Evaluation of a neural network for predicting muscle activity based on intracortical signals during gait in healthy rats - towards restoration of locomotion of SCI using BMI

Evaluation of a neural network for predicting muscle activity based on intracortical signals during gait in healthy rats

Riis, Hans Christian ; Schneider, Gynter

4. term

Biomedical Engineering and Informatics, Master

2012

2012-06-01

117

Rygmarvsskader (SCI) rammer cirka 12.000 mennesker årligt i USA og medfører høje udgifter til pleje og behandling. At genvinde kontrollen over gang er blandt de højest prioriterede mål for personer med SCI. En mulig vej er at omgå den skadede rygmarv ved at oversætte hjernens signaler for viljestyrede bevægelser til muskelaktivering. Denne rapport undersøger, om et kunstigt neuralt netværk (ANN) kan forudsige muskelaktivitet under gang hos raske rotter. Aktivitet fra den motoriske cortex (M1) blev registreret med 16-kanals intracortikale elektroder, og muskelaktivitet (EMG) fra biceps femoris (BF) og vastus lateralis (VL) blev målt med intramuskulære elektroder. Fire rotter blev trænet til at gå på et løbebånd med 0° og 15° hældning. Højhastighedsvideo med markører på tå, hæl, knæ, hofte og et referencepunkt blev brugt til at beregne ledvinkler og udvælge ensartede gangcyklusser. Hjernedata blev opsummeret som peri-stimulus time histograms (PSTH; en måde at sammenfatte signalmønstre pr. skridt), og EMG-signaler blev omdannet til gennemsnitlige konvolutter (glatte profiler for muskelaktivering). Optagelser fra to rotter, der opfyldte inklusionskriterierne, blev brugt til at træne en ANN baseret på en tidligere konfiguration. Modellen modtog PSTH som input og forudsagde EMG-konvolutter for BF og VL som output. Den første model havde lav forklaringsgrad (R2 = 0,1020). En systematisk optimering forbedrede præstationen (R2 = 0,4146) og viste, at aktivitet i M1 delvist kan forudsige benmusklernes aktivitet under gang. Yderligere forbedring af modellen og data er nødvendig, men sådanne forudsigelser kan på sigt bruges som styringssignaler til funktionel elektrisk stimulation (FES; svage elektriske impulser, der aktiverer muskler) og indgå i hjerne–maskine-grænseflader (BMI) med henblik på at genoprette gangkontrol efter rygmarvsskade.

Spinal cord injury (SCI) affects about 12,000 people each year in the United States and leads to high care and treatment costs. Regaining control of walking is among the top priorities for people with SCI. One potential approach is to bypass the injured spinal cord by translating brain signals for voluntary movement into muscle activation. This report evaluates whether an artificial neural network (ANN) can predict muscle activity during walking in healthy rats. Activity from the motor cortex (M1) was recorded with 16-channel intracortical electrodes, and muscle activity (EMG) from the biceps femoris (BF) and vastus lateralis (VL) was measured with intramuscular electrodes. Four rats were trained to walk on a treadmill at 0° and 15° incline. High-speed video with markers on the toe, heel, knee, hip, and a reference point was used to compute joint angles and select consistent gait cycles. Brain signals were summarized as peri-stimulus time histograms (PSTH; a way to condense signal patterns for each step), and EMG signals were converted to mean envelopes (smoothed profiles of muscle activation). Recordings from two rats that met the inclusion criteria were used to train an ANN based on a previous configuration. The model took PSTH as input and predicted EMG envelopes for BF and VL as output. The initial model showed low explained variance (R2 = 0.1020). A systematic optimization improved performance (R2 = 0.4146), indicating that M1 activity can partly predict leg muscle activity during walking. Further refinement of the model and data is needed, but such predictions could eventually provide control signals for functional electrical stimulation (FES; small electrical pulses that activate muscles) and be integrated into brain–machine interfaces (BMI) aimed at restoring gait control after spinal cord injury.

[This abstract was generated with the help of AI]

Download
View record in AAU Student Projects