Multiblock PLS model for group corticomuscular activity analysis in Parkinson disease

To explore the cross-information in multi-modal data, several multivariate data fusion techniques have been proposed. Partial least square (PLS) has great potential for neuroimaging studies. However, when performing group analysis with PLS, the presence of inter-subject variability makes the conventional technique of simply pooling data from different subjects problematic. To circumvent this issue, we introduce the idea of multiblock PLS (mbPLS) which incorporates a hierarchical structure into the ordinary two-block PLS. The mbPLS model groups each subject´s detail in individual data blocks on the sub-level, while aggregates the sub-level information to obtain a group “consensus” on the super-level. With the hierarchical, two-level design, the mbPLS provides a trade-off between modeling simplicity and preservation of subject specificity. We applied the proposed mbPLS method to concurrent EEG and EMG data collected from normal subjects and patients with Parkinson´s disease. The decomposition identifies active brain rhythms and spatial activation pattern involved in the generation of the extracted EEG temporal patterns that show high correlation with EMG signals. The proposed mbPLS framework is a promising technique for performing multi-subject, multi-modal data analysis and it allows for robust group inferences even in the face of large inter-subject variability.