Mental Effort Detection When Using a Motor Imagery-Based Brain-Computer Interface

A wearable electroencephalographic (EEG) device-based method for the classification of the mental effort during motor imagery is presented. The solution can be used to improve the training of novice surgeons involved in minimally invasive surgery. The method was validated on a public dataset comprising recordings from two participant groups: the control group (engaging in pure motor imagery tasks without feedback) and the neurofeedback group (receiving feedback during their mental tasks). In particular, a previous work on the same dataset found a higher cognitive effort for the neurofeedback group than for the control group, which was confirmed by the results of the NASA-TLX questionnaire. EEG signals were acquired with an 8-electrode dry device. The EEG features of the mental effort were identified by using the Sequential Feature Selector (SFS), in combination with different classifiers, on the EEG data of the control group. In order to classify between low and high mental effort (baseline accuracy of 50 %), four EEG features and the Multi-Layer Perceptron classifier resulted in the best combination for the mental effort assessment in the control group, achieving an average accuracy of 82.1 ± 8.7 %. The 4 features identified were: (i) theta-to-alpha ratio on Fz channel, (ii) beta-to-delta ratio on O1 channel, (iii) theta-to-beta ratio on FP1 channel, and (iv) (thcta+alpha)/beta on FP1 channel. The same pipeline was employed on the neurofeedback group, achieving an average accuracy of 84.0 ± 6.8 %. These findings are in accordance with the results of NASA - TLX questionnaire. This work demonstrated the feasibility of assessing cognitive effort in real-time by means of wearable EEG device during motor imagery tasks. Thus, neurofeedback-supported motor imagery systems can be enriched by a new module to adapt the training to the novice surgeons and optimise learning outcomes.