Multiclass classification of hemodynamic responses for performance improvement of functional near-infrared spectroscopy-based brain-computer interface.

摘要: We improved the performance of a functional near-infrared spectroscopy (fNIRS)-based brain–computer interface based on relatively short task duration and multiclass classification. A custom-built eight-channel fNIRS system was used over motor cortex areas in both hemispheres to measure hemodynamic responses evoked by four different tasks (overt execution arm lifting knee extension for sides) instead finger tapping. The were classified using naive Bayes classifier. Among mean, max, slope, variance, median signal amplitude time lag signal, several features are chosen obtain highest classification accuracy. Ten runs threefold cross-validation conducted, which yielded accuracies 87.1%±2.4% 95.5%±2.4%, 77.5%±1.9% 92.4%±3.2%, 73.8%±3.5% 91.5%±1.4% binary, ternary, quaternary classifications, respectively. Eight seconds obtaining sufficient accuracy suggested. bit transfer rate per minute (BPM) investigated. BPM can be achieved from 2.81 5.40  bits/min.