In 2015, a man named Ian Burkhart picked up a glass bottle, poured its contents into a jar, and then stirred them with a stick. None of this would be remarkable—except for the fact that Ian had been paralyzed from the neck down for five years.

The technology that enabled Ian to regain control of his own hand and fingers to perform this task is called Battelle NeuroLife. The groundbreaking neural bypass device picks up signals from a chip implanted in Ian’s motor cortex and translates them into electrical signals that directly stimulate his muscles using a specialized sleeve.

Between the chip in Ian’s brain and the sleeve that stimulates his muscles lies a powerful analytical engine that uses machine learning to analyze, decode and interpret brain signals and turn them into usable data. These same “big data” methods could be used for a broad range of medical and wellness applications, from stroke rehabilitation to medical imaging.

Machine Learning and Medicine

Our bodies produce millions of data points, from brain signals to vital signs to observable behaviors, but turning this data into usable information is a daunting analytical problem. How do you find meaningful signals in the middle of masses of noise, correctly interpret those signals and apply them for a useful purpose?

Traditional methods require human programmers to tell the software exactly what to look for and what to do with it once it finds it, but big data problems involve far more data than humans can process. Given the volume and complexity of the problem, there is no way to tell the program in advance what solution to look for.

Machine learning—a type of artificial intelligence—is the key to solving the big data conundrum. Machine learning uses classification algorithms that enable analytical programs to detect patterns in large data sets. Instead of providing explicit directions to follow, a computer scientist feeds the program with large sets of training data. For example, an image interpretation program can be trained to recognize cancer cells by processing large data sets with images known to be with and without cancer cells. Once the program has learned from the training set, it can be used to analyze new images. Battelle applied this method to develop an imaging device for detection of cervical cancer.

The same training methods can be applied to virtually any big data problem. In medicine, machine learning is already tackling a variety of problems, including the interpretation of neurological signals.

Turning Brain Signals Into Motion

The brain is the ultimate big data problem: it produces millions of electrical signals per minute to support conscious and unconscious processes that keep us alive and allow us to perceive, think and act. When the pathway between the brain and the muscles is interrupted by a spinal cord injury or other organic problem, we lose the ability to move our muscles consciously.

NeuroLife solves this problem by bypassing the damaged area of the spinal cord entirely. The chip implanted in the motor cortex detects electrical signals that indicate an intention to move in a specific manner. That intention is translated into another set of electrical signals that go to electrodes in a sleeve that wraps around the wrist and forearm. Different patterns of electrode stimulation cause different sets of muscles to contract, allowing the wrist, hands and fingers to move in various ways.

To make this all work, researchers first had to figure out how to separate the signals in the motor cortex that drive intentional movement from the background noise of electrical activity in the brain. Then, they had to figure out how to match different patterns of brain signals to specific intended movements. In other words, the program had to learn which sequence of neurons firing meant “close the hand into a fist,” which meant “point the index finger,” and which meant “bring the thumb and index finger together.”

The NeuroLife algorithm was trained by having Ian repeatedly imagine making specific movements with his wrist, hand and fingers while watching an animated hand perform the same motion on a computer screen. Imagining these motions causes neurons in his motor cortex to fire in the same sequences they would if he were actually making the motion. The chip in his brain collected the data, which was then processed and compressed before sending it to the machine learning algorithm. After many repetitions, the algorithm learned which neural firing patterns corresponded to each motion.

Translating the intended motion into a specific pattern of electrode stimulation in the sleeve also required specialized algorithms that had to be refined over time. Eventually, the program learned how to correlate the neural data into specific patterns of electrode stimulation in order to evoke the intended response in the muscles. For the first time in five years, Ian could make intentional movements with his right hand just by thinking about it.

Beyond Spinal Cord Injuries: Stroke and Parkinson’s

The technology and algorithms that power NeuroLife could also be applied to help people with other kinds of neurological damage or disease.

For example, stroke victims often have movement impairments caused by interruptions in the signals between the brain and the muscles. Sometimes these connections can be regained as the body heals itself, and NeuroLife could be used as part of a physical therapy program to help patients maintain or regain strength in their muscles until they regain conscious control. Alternatively, in cases where full function cannot be restored due to permanent brain damage, a permanent assistive device could be used to enhance strength and dexterity. In either case, the sleeve that stimulates the muscles would not be connected to a brain chip. Instead, the system could be trained to interpret other non-invasive signals—such as eye movements or contraction of muscles that the patient retains control over—to initiate specific motion sequences.

People with Parkinson’s disease could also be helped by technology that bridges the brain and the body. Instead of detecting brain signals and sending them to muscles, the NeuroLife system would detect motion in the hands and use the information to adjust a deep brain stimulation (DBS) device, which has shown tremendous promise in reducing the troublesome tremors experienced by most Parkinson’s patients. A machine learning algorithm connected to a motion sensor could be used to inform calibration of the DBS device by doctors or, eventually, provide real-time calibration based on a patient’s current tremor levels.

Diagnostics and Disease Management

The software used for NeuroLife is not limited to applications that send or receive neurological data. In fact, the machine learning algorithms can be used to detect patterns in any kind of data sets and be applied to a broad range of medical challenges.

Diagnostics, including but not limited to diagnostic imaging, is one of these challenges. Beyond potential applications for smart imaging devices for better cancer detection, machine learning could analyze a broad spectrum of diagnostic lab results and identify biomarkers of disease that may lead to new assays and diagnostic tests. Eventually, machine learning may be used to analyze massive data sets contained in electronic health records to identify patterns for better early detection of disease.

Machine learning is not limited to using official medical records; it can also be applied to data from a device most of us carry every day—a smartphone. One promising application could be in the detection of early warning signs of progressive diseases such as Alzheimer’s. People in the early stages of Alzheimer’s disease often have subtle changes in speech patterns, activity levels or performance on cognitive tasks. Because the change is so gradual, it may not be noticed by the affected person or family members. Smartphone data could detect these early signs or monitor disease progression. Such data could include analysis of speech patterns in phone calls, performance on certain apps or games, and data from motion sensing software or GPS apps. While this kind of tracking raises potential privacy and patient rights concerns, people with high genetic risk factors may choose to move forward for peace of mind or to help them decide when to seek treatment.

Smartphone apps can also be used to collect data for better management of chronic conditions such as diabetes. For example, data from apps that monitor activity levels and food intake in real time could help better regulate a closed-loop insulin pump system.

The Digitized Human: Wellness and Peak Performance

Beyond disease detection and management, there is a world of opportunity for machine learning and data analytics to help people improve overall health and wellness. Millions of people already use diet tracking apps, wearable fitness monitors, sleep monitors and mood tracking programs. Applying machine learning algorithms to all of this data would enable powerful new insights into how our behaviors, health and mood are linked and how to prioritize healthy choices.

This kind of analytics would benefit everyone, from ordinary people trying to live their best lives to elite athletes working to fine tune their performance. An analytics-based training program could tailor individualized exercise programs and recovery periods to maximize performance. Programs that combine multiple types of data might find hidden patterns between food intake or sleep patterns and athletic performance, mood or pain levels. Beyond the individual level, analytics could be used to monitor overall force health and military readiness.

Machine learning will be used to enhance and improve human health, wellness and potential in countless ways. Today, we are using big data analytics to help a paralyzed man regain control of his hand. We are only beginning to imagine the possibilities for tomorrow.