A stroke in the brain is characterized by a deprivation of oxygen in some brain regions. When this happens in the motor cortex, one or more parts of the body result impaired. This causes a dramatic alteration on the brain network and a consequent process of reorganization of the different cerebral districts (brain plasticity). In order to restore the motor function, different rehabilitative treatments have been introduced. Among these, Brain Computer Interfaces (BCIs), have been aroused considerable interest in recent years due to the experimental evidences that have been demonstrated their effectiveness in the rehabilitative context. Many studies have shown how the restoring of motor function with BCI is related with the increase of connections between the two hemispheres. Thus, the mere study of changes in lesioned brain areas is not sufficient to describe the brain plasticity and it is necessary to use brain connectivity. Although scientific evidences support the use of brain connectivity to study the effects of stroke on brain network, monitoring the evolution of the connections between hemispheres, within (intra-session) and across (cross-session) different sessions, it is still a challenging issue. This is mainly due to methodological limitations of the brain connectivity estimators. In fact, in the intra-session case the number of data samples available for the estimation process is not adequate to provide an accurate estimation. Instead, in the cross-session case the accuracy and reliability of brain connectivity estimators have not been tested. The main objective of this project consist in the development, implementation and testing of an approach based on artificial neural network (ANN) that can overcomes these limitations. The final toolbox will be validated on 28 stroke patients involved in an rehabilitative intervention based on BCI technology. In conclusion, this project could provide an ANN based toolbox for monitoring the recovery of stroke patients.
This project will provide a toolbox for brain connectivity analysis based on artificial neural network, to be used when the number of data samples available for the estimation process is enough as well as the number of data samples available is not adequate. This new unified approach will be tested on 28 stroke patients to demonstrate the ability to follow the changes in the interhemispheric connections within and across the different sessions of a BCI-based motor training. The toolbox and the knowledge acquired after the test on stroke patients could have a strong impact on several clinical applications.
In conclusion, the innovation of the project would cover several aspects:
-methodological: the project will produce a new toolbox for brain connectivity evaluation able to track brain circuits even when few data samples are available for the estimation process. The low computational cost due to the use of computational graph could open the way to the evaluation of brain connectivity in real-time. Thus, it could be possible to integrate the information provided from a brain connectivity analysis in a real BCI system.
-clinical: the capability to follow in real-time the variability of interhemispheric connections will be exploited as a neurophysiological indicator of recovery outcome at single patient level. Furthermore, most of the studies do not take into account on the wide heterogeneity of patients and their different lesion size and location. In such sense it could be possible to assess on the recovery of the single patients instead of making an hypothesis on the entire experimental group.
The new toolbox could be also used in order to understand the dynamic of cortical motor activity that is not completely clear. As well as in stroke patients, the new approach can be used to study other forms of pathologies affecting the central nervous system in which a variation in interhemispheric connections at rest has been highlighted. For instance, epilepsy, language disorders, schizophrenia, autism and mental retardation.
- neurophysiological: because many cognitive processes are not stationary and nonlinear, a toolbox based on ANN could take into account the non-idealities of such processes. The toolbox could be employed in all those contexts in which it is important to evaluate the inter-trial variability such as social neuroscience or during cognitive task. In this context, it has been proved that exists a relationship between the fluctuation of EEG signals and behavioural performance during a cognitive task. These fluctuations are often considered as noise and treated as a variance of no explanatory value. For this reason, this new toolbox could be used for evaluating inter-trial variability in terms of brain connectivity instead of considering behavioural performance.