Optimizing electrical stimulation therapies using machine learning

Similar to a pacemaker, nerve stimulation devices are implanted to send electrical impulses to activate nerves throughout the body. These electrical stimulation devices have been used to treat and control many disorders, including heart disease, epilepsy, depression, and rheumatoid arthritis.

But many variables affect exactly how a nerve responds to stimulation, making the development and use of nerve stimulation therapies difficult and complex.

Neuroscientists at Duke University have developed a computer model that makes it easier to simulate nerve responses to electrical stimulation. The model can simulate the action of more than 50,000 nerve fibers in the time it takes the current industry standard to simulate one. The researchers say the new tool will help design more effective and targeted neuromodulation therapies.

The research appears in the journal Nature Communications and the new tool is available for free.

“There are many possible adjustments that need to be considered to optimize these devices for effective clinical treatment, whether it’s changing the amplitude, duration, shape or frequency of the pulse, or changing the placement of the electrodes,” said Warren Grill, Edmund T. Pratt, Jr. School of Medicine Distinguished Professor of Biomedical Engineering at Duke.

“Neural responses are affected by the anatomy and characteristics of the nerves themselves. You have many options for changing the stimulation parameters, and it’s difficult to know which changes will provide the greatest improvement.”

Engineers have long used a platform called “NEURON” to model how nerve fibers respond to electrical stimulation. The “MRG” model of a nerve fiber is implemented in NEURON and has been widely used in academic research and industry.

Although the MRG model is highly accurate, the computing power required to simulate neural responses limits its speed, creating a bottleneck that prevents the use of MRG in real-time modeling and slows research aimed at improving existing therapies.

To overcome this long-standing obstacle, Grill, Minhaj Hussain, a doctoral student in the Grill lab, and Nicole “Nikki” Pelot, the lab’s research director, developed S-MF (pronounced “smurf”), an alternative to the MRG nerve fiber model. Simulating a population of S-MF nerve fiber models runs thousands of times faster than a population of MRG nerve fiber models, without sacrificing accuracy or detail.

Unlike NEURON and the MRG model, which run on CPUs (central processing units), S-MF runs on GPUs (graphics processing units), a kind of computer chip capable of running thousands of calculations in parallel.

“If we’re modeling a single fiber, S-MF isn’t much faster than NEURON,” Hussain said. “But the big leap is that S-MF takes the same amount of time to simulate several thousand nerve fibers as it does to simulate a single MRG nerve fiber. The human vagus nerve alone contains 100,000 nerve fibers, so this new efficiency is incredibly useful.”

The vagus nerve is a key target for stimulation therapies because it connects the brainstem to most organs in the torso, including the heart, lungs, pancreas, stomach, and liver. Effective stimulation has been shown to safely treat conditions such as drug-resistant epilepsy, depression, and heart failure. However, stimulating off-target fibers in the nerve can cause side effects.

The team simplified how nerve fiber anatomy is represented in their models: the MRG model represents different anatomical features at the micron level along the length of a neuron, while the S-MF model focuses on key features that initiate and propagate neuronal activity. The team used machine learning approaches to define the electrical parameters of the S-MF model to ensure accuracy comparable to that of the MRG model.

“Unlike other studies that have used alternative approaches to speed up these simulations, S-MF is accurate across a wide range of neuronal anatomies and stimulation parameters,” Pelot said. “S-MF also preserves many details that other simplifications have overlooked, providing important insights for designing better therapies.”

The team used S-MF technology to test different stimulation scenarios on thousands of different nerve fibers at once and quickly identify the best conditions for optimal nerve stimulation. The GPU-based design of S-MF technology allowed the team to use machine learning optimization techniques, which are faster than optimization techniques available for NEURON-based models.

To demonstrate the power of S-MF and its optimization through machine learning, the team predicted stimulation parameters that would initiate neuronal activity only in one half of the vagus nerve, while leaving fibers inactive in the other half.

The team’s platform quickly and correctly predicted the stimulation levels and patterns that triggered the desired response in human and porcine vagus nerve models, activating target nerve fibers while avoiding off-target nerve fibers.

Although the S-MF was trained to mimic the MRG model of myelinated fibers, an important target for neuromodulation therapies, the team also demonstrated that their platform could easily be adapted to simulate other types of nerve fibers.

They are studying how their approach can be extended to other neuromodulation techniques, including transcranial magnetic stimulation of the brain, which would require modeling more complex neuronal anatomy and multiple types of neurons in the brain.

“Neural engineering is a field that benefits from having access to scalable, efficient, and anatomically realistic models,” Hussain said. “We hope that as we continue to use this platform, it will tell us more about the design decisions we should make with our stimulation therapies so we can achieve the best possible outcomes.”