Preprint: an algorithm found the optimal interval between nerve and brain stimulation faster in adult recordings

Preprint: an algorithm found the optimal interval between nerve and brain stimulation faster in adult recordings

In a preprint posted on September 8, the authors measured how the interval between two stimuli affected the electrical response of a leg muscle in ten adults and eight children, then retrospectively simulated sequential interval selection on these recordings. In nine out of ten adults, the algorithm identified an interval producing a high individual response within a median of 10 stimulus pairs, compared to 20 for random search. In children, the model predicted individual response profiles less accurately.

In the experiment, an electrical pulse was delivered to the common peroneal nerve at the knee, which is involved in lifting the foot, and then a magnetic pulse was applied to the motor cortex, the brain region that controls movement. Surface electrodes recorded the electrical response of the tibialis anterior, the muscle that lifts the foot. By varying the interval between the two pulses, the researchers controlled the timing of the cortical stimulus relative to the peripheral nerve signal.

First, for each participant the authors measured the latency of the muscle response to a single magnetic pulse. Relative to that latency, they tested 26 intervals ranging from an additional 5 to 130 milliseconds in steps of 5 milliseconds. Each interval was repeated ten times in adults and six times in children. From these measurements a personal response profile was constructed, showing at which time points the muscle response was higher or lower than baseline. On average, adults showed increased responses at longer intervals, but individual participants differed in the shape of their profile, the position of the peak, and even the direction of the response change.

The algorithm sequentially revealed pre-recorded responses, simulating the course of an experiment. After three randomly chosen stimulus pairs, a Gaussian process (a statistical model that predicts the response from nearby intervals and indicates where the prediction is still uncertain) selected the next interval to test. Each recommendation was checked against the participant's complete response profile. The target was an interval yielding a response at the 90th percentile between that person's weakest and strongest recorded responses.

In nine out of ten adults, both strategies reached this target, but the median was 10 stimulus pairs for the algorithm and 20 for random selection. In children, the algorithm reached the target in five out of eight participants, and random search in four; the medians were 45 and 56 pairs, respectively. In a validation step, the model predicted a single held-out interval from the remaining data. The mean R², a measure of how well predictions matched the measurements, was 0.71 in adults and 0.18 in children. The authors attribute the less accurate profiles in children to greater response variability and to the fact that each interval was repeated six rather than ten times.

In a 2023 study, Marco Bonizzato and colleagues had already used a similar algorithm to select neurostimulation parameters in real time in rats and monkeys. The authors of the current preprint applied the algorithm to a single parameter in humans: the interval between electrical stimulation of a peripheral nerve and a magnetic pulse over the motor cortex.