Neuroscience and Machine Learning Restore Movement in Paralyzed Man’s Hand

Last week, the New York Times reported the first successful “limb reanimation” in a person with quadriplegia. Ian Burkhart, 24, had broken his neck as a teen in a diving accident. His spine was damaged at the fifth cervical vertebra, leaving him paralyzed from the shoulders down. Using nerve bypass technology that transmits his thoughts directly to his hand muscles, he has regained control over his right hand and fingers.

The system effectively bypasses his spinal injury. This is the first time a brain-computer interface has been used to help an individual move his own hands. The research was published in in the journal Nature, and explains how machine learningand MATLABwere used in this project.

This remarkable breakthrough is the result of years of research and a cross-discipline project. Engineers, bioengineers, neuroscientists, and surgeons contributed to the prototype medical system and clinical study. The device, called NeuroLife, was invented at Battelle Institute. The engineers at Battelle worked with physicians and neuroscientists from Ohio State University Wexner Medical Center to develop the research approach and perform the clinical study.

In 2014, Ohio State surgeons implanted a chip in Ian’s brain. The Utah Array chip from Blackrock Microsystems consists of 96 microelectrode sensors that record the firing of individual neurons. The team used brain imaging to identify and isolate the part of Mr. Burkhart’s brain that controls hand movements. During the surgery, the team repeatedly tested exposed brain tissue to pinpoint the correct location for the chip.

After the surgery, Mr. Burkhart spent hours upon hours watching an avatar of a hand on a screen, concentrating on the thoughts it would take to make the movement.  During these sessions, the activity in his brain was recorded and analyzed. The goal was to understand which signals in his brain corresponded with the hand movements he was mentally emulating.

The firing patterns in his brain were transmitted to a computer and recorded. The amount of data this requires is immense. Three million samples of neural activity are collected each second.  Approximately one gigabyte of data is generated from neurons in the motor cortex every three minutes.