Modified Electric Convulsive Therapy (MECT) is an efficacious physical therapy in treating mental disorders. The occurrence of epilepsy is a crucial benchmark for evaluating therapeutic effectiveness. However, the medical field still lacks relevant research on automatically detecting epileptic waves in MECT. Therefore, this article proposes a novel automatic detection method of epileptic waves in MECT. In this article, EEG local features (time, frequency, and time-frequency domains) and global features (Pearson correlation coefficient) are combined for epileptic wave detection with SVM (Support Vector Machine). We researched the system with 15 EEG detection channels. The dataset under investigation contains EEG data from 22 patients who received MECT and presented with epileptic seizures. The results revealed that LA (Logarithm of Activity) feature exhibits the best classification significance. When epileptic waves appear, there is a decrease in the power ratio of delta waves and an increase in the power ratio of theta waves. Additionally, the complexity of EEG decreases while the correlation between EEG channels increases. The Cz, F4, and P3 channels exhibit the highest classification significance among all EEG channels. Furthermore, based on the channel classification significance, the EEG detection channels number can be reduced to 8. Similarly, based on the feature classification significance, the local feature number can be reduced from 9 to 3. These conclusions can improve detection efficiency and reduce the cost for MECT. Moreover, the method we proposed can effectively detect epileptic waves in MECT. This work can provide physicians with a reference for evaluating the effectiveness of MECT.
Keywords: Detection of epilepsy; MECT; channel and feature significance; global features; local features.