Homepage Research Patents Academic Activities Contact me
Introduction:
Neural system is the significant coordinator of the human body, which makes the body as an integral system. At the same time, physical changes of any organ or tissue will also directly or indirectly influence the work of neural system. Hence, neural system detection is not only the main way to find out and cure the neural problems, but also will be helpful for other disease diagnosis. As the control center, brain can reflect any neural action, and furthermore, produce corresponding physical electricity. Electroencephalograph (EEG) is the very production. Compared with some other technology in the field of Biomedical Engineering, EEG holds the advantage of high time-resolution, easy acquisition as well as low cost, and therefore attracts more and more attention from psychology, neural informatics and many other fields. In this project, the fluctuation of EEG is tracked, and then event-related potential (ERP) is extracted from EEG to evaluate the function of sensory system. During the project, considering the adaptability of the skin sense, we try to do ERP extraction from single-trial EEG. Final experimental result demonstrates that our research offers effective tries on practical study and algorithm selection for ERP based Current Perception Detection. The recognition rate is around 91.75%, which indicates that our method is promising to detect the threshold of the current perception objectively.
Research Process:
The research includes mainly two parts: signal sampling and analysis algorithm development:
1. Signal Sampling: During experimentation we generate current with magnitude of 0-2mA and a frequency of 250Hz. We choose the distal end of the finger joint as our point of stimulation since it's where most sensory nerve endings distribute. Our sampling frequency is 2*250=500Hz and we select two point behind ears as reference points. By adding the stimulation, the EEG signal is transfer to the computer to be record. 5 healthy people aged from 20-21 are selected as our subjects. They are sampled consecutively for 15 sets of data, 7 of which are without stimulation and 8 with different levels of stimulation. The blank comparison group and the simulation groups are sampled alternatively with 30 seconds interval. Detailed process is shown in Fig1.
2.Analysis algorithm development: The main process can be generalized as the following steps:
Pre-analysis: After sufficient sampling, we use power spectral density analysis to see that most energy of EEG signal is less than 30Hz, with a 10Hz crest indicating the ยช wave. Power spectral density graph shows that signals less than 3Hz are most distinguishable for stimulations. Hence we use Butterworth filter and set 3Hz as the upper limit to filter the original signal.
Feature point analysis: We take 1500 points both prior and after the stimulation center (approximately 6 seconds) as our main analysis period. According to the lasting time of the stimulation, we take 128 sampling points (about 256ms). Since a vector with 128 dimensions are too complicated spatially and timely, we tend to reduce the dimensions by randomly selecting data from the training set to calculate pattern center of the stimulation group and reference group. Then we calculate the Euclidean distance from individual sample to two central points. With each data set, we get two distances data, hence reducing the dimension from 128 to 2.
Pattern Recognition: By drawing points' distribution of 2 dimension samples, we are able to use SVM to find a linear kernel function that is most suitable for classification.
3. Result analysis: Our analysis has proven to be robust in terms of individuality.The threshold of our data set is approximately 0.36mA, which is consistent with emperical value.
Fig 1 Experiment procedure Fig2 Euclidean distance
My role in the project:
In this project, I'm the project manager in a group of 4 people. Our research is based on a strict timeline I worked out with every step planned meticuluously. Besides managing the team, I'm also responsible for both experimentation and algorithm developmeng, with emphasis on the latter.
Experimentation: In this project, I'm tested as one of the major subject for EEG signal sampling. Our human-based experiments enriched my experiences as a researcher, laying the foundation for furture analysis.
Algorithm development: I came up with the ultimate plan for signal analysis by both referring to previous research results and trials of different methods. Especially in the Patter Recognition part where back propagation artificial neural network, K-neighbor and SVM methods are tried. Eventually I decided that SVM is the most efficient method and successfully lauched our project.
Uploading ....