A Detection and Prediction Method of Epilepsy Ictal Based on Electroencephalogram by Using Artificial Intelligent Computing

• Main Authors: Chen-hao Kuo, 郭宸豪
• Other Authors: Tsu-wang Shen
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/91583437562403621077

碩士 === 慈濟大學 === 醫學資訊研究所 === 97 === Epilepsy affects a population of 50 million in the world. The electroencephalogram (EEG) is very important biomarker to apply for epilepsy monitoring, seizure detection and seizure prediction. Seizure detection is often combined with spike detection because epilepsy spikes are highly related with epilepsy diagnosis. In general, the five criteria to achieve seizure detection are as follows: 1) sudden desynchronization of background EEG pattern, 2) changing of frequency into a distinct rhythm, 3) showing spiky phase of the oncoming rhythmical waves, 4) increasing in voltage of the new rhythm, and 5) propagation of the new EEG activity. In addition, seizure prediction provides an new solution for drug treatment in order to maximize drug effects, to minimize possible side-effects, and to guide an effective acute invention in the early phase. Unfortunately, seizure prediction existence and predication time horizon are remained as an open issue, and theclinical applicable solution is still not available. The aim of our research is to detection and prediction of epilepsy ictal based on artificial intelligent computing. A novel seizure prediction method based on probability density distribution of Poincaré chaos is introduced. This method not only predicts seizure risk, but also detects the risk of seizure onset. Also, the ant k-means (AK) clustering automatically constructs spike model of each individual for spike detection. Finally, the AI reasoning method is at the end to decide seizure duration. In this research, two databases are used in this research: (1) Freiburg Seizure Prediction EEG (FSPEEG) database with 21 individual EEG data, total 81 seizures are detected. (2) Total 50 EEG spike template and 50 background EEG template from Tzu Chi general hospital. The overall system performance is 88.9% accuracy on seizure detection rate, and average prediction time is 24.96±8.09 minutes on FSPEEG database. Although the EEG analysis system still has room for improving, the preliminary results are encouraging. The tools developed for seizure identification should serve in future neurological expert system development, brain computer interface (BCI), or mental tasks on a patient.