基于多模态数据机器学习的方法构建VNS治疗儿童难治性癫痫的疗效预测模型研究

迷走神经刺激（Vagus nerve stimulation, VNS）是通过刺激左侧迷走神经，治疗药物难治性癫痫的一种植入式神经调控技术，已成为儿童难治性癫痫的重要治疗手段。但儿童难治性癫痫受多种因素的影响，VNS的疗效存在很大的个体差异，因此实现儿童难治性癫痫VNS疗效的前瞻性预测具有很大的临床需求和应用价值。国内外既往研究多采用单一的数据类型预测VNS的疗效，尚无采用多模态数据机器学习的方法构建儿童难治性癫痫的疗效预测模型研究。本研究拟首先深入分析既往开展的多中心临床试验中所采集的儿童难治性癫痫患者术前的多模态数据，探索与VNS疗效相关的多模态特征集合；继而将其用于SVM和XGBoost算法的机器分类学习，并建立性能最佳的VNS治疗儿童难治性癫痫的疗效预测模型；最后前瞻性收集多中心、大样本量的多模态数据，检验疗效预测模型的预测性能并进一步进行优化。

Vagus nerve stimulation (VNS) is an implantable neuromodulation technology that stimulates the left vagus nerve to treat drug-refractory epilepsy, which has become an important treatment method for children with refractory epilepsy. However, children with refractory epilepsy are affected by many factors, and the efficacy of VNS has great individual differences. Therefore, the prospective efficacy prediction of VNS for children with refractory epilepsy has great clinical needs and application value. Previous studies at home and abroad mostly used a single data type to predict the efficacy of VNS. There is no study on establishment of efficacy prediction model for children with refractory epilepsy based on the multi-modal data machine learning. This study intends to firstly in depth analyze the preoperative multimodal data of children with refractory epilepsy collected in previous multicenter clinical trials, and explore the set of multimodal features related to the efficacy of VNS. Then the multimodal features will be trained in SVM and XGBoost algorithm machine learning, and the best performance prediction model of VNS for refractory epilepsy in children will be established. Finally, large sample size multimodal data in multi-center are collected to test individualized prediction performance of the efficacy prediction model and further optimize it.