Prognostic Analysis of KCNQ2 Patients via Combining EEG Deep Features and Machine Learning Classifiers

Zheng Zeng,Yan Xu,Yuanfeng Zhou,Ruizhi Su,Linkai Tao,Zaihao Wang,Chen Chen,Wei Chen

DOI: https://doi.org/10.1109/EMBC40787.2023.10341098

2023-01-01

Abstract:Pathogenic variants of the KCNQ2 gene often induces neonatal epilepsy in clinical. For better treatment, infants with confirmed KCNQ2 pathogenic variant and epilepsy symptoms need to adjust their treatment plans according to the outcome after taking antiseizure medicines (ASMs). This process is often time-consuming and requires long-term follow-up, which undoubtedly causes unnecessary psychological and economic burdens. In this study, we investigate the feasibility to predict the outcome of KCNQ2 patients via Electroencephalogram (EEG). By using the combination of deep networks and classical classifiers, the abnormal brain pathological activities recorded in EEGs can be encoded into deep features and decoded into specific KCNQ2 outcomes, thus taking the advantage of both powerful feature extraction capability from deep networks and stronger classification ability from classical classifiers. Specifically, we acquire 10-channel EEG signals from 33 infants with KCNQ2 pathogenic variants after taking ASMs. Two well-trained models (Resnet-50 and Resnet-18) are employed to extract deep features from the EEG spectrums. We achieve an accuracy of 78.7% to predict the KCNQ2 outcome of each infant. To our best knowledge, this is the first study to employ potential EEG pathological differences to predict the outcomes of KCNQ2 patients. The investigation of automatic KCNQ2 outcome prediction may contribute to a more convenient diagnosis mechanism for KCNQ2 patients.