Individualized Inter-Stimulus Interval Estimation for Neural Facilitation in Human Motor System: A Particle Filtering Approach

Abstract

Quantitative understanding of the human neuromotor system is essential for the implementation of the future robotic therapeutic exercises. For this purpose, sensorimotor adaptations in voluntary and involuntary movements facilitated by peripheral stimulation and resultant motor-evoked potentials (MEP) must be well characterized. One such facilitation exercise is paired associative stimulation (PAS). However, effective inter-stimulus intervals between cortical and peripheral stimulations are highly variable between individuals due to different physiological characteristics. Past studies measured MEPs in a wide range of time by incrementally varying inter-stimulus intervals to find the optimal interval in a specific subject, which has been a time-consuming process. This paper develops a search algorithm based on particle filtering to estimate individualized inter-stimulus intervals for PAS with mechanical muscle tendon stimulation realized by a pneumatically-operated robotic neuromodulatory system. The particle filter-based method reduces the number of PAS trials 70%–80% in comparison to the conventional incremental method. An accelerometer attached to the robotic system that measures exact timings of tendon stimulation can further reduce the number of trials.

Statistical Inter-stimulus Interval Window Estimation for Transient Neuromodulation via Paired Mechanical and Brain Stimulation

“…On the other hand, nonparametric methods require many data points and relatively long processing time. PF was previously implemented by the authors (Takemura et al, 2018) where 30 particles were adopted.…”

“…Based on the locations and weights of particles, particles were redistributed in the next iteration to update the estimation of ISI-W. Detailed explanation on PF can be found in Takemura et al (2018). NR, GPR, and SVMR first measured MEP for (E) Determination of ISI-W from single-Gaussian fitting in NR and PF.…”

“…For this reason, the figure evaluated the initial number of observations only for NR, SVMR and GPR. For PF, 30 particles were used from our previous study (Takemura et al, 2018) that resulted in the best estimation performance among five different numbers of particles. In each graph, statistical significance was evaluated between neighboring bar plots within the same condition: without ACC and with ACC.…”

“…ISI-W estimation methods must consider the stochastic nature of the human neuromotor system responses as shown in Figure 2E where large variation in MEP amplitudes was observed. Particle filtering technique was applied earlier (Takemura et al, 2018) in which particles determined when to apply TMS relative to Mstim to update an estimate of the MEP profile in an iterative manner. ISI-W estimation based on particle filtering showed potential to reduce the required number of observations.…”