Online learning of stimuli parameters for standing in spinally stimulated paraplegia

Authors: *Y. SUI1, J. W. BURDICK2;
1CMS, California Institute of Technology, Pasadena, CA; 2Caltech, Pasadena, CA

Yanan Sui, Graduate Student at Caltech

Previous research has shown that spinal stimulation via electrode arrays implanted in the epidural space over the lumbosacral area enables paralyzed patients to achieve full weight-bearing standing, improvements in stepping, and partial recovery of lost autonomic functions. The optimal stimulus varies significantly across patients. And even for the same patient, the outcome of the same stimulus varies from trial to trial. Thus, clinicians must determine the optimal stimulus for each patient, under noisy conditions, from a large decision space. Currently, the search for the optimal stimulating parameters is a laborious approach that consumes valuable clinician and patient time, and does not guarantee an optimal outcome.

We developed and tested a Correlated Dueling Bandit algorithm to automatically plot the large decision space of possible stimuli. This algorithms selects and improves the spinal stimulation parameters for standing ability in epidurally stimulated paraplegics, and optimizes them over time. In practice, the algorithm chooses a stimuli, whose effect on the subject is tested and ranked by observing clinicians. The algorithm then balances the exploration for more optimal stimuli while also exploiting currently known good ones to provide effective therapy. The algorithm also seeks to maximize total performance during the limited clinical period within which we can search for the optimal solution.

The standing skill of two paraplegic subjects implanted with 16-electrode epidural implants was tested in response to 90 and 117 different stimuli respectively over two non-consecutive weeks. The algorithm chose stimuli that enable the subjects to achieve full weight-bearing standing, consistently improved standing performance over the evaluation period. Moreover, the optimal stimuli sets found by the algorithm included the same stimulating parameters that were selected for each subject by clinical staff using an intuitive search process, validating the effectiveness of the approach.

Disclosures: Y. Sui: None. J.W. Burdick: None.

Grant Support
NIH U01 EB007615-08
NIH U01 EB015521-05

Society for Neuroscience LINK

Yanan Sui Homepage LINK

SCRIR&A: By utilizing an algorithm program, scientists can detect the “sweet spot” to increase efficacy of the movement they are attempting to create with the implanted stimulator. The parameters are being validated by the staff observations in real time. The amount of stimulation needed can be different for each patient and each training session, so by utilizing a computer program, it can help make the decision changes faster and more precise.

This entry was posted in Chronic Spinal Cord Injury Research, Neuroscience Abstracts, Rehabilitation, Spinal Research and tagged , , . Bookmark the permalink.