Titulo Estágio
Connecting the Dots in the Brain: Conciliating Functional Connectivity with Distinct Epiphenomena via Artificial Neural Networks
Áreas de especialidade
Sistemas Inteligentes
Sistemas de Informação
Local do Estágio
DEI-FCTUC
Enquadramento
The Brain remains a black-box: The underlying dynamical laws governing the Brain’s activity lie still elusive. This renders several Brain related inference problems blatantly hard. Of particular interest is the question: How are distinct regions of the Brain communicating? What is the underlying network of causal relationships linking these regions? If this is enticing on its own right, recent evidences show that this causal structure (or certain proxys thereof) entails relevant information about several epiphenomena, e.g., cognitive disorders or motor activities. How can we unveil this interaction profile? It can only be inferred from the activity itself, i.e., from the time-series reflecting the state-evolution over time of the distinct regions of the Brain (observed as FMRI or EEG signals). But how to translate the observed time-series into structural information revealing the causal interactions?
The Brain is a complex networked dynamical system, possibly spatially hybrid in that distinct regions may abide by distinct dynamical laws. To reverse-engineer the network or other built-in constructs of a complex networked system from its time-series, knowledge of its underlying nature, i.e., of the particular (dynamical) law is critical. Alike comprehending distinct languages (German, French, Portuguese, etc.), we need a French translator to leverage faithful information entailed in a French-written sentence. In our context, the same applies: i) If the observed time-series obey to a Gaussian multivariate distribution, the inverse of the correlation matrix is the right translator (i.e., a consistent estimator) for the network structure; ii) if the time-series reflect the state evolution of a linear networked dynamical system, then regression (or Granger) may be appropriate; iii) if it abides by an Ising model distribution, then the correlation matrix is a consistent one; iv) if it emerges from other (dynamical) laws, distinct estimators need to be on demand. In other words, the (structural) consistency of the estimator/translator ties closely with the underlying nature of the observed time-series, i.e., with the particular language uttered by the Brain. Distinct from cosmology, these dynamical laws lie concealed within the Brain itself.
Functional connectivity (FC) stands for the network of interactions among active regions of the Brain in a particular moment in time [1]. It entails the main avenues of information flow across the distinct regions. Due to the aforementioned reasons, the actual FC conspicuously lurks underneath the Brain’s activity with no (provably consistent) estimator to transparently translate it out. There are, however, proposed proxys based upon correlation or other estimators that seem to convey nontrivial structural signatures about several epiphenomena.
[1] Raphaël Liégeois, Augusto Santos, Vincenzo Matta, Dimitri Van De Ville, Ali H. Sayed, "Revisiting Correlation-based Functional Connectivity and its Relationship with Structural Connectivity", Network Neuroscience, December 2020.
Objetivo
The goal of this internship is to tighten the chasm of our understanding about the actual Functional Connectivity -- hence, further tapping on the fundamental laws of the Brain, -- via resorting to Artificial Neural Networks. We further (co)relate the structural attributes of the estimated FC with various cognitive disorders (Alzheimer, Seizures, etc.) and motor activities. Next, we state more concretely the planned path to pursue our goals.
1) [Inference of the Functional Connectivity]. Given the time-series (either from FMRI or EEG signals), we devise Deep Learning based tools to infer the underlying FC as consistently as possible. In this regard, we plan to proceed as follows: we train two Convolutional Neural Networks (CNNs), one for the classification of the underlying dynamical law in course and another for the structural inference of the functional connectivity conditioned on the dynamical law obtained by the first CNN. Ideally, the training data should be a properly pre-processed version of the actual time-series. In this module, the student will enjoy great expertise of its advisors with regards to structural inference of networked dynamical systems (via Artificial Neural Networks).
2) [Relating the FC with distinct Epiphenomena]. Once the matrix estimate of the Functional Connectivity has been obtained in module 1, the goal is to conciliate distinct structural attributes appertaining to this matrix with several forms of cognitive disorders or motor activities. For example, how is the spectra of the estimated FC matrix related with the underlying observed disorder? Does a larger spectral gap of the FC foster a particular disorder? These and other structural variables will be exploited, and the major goal is to ascertain a vector of informative structural attributes as a consistent biomarker to the phenomena studied.
We resort to data available online across distinct platforms, e.g.,
http://preprocessed-connectomes-project.org/abide/
CRCNS.org
https://search.kg.ebrains.eu/ (Brain Project)
https://www.fil.ion.ucl.ac.uk/spm/data/
https://www.oasis-brains.org/
In this internship, the student will be exposed to four hot topics/fields of current research: Artificial Intelligence, Deep learning, Brain Functional Connectivity (Neuroscience), Model Identification (especially, causal inference).
Plano de Trabalhos - Semestre 1
[Some tasks might overlap; M=Month]
T1 (M1-M2): Knowledge transfer and State of the art analysis: Study the concepts underneath Functional Connectivity, network identification of dynamical systems, Deep Learning tools applied to structure identification, etc.
T2 (M3): Data collection and mining.
T3 (M4-M5): Devise methods for network identification (or causal inference) via deep learning tools.
T4 (M6): Writing the Intermediate Report.
Plano de Trabalhos - Semestre 2
[Some tasks might overlap; M=Month]
T5 (M7-M8): Implementation of the proposed methods for network structure identification (via deep learning tools) and test them across synthetic/real datasets.
T6 (M9-M10): Relate the resulting Functional Connectivity attributes with the distinct disorders or activities considered in the underlying dataset.
T7 (M11): Writing the thesis.
Condições
Note: there is a possibility of having a 6-month scholarship.
Type of scholarship: Bolsa de Investigação para Licenciado
Value of the scholarship: 752,38€
Observações
Please contact the advisors for any questions or clarifications needed.
Advisors: Augusto Santos (augustosantos@dei.uc.pt); Paulo Gil (pgil@dei.uc.pt); Jorge Henriques (jh@dei.uc.pt)
Orientador
Augusto Santos / Paulo Gil / Jorge Henriques
augustosantos@dei.uc.pt 📩