- Hans Christian Riis
- Gynter Schneider
4. term, Biomedical Engineering and Informatics, Master (Master Programme)
Spinal cord injury (SCI) occur to a great extent with approximately 12.000 incidents annually in the US, inducing great expenses for carrying and treatment. SCI patients encourage restoration of gait control as one of the four most prioritized functions to regain for improving quality of life. In order to aid and restore gait control it is necessary to bridge and understand the gab between neural activity for voluntary movements and themuscle response, going around the spinal cord. Objective of this report was to evaluate an artificial neural network (ANN) for predicting muscle activity during gait in
healthy rats. Neural activity from motor cortex (M1) and EMG signals from biceps femoris (BF) and vastus lateralis (VL) were obtained by use of 16 channel intracortical electrode arrays and intramuscular
EMG electrodes in a bipolar configuration. Four Sprague-Dawley rats were trained to walk on a treadmill with 0o and 15o inclination. Kinematics were calculated from high-speed camera recordings by digitizing toe, heel, knee, hip and reference markers attached on the rats. Joint angles were calculated and used to detect and extract specific gait cyclesmeeting the inclusion criteria.
Peri-stimulus time histograms (PSTH) were calculated for the intracortical signals and mean envelopes of maximal EMG value for BF and VL. Recordings from two rats, meeting the inclusion criteria, were used in an ANN configured on the basis of a previous study. PSTH of neural activity were used as input and envelopes of average maximal EMG value for BF and VL for output. Results from the ANN gave low R2 values (R2 = 0.1020) yielding an optimization problem. A systematic optimization process of the ANN improved the R2 value (R2 = 0.4146) and demonstrated possibilities in predicting
muscle activity by use of an ANN with neural activity from M1 as input. Further advancement and usage of prediction for control signal of FES further attention and evaluation is needed,
Findings implied future possibilities for integration in BMI applications for restoring gait control of SCI patients, if further refinement of the ANN and data are done.
healthy rats. Neural activity from motor cortex (M1) and EMG signals from biceps femoris (BF) and vastus lateralis (VL) were obtained by use of 16 channel intracortical electrode arrays and intramuscular
EMG electrodes in a bipolar configuration. Four Sprague-Dawley rats were trained to walk on a treadmill with 0o and 15o inclination. Kinematics were calculated from high-speed camera recordings by digitizing toe, heel, knee, hip and reference markers attached on the rats. Joint angles were calculated and used to detect and extract specific gait cyclesmeeting the inclusion criteria.
Peri-stimulus time histograms (PSTH) were calculated for the intracortical signals and mean envelopes of maximal EMG value for BF and VL. Recordings from two rats, meeting the inclusion criteria, were used in an ANN configured on the basis of a previous study. PSTH of neural activity were used as input and envelopes of average maximal EMG value for BF and VL for output. Results from the ANN gave low R2 values (R2 = 0.1020) yielding an optimization problem. A systematic optimization process of the ANN improved the R2 value (R2 = 0.4146) and demonstrated possibilities in predicting
muscle activity by use of an ANN with neural activity from M1 as input. Further advancement and usage of prediction for control signal of FES further attention and evaluation is needed,
Findings implied future possibilities for integration in BMI applications for restoring gait control of SCI patients, if further refinement of the ANN and data are done.
| Language | English |
|---|---|
| Publication date | 1 Jun 2012 |
| Number of pages | 117 |