Almost all BMIs used for stroke neurorehabilitation have been noninvasive, so far. Yet having provided remarkable results, there are some challenges ahead that have to be mastered before broad and out-of-the-lab application of assistive and rehabilitative BMI will belong to the standard treatment options of severe motor paralysis. While low-frequency rhythms, such as SMR, showed limited correlation and contingency with intended movements [80–83], decoding of other brain signals, e.g., high gamma band and action potentials (single- or multiunit) [81–83], allowed for control of high degree-of-freedom prosthetic limbs [9, 36] or FES [12]. As it is hypothesized that Hebbian plasticity involving simultaneous activation of pre- and postsynaptic neurons is critical to driving functional connectivity in an optimal way, invasive BMIs or BMIs providing repetitive and precise contingency between efferent and afferent neural fibers’ activity may prove superior compared to previously used approaches [84]. Recently, it was demonstrated that high gamma band signals allow reconstruction of highly fractionated movements [28] most effectively obtained by invasive recordings using intracortical, subdural, or epidural electrodes [81, 82]. While intracranial electrodes require implantation, epidural or subdural electrodes could ultimately be implanted through a small burr hole reducing risk and cost of surgery.

A substantial barrier here is the lack of fully implantable, wireless intracranial devices. Once fully internalized systems are available, the risk–benefit ratio for paralyzed individuals may change significantly. Similarly, as noninvasive BMIs advance, these may, if providing comparable motor function similarly, change the risk–benefit ratio. Ultimately, the decision of a BMI implantation will be strongly influenced by individual patient circumstances.

In a recent study, the primary motor cortex of a rat was injured leading to a disruption of fibers between the motor and somatosensory areas [85]. An implanted neural prosthesis that translated action potentials in premotor cortex into contingent electrical stimulation in somatosensory cortex led to motor restoration of grasping movements that was indistinguishable from movements performed before the lesion.

Another study used ECoG signals and linked them to direct electric stimulation of a monkey’s spinal cord’s anterior horns. Besides reestablishing motor control of lower limb movements, this approach may prove useful for the rehabilitation of spinal cord injuries or subcortical stroke [86].

Although there is increasing clinical evidence for the efficacy of BMI-related tools in stroke neurorehabilitation, more and larger clinical studies are needed. Also, it is critical to further investigate the underlying mechanisms of BMI-induced functional recovery. A better understanding of the mechanisms underlying motor recovery could also lead to identification of specific and reliable biomarkers predicting treatment response (e.g., [57, 65]. The optimal dosage (frequency and intensity) of BMI training and/or brain stimulation applied in its context still needs to be investigated. Expansion of BMIs from hospital-based to home-based applications may further facilitate applicability of BMI in stroke neurorehabilitation and lower the threshold of including these devices in daily rehabilitation programs. In cases in which ipsilesional BMI training is not feasible, exploration of other strategies, e.g., training of contralesional, ipsilateral brain activity, may be effective [87, 88].

Another challenge is to identify and provide optimal frameworks for generalization of skills learned in the lab or hospital to the patient’s daily life activities. BMI systems following a hybrid assistive and rehabilitative approach may fill this gap and facilitate generalization of motor function reducing the necessity of behavioral physiotherapy. An important aspect when evaluating the efficacy of novel treatment strategies in severe stroke with no residual movements is the availability of reliable and valid instruments for assessment of motor function, as the established assessment scales (e.g., the Fugl-Meyer scale, Wolf test, NIH stroke scale, or ARAT) are rather insensitive to small but significant changes in upper limb movements. Kinematic measures such as sub-movement changes, joint flexibility, and proximal and distal limb movement accuracies [84, 89] should be taken into account as complementary measures to assess BMI-related functional motor recovery in severely affected stroke survivors.