The combined analyses of brain and muscle signals can be very useful in assessing the cortico-muscular connectivity and reorganization in brain injury. A basic approach would be to assess the sensorimotor (SMR) rhythms with electroencephalography (EEG) during the muscle activation in a pinch-and-hold task. For this paradigm, a cue is presented to the subject to initiate and stop the pinch-and-hold. However, we have found that this simple task can become quite challenging to perform repeatedly and robustly in children with hypertonic or hypotonic muscles, such as in hemiplegic cerebral palsy (HCP). We observed that many children found it difficult to respond immediately to a cue, hold and maintain the pinch on a force gauge, with a specific force, for the length of a trial for multiple trials. In these cases, a self-paced movement execution task appears more appropriate, albeit without an EEG-EMG time-locking event. Inspired by these experimental challenges in assessing the cortico-muscular connectivity in children with HCP, we developed a novel approach to extract SMR spatial patterns from EEG signals using the dynamic information of the corresponding EMG activation. Essentially, in this method the aim is to find EEG sources that can explain the variation of muscle activity. However, due to the different nature of frequency components between EEG and EMG, usual cross decomposition algorithms (CCA, PLS) are not easily applicable towards this goal. Recently, mSPOC was proposed as a novel approach to solve such an issue employing a complex iterative procedure to find components that maximized the envelop of correlation between sources of two sets of multivariate signals. Here, we introduce a simpler alternative methodology, based on Riemannian geometry, that allows to objectively extract EEG spatial patterns that explain variations of EMG power. We show a proof-of-concept in healthy subject’s dataset.