Automatic Annotation of Hyperventilation and Sleep Stages in Electroencephalogram Examination

The brain is the body's control center, which possesses the ability to regulate thinking, memory, voice, and motion, and also control the function of many organs. The presence of a disease is one of the most complicated disorders in humans, and Electroencephalography (EEG) is a popular method used to make diagnosis in hospitals. In the diagnosis using EEG, a patient is typically attached more than 20 electrodes to the scalp, then recorded the electrical waves in the brain. During a typical test menu around one hour, a supine patient on the bed with EEG and various equipment receives various instructions and stimulations. EEG automatically records the brain waves, however, such events and specific states of brain wave (e.g. sleep and awake) are annotated manually by technicians. To reduce the workload of human annotators, we tried two things for automating the annotation. Firstly, we proposed a new approach of the non-contact capturing method of breathing activities using the Kinect depth sensor for automatic annotation of hyperventilation, which is one of the important events in EEG diagnosis. The time-series mean depth value between Kinect and subject’s breast are further processed by feature reduction step, then classified by Support Vector Machine (SVM). This approach achieved 99% accuracy in the classification of three breathing states including hyperventilation. Secondly, we proposed a method of sleep stage classification. Unlike various existing methods using complicated process of signal filtering, feature extraction, and feature selection, we used high-dimensional features calculated by Fast Fourier Transform (FFT) from single- or multi-channel EEG signals. In the classification of the expanded version of Sleep-EDF dataset with 61 recordings, our method using SVM achieved better or nearly equal performance in comparison with the most recently reported and state of the art method. It means that that our method is useful for the automatic sleep stage annotation in EEG diagnosis.