In the framework of network physiology, the human body can be seen as a functional network depicting the interactions between different complex systems. Due to the different oscillatory activity of each biological system, this exchange of information is usually evaluated with a multiscale approach based on the information theory. This approach allows to consider in a single model all the time series recorded by the different anatomical districts. However, there is a problem of loss in accuracy when the data available, for the evaluation of the physiological interactions, are not enough. In addition to the description of the functional links, in recent years many efforts have been made for the simultaneous analysis of multiple signals for the physiological state evaluation. Many of the approaches developed are based on scientific evidence that correlates the variations of certain biological signals with the state of the subject. In fact, machine learning models used for these purposes are trained with features that do not take into account the relationship between the different biological signals, but only how they individually vary with the state of the subject. To overcome these methodological limitations, this project aims to develop an approach based on information theoretic measures and machine learning models for the detection of the physiological state in humans. The approach will be tested on 20 healthy subjects during the execution of a motor task on which muscular and brain signals were simultaneously recorded. The entire dataset will be made available by the involvement of the Fondazione Santa Lucia in Rome.
This project will provide a new approach for biological signals analysis with the possibility of: 1) performing a multiscale analysis based on the information theoretic measures, even when a high number of time series are considered or when few data samples are available for the estimation process. 2) performing a detection of the physiological state of a subject, by introducing a second step of analysis based on machine learning models. These will be trained with features derived directly from the information theoretic estimation that represent physiological interactions between different biological systems. This framework will be tested on 20 healthy subjects during a motor task execution considering as biological signals those recorded from 31 different sites of the brain and 12 different sites of the upper limb. In the future, the results of this project could have a strong impact on different contexts:
- Methodological: the project will produce a new framework for the information dynamics evaluation able to track the complex dynamics of different physiological systems at different time scales even when few data samples are available for the estimation process. We will identify consistent patterns of network activity and network structure quantified in term of information dynamics, which are peculiar of stable physiological state.
- Clinical: the ability of the new approach to estimate the information flow between different biological systems and exploit this information for the recognition of the physiological state, could be used in the evaluation of the engagement level of a patient with severe motor disability involved in a rehabilitation task, in order to evaluate his intention in performing the task.
- Neurophysiological: the developed methodological approach will allow to overcome some limits emerged the social science field from analysis considering biological signals independently each other. The possibility to identify the emotional state of a subject during a social task allows a greater understanding of the physiological mechanisms at the basis of complex social behaviours.