Stochastic Approximator of Motor Threshold (SAMT) for transcranial magnetic stimulation: Online software and its performance in clinical studies

Background: The motor threshold (MT) plays a central role in probing brain excitability and individualizing transcranial magnetic stimulation (TMS). Previously, we proposed stochastic approximation (SA) as a new method for determining TMS MT and demonstrated its excellent speed and accuracy via simulations. SA also has low computational requirements and is theoretically robust to potential model flaws. 

Objective: This project aimed to develop a practical SA thresholding method and assess its performance in clinical studies. 

Methods: The SA thresholding method was implemented as an online software application—SAMT (Stochastic Approximator of MT)—that incorporates features for warning against likely inaccurate MT estimates. Two clinical studies used SAMT and collected 365 small hand muscle MTs from 179 participants to date. SAMT's misestimation warning method marked MTs of 7 thresholding trials as likely inaccurate, and SAMT's performance in the remaining 358 trials was assessed by comparing the MT at each step to the threshold estimated by fitting a sigmoidal probability distribution to the complete muscle response data from the session using maximum likelihood estimation (MLE). 

Results: By the 25th TMS pulse, 99% of the SAMT MTs differed by less than 3.0% (relative) and 1.3% of maximum stimulator output (absolute) from the corresponding fitted MLE sigmoid thresholds and were within the 95% confidence intervals of the MLE thresholds. 

Conclusions: We provide the TMS community with a new thresholding tool, SAMT. Combined with the prior simulation results, the experimental assessment presented here supports the practicality and accuracy of the SA thresholding method and the SAMT software.