial epilepsy is mainly based on an electrophysiological-clinical diagnostic system.The advantage of this diagnostic system is that it is simple and easy to obtain.EEG acquisition is convenient,and diagnosis based on clinical symptoms is easier to promote,so the system has become the mainstream diagnostic tool.However,because of the cross-section between the symptoms of different temporal lobe epilepsy syndrome and EEG features,the diagnosis based on this system may be subject to subjec.tive factors.At the same time,due to the individual differences in the doctor's clinical experience,the occurrence of error diagnosis is diffi.cult to avoid.More importantly,the classification system based on limited information may not really show the true difference between the different temporal lobe epilepsy,which in turn affects the choice of treatment options and the prognosis.The integration of data by artificial intelligence algorithms has enabled us to obtain new tools for analyzing multidimensional data of patients with epilepsy.Depending on the EEG signal,the diagnostic observation period is long,and magnetic resonance imaging has a greater advantage in time.However,due to the poor readability of magnetic resonance images,predecessors rarely identify epilepsy based on magnetic resonance imaging.Different from the traditional EEG based diagnosis method,proposes a multi-modal fusion method based on deep neural network for epilepsy diagnosis.