Decoding of Self-paced Lower-Limb Movement Intention: A Case Study on the Influence Factors

Abstract

Brain-machine interfaces (BMIs) have been applied as new rehabilitation tools for motor disabled individuals. Active involvement of cerebral activity has been shown to enhance neuroplasticity and thus to restore mobility. Various studies have focused on the detection of upper-limb movement intention, while the fewer study has investigated the lower-limb movement intention decoding. This study presents a BMI to decode the self-paced lower-limb movement intention, with 10 healthy subjects participating in the experiment. We varied four influence factors including the movement type (dorsiflexion or plantar flexion), the limb side (left or right leg), the processing method (time-series analysis based on MRCP, i.e., movement-related cortical potential or frequency-domain estimation based on SMR, i.e., sensory motor rhythm) and the frequency band (e.g., delta, theta, mu, beta and MRCP band at [0.1 1] Hz), to estimate both single-trial and sample-based performance. Feature analysis was then conducted to show the discriminant power (DP) and brain modulations. The average detection latency was -0.334 ± 0.216 s in single-trial basis across all conditions. An average area under the curve (AUC) of 91.0 ± 3.5% and 68.2 ± 4.6% was obtained for the MRCP-based and SMR-based method in the classification, respectively. The best performance was yielded from plantar flexion with left leg using time-series analysis on the MRCP band. The feature analysis indicated a cross-subject consistency of DP with the MRCP-based method and subject-specific variance of DP with the SMR-based method. The results presented here might be further exploited in a rehabilitation scenario. The comprehensive factor analysis might be used to shed light on the design of an effective brain switch to trigger external robotic devices.

Keywords: brain-machine interface (BMI); electroencephalography (EEG); lower-limb movement; movement-related cortical potentials (MRCPs); onset detection; sensory motor rhythms (SMRs).

Figure 1

Protocol of the experiment. The movement type refers to ankle dorsiflexion and plantar flexion, and the limb side refers to left and right legs. The movement type was consistent within each run and the limb side was indicated by the directional cues.

Figure 2

Topographic representations of the brain activity at different time points from −2.5 s to 1.5 s with respect to the actual movement onset.

Figure 3

Grand average electrophysiology signals around the motor cortex over all subjects and conditions. The actual movement onset is shown with vertical dashed lines. (A) Grand average time-amplitude representations at [0.1, 1] Hz. (B) Grand average time-frequency representations at [0.1, 30] Hz. Only significant values (bootstrap p < 0.05) are colored and non-significant values are plotted in green.

Figure 4

Sample-based performance (AUC) for different movement types and processing methods. We use dorsi and plantar to represent dorsiflexion and plantar flexion for simplicity. The boxplot shows the median (central mark) and 25th/75th percentiles (edges of the box) of the AUC values.

Figure 5

Single-trial detection performance using the MRCP-based method (upper), the SMR-based approach (middle), and the concatenate model combining MRCPs and SMRs (lower) from a typical subject (s10). The detection was performed from −4 to 2 s with respect to the actual movement onset. The blue lines display the detection points when consecutive 3 samples have a detection rate significantly above the chance level (p < 0.05). The chance level is shown in red dashed lines. The black curves and gray regions depict the mean and standard deviation of the detection rate.

Figure 6

DP maps calculated by the modified CVA to show the consistency in feature selection across all subjects and conditions. The features are the channels and time points in the 1-s window for the MRCP-based method and channels and frequency bands for the SMR-based method.

Figure 7

Topographic maps to show the normalized DP index of each channel averaged across all subjects and conditions. The weights were assigned by the modified CVA for the MRCP-based and SMR-based methods.

Figure 8

Sample-based performance (AUC) across different frequency bands using the SMR-based processing method. The meaning of the boxplot was the same as that in Figure 4.

Figure 9

Mean and standard deviation of sample-based performance (AUC) across different frequency bands using the combination of MRCP and SMR features.