OBJECTIVE: Recent reports indicate that making better assumptions about the user's intended movement can improve the accuracy of decoder calibration for intracortical brain-computer interfaces. Several methods now exist for estimating user intent, including an optimal feedback control model, a piecewise-linear feedback control model, ReFIT, and other heuristics. Which of these methods yields the best decoding performance? METHODS: Using data from the BrainGate2 pilot clinical trial, we measured how a steady-state velocity Kalman filter decoder was affected by the choice of intention estimation method. We examined three separate components of the Kalman filter: dimensionality reduction, temporal smoothing, and output gain (speed scaling). RESULTS: The decoder's dimensionality reduction properties were largely unaffected by the intention estimation method. Decoded velocity vectors differed by <5% in terms of angular error and speed vs. target distance curves across methods. In contrast, the smoothing and gain properties of the decoder were greatly affected (> 50% difference in average values). Since the optimal gain and smoothing properties are task-specific (e.g. lower gains are better for smaller targets but worse for larger targets), no one method was better for all tasks. CONCLUSION: Our results show that, when gain and smoothing differences are accounted for, current intention estimation methods yield nearly equivalent decoders and that simple models of user intent, such as a position error vector (target position minus cursor position), perform comparably to more elaborate models. Our results also highlight that simple differences in gain and smoothing properties have a large effect on online performance and can confound decoder comparisons.
OBJECTIVE: Recent reports indicate that making better assumptions about the user's intended movement can improve the accuracy of decoder calibration for intracortical brain-computer interfaces. Several methods now exist for estimating user intent, including an optimal feedback control model, a piecewise-linear feedback control model, ReFIT, and other heuristics. Which of these methods yields the best decoding performance? METHODS: Using data from the BrainGate2 pilot clinical trial, we measured how a steady-state velocity Kalman filter decoder was affected by the choice of intention estimation method. We examined three separate components of the Kalman filter: dimensionality reduction, temporal smoothing, and output gain (speed scaling). RESULTS: The decoder's dimensionality reduction properties were largely unaffected by the intention estimation method. Decoded velocity vectors differed by <5% in terms of angular error and speed vs. target distance curves across methods. In contrast, the smoothing and gain properties of the decoder were greatly affected (> 50% difference in average values). Since the optimal gain and smoothing properties are task-specific (e.g. lower gains are better for smaller targets but worse for larger targets), no one method was better for all tasks. CONCLUSION: Our results show that, when gain and smoothing differences are accounted for, current intention estimation methods yield nearly equivalent decoders and that simple models of user intent, such as a position error vector (target position minus cursor position), perform comparably to more elaborate models. Our results also highlight that simple differences in gain and smoothing properties have a large effect on online performance and can confound decoder comparisons.
Authors: Andrew H Fagg; Gregory W Ojakangas; Lee E Miller; Nicholas G Hatsopoulos Journal: IEEE Trans Neural Syst Rehabil Eng Date: 2009-08-07 Impact factor: 3.802
Authors: Shinsuke Koyama; Steven M Chase; Andrew S Whitford; Meel Velliste; Andrew B Schwartz; Robert E Kass Journal: J Comput Neurosci Date: 2009-11-11 Impact factor: 1.621
Authors: Francis R Willett; Brian A Murphy; William D Memberg; Christine H Blabe; Chethan Pandarinath; Benjamin L Walter; Jennifer A Sweet; Jonathan P Miller; Jaimie M Henderson; Krishna V Shenoy; Leigh R Hochberg; Robert F Kirsch; A Bolu Ajiboye Journal: J Neural Eng Date: 2017-02-08 Impact factor: 5.379
Authors: Jennifer L Collinger; Brian Wodlinger; John E Downey; Wei Wang; Elizabeth C Tyler-Kabara; Douglas J Weber; Angus J C McMorland; Meel Velliste; Michael L Boninger; Andrew B Schwartz Journal: Lancet Date: 2012-12-17 Impact factor: 79.321
Authors: Leigh R Hochberg; Daniel Bacher; Beata Jarosiewicz; Nicolas Y Masse; John D Simeral; Joern Vogel; Sami Haddadin; Jie Liu; Sydney S Cash; Patrick van der Smagt; John P Donoghue Journal: Nature Date: 2012-05-16 Impact factor: 49.962
Authors: Jose M Carmena; Mikhail A Lebedev; Roy E Crist; Joseph E O'Doherty; David M Santucci; Dragan F Dimitrov; Parag G Patil; Craig S Henriquez; Miguel A L Nicolelis Journal: PLoS Biol Date: 2003-10-13 Impact factor: 8.029
Authors: Sergey D Stavisky; Francis R Willett; Donald T Avansino; Leigh R Hochberg; Krishna V Shenoy; Jaimie M Henderson Journal: J Neural Eng Date: 2020-02-05 Impact factor: 5.379
Authors: Francis R Willett; Daniel R Young; Brian A Murphy; William D Memberg; Christine H Blabe; Chethan Pandarinath; Sergey D Stavisky; Paymon Rezaii; Jad Saab; Benjamin L Walter; Jennifer A Sweet; Jonathan P Miller; Jaimie M Henderson; Krishna V Shenoy; John D Simeral; Beata Jarosiewicz; Leigh R Hochberg; Robert F Kirsch; A Bolu Ajiboye Journal: Sci Rep Date: 2019-06-20 Impact factor: 4.379