融合多特征脑电评估孤独症儿童

孤独症是一种复杂的神经发育性脑疾病,其早期发现和精确诊断非常重要。从54名孤独症和50名正常儿童的脑电信号中提取功率谱、熵、双谱相干性以及相干性等多特征进行分析研究,并对每组特征进行独立样本t检验分析组间差异;为提高分类性能,提出融合多特征脑电进行分析,进一步采用最大相关最小冗余算法进行特征选择,最后利用支持向量机建立分类模型。结果显示,用单一特征分类,得到的分类准确率为72%,灵敏度为73.94%,特异性为67.74%,F1分数为69.74%,因此单一特征所建立的分类模型性能较差;在融合多特征进行分类时,选择前25个特征建立模型,具有较高的分类精度(93.45%±0.79%),此时灵敏度为91.73%±0.42%,特异性为94.01%±0.36%,F1分数为92.54%±0.31%,且AUC达到0.96,相比单一特征分类模型具有良好的性能。研究结果可以为孤独症的辅助诊断提供科学客观依据,为孤独症儿童后期康复提供可靠参考。

Study on Multi-Feature Fusion of EEG to Evaluate Children with Autism

Autism is a complex neurodevelopmental brain disease, early detection and accurate diagnosis are very important. This paper extracted power spectrum, entropy, bispectral coherence, and coherence from the EEG signals of 54 autistic and 50 normal children for analysis and research and conducted independent sample t test for each group of characteristics to analyze the differences between the groups. To improve the classification performance, the fusion of multi-feature EEG was proposed for the analysis, and the maximum correlation minimum redundancy algorithm was further used for the feature selection, and at last the support vector machine was used to establish the classification model. Results showed that the classification accuracy rate obtained by single feature classification was 64%, the sensitivity was 68.25%, the specificity was 65.25%, and the F1 score was 69.19%. The single feature was established and showed poor performance. When fusing multiple features for classification, the first 25 features were selected to build a model that had a high classification accuracy of 93.45%, the sensitivity was 91.73%, the specificity was 84.01%, and the F1 score was 92.54%, and the AUC reached 0.96, which had better performance than the single feature classification model. The results of this study provided a scientific and objective basis for the auxiliary diagnosis of autism, as well as a reliable reference for the later rehabilitation of autistic children.