Current project: Functional Connectivity, Information Fusion and Explainable AI for
Modelling Neural Synchronization in Dyslexia Using High-Performance and
Energy-Efficient Computing Algorithms  (DYSCONNECT)

Funded Research Project – PID2022-137461NB-C32. Plan Nacional de I+D+i. Proyectos de I+D de
Generación de Conocimiento. Ministerio de Ciencia, Innovación y
Universidades. 2023 – 2027

Abstract

Recent technological advances allow storing and  processing vast amounts of data of different nature. However, harnessing the information to transform it into knowledge is a challenging task. In the field of biomedical research and, specifically in neurosciences, new acquisition methods provided a way to obtain detailed structural or functional brain information. However, the identification of patterns associated with different pathologies is a manual and time-consuming task, which requires the use of new computer-based processing methods.
This is the case of learning disabilities such as developmental dyslexia (DD), a condition that hinders the learning of reading and spelling in 5%-12% of the worlds population and a major determinant of school failure. The use of behavioral or neuropsychological data is insufficient to accurately diagnose DD or investigate its cognitive or neurological underpinnings. This knowledge results essential for the early and objective diagnosis of DD, paving the way to reduce its psychosocial consequences.
The present project aims to continue the lines of research initiated with the previous project with two complementary objectives: 1) The development of exploratory techniques that allow us to delve into the biological origin of this brain disorder and 2) to transform the exploratory information nto knowledge for the clinical use. In this case, the aim is to provide methods for an early and objective identification of DD allowing the development of individualized intervention tasks.
Specifically, we propose the development of methods and algorithms for functional connectivity estimation by means of local and long-range neural synchronization. Moreover, this connectivity information will be used for constructing network models to explain conditional,
unconditional or causal interactions between brain regions. To do so, we will employ graph theory to infer the network that processes different phonological units. Indeed, complex network analysis allows computing descriptors regarding their efficiency, in the search of
differential patterns between controls and DD subjects. Connectivity information can be fused to Heart Rate Variability (HRV) data and the results of behavioural tests to identify implications in reading performance. Additionally, the analysis of network dynamics would reveal differential patterns in neural adaptation.
These objectives require the combination of signal processing and explainable AI techniques that allow the extraction of knowledge with clinical implications. Moreover, the computational burden associated to the tasks proposed here requires the use of HPC architectures and the parallelization of the algorithms, striking a balance between performance and energetic efficiency.
EEG and fNIRS signals were acquired during a semi-resting state experiment, using a specific experimental procedure developed by our research team during our previous projects PSI2015-65848-R, PGC2018-098813-B-C32. Subjects for EEG/fNIRS acquisitions were selected from the leeduca database (https://leeduca.es), which currently contains behavioural results from more than 7000 subjects (4-7 years old). The multidisciplinary research team (consolidated in the leeduca project) and their synergistic interaction will enable the transference of the results, representing a game-change in the treatment of learning disabilities in children with the associated socioeconomic impact.

Main results

Current Publications (Indexed Journals)

1. Castillo-Barnes, D., Gallego-Molina, N.J., Formoso, M.A., Ortiz, A., Figueiredo, P., Luque, J.L., «Probabilistic and explainable modeling of Phase–Phase Cross-Frequency Coupling patterns in EEG. Application to dyslexia diagnosis,» *Biocybernetics and Biomedical Engineering*, 2024.

2. Gallego-Molina, N.J., Ortiz, A., Arco, J.E., Martínez-Murcia, F.J., Lok, W.W., «Unraveling brain synchronisation dynamics by explainable neural networks using EEG signals. Application to dyslexia diagnosis,» *Interdisciplinary Sciences: Computational Life Sciences*, 2024.

3. Arco, J.E., Gallego-Molina, N.J., Ortiz, A., Arroyo-Alvis, K., López-Pérez, P.J., «Identifying HRV patterns in ECG signals as early markers of dementia,» *Expert Systems with Applications*, Vol. 240(1), 2024, pp. 1-14.

4. Rodríguez-Rodríguez, I., Ortiz, A., Gallego-Molina, N.J., Formoso, M.A., Woo, L.W., «EEG Interchannel Causality to Identify Source/Sink Phase Connectivity Patterns in Developmental Dyslexia,» *International Journal of Neural Systems*, Vol. 33(4), 2023, pp. 1-18.

5. Gallego-Molina, N.J., Ortiz, A., Martínez-Murcia, F.J., Rodríguez-Rodríguez, I., Luque, J.L., «Assessing Functional Brain Network Dynamics in Dyslexia from fNIRS Data,» *International Journal of Neural Systems*, Vol. 33(4), 2023, pp. 1-19.

6. Escobar, J.J., Rodríguez, F., Prieto, B., Kimovski, D., Ortiz, A., Damas, M., «A distributed and energy-efficient KNN for EEG classification with dynamic money-saving policy in heterogeneous clusters,» *Computing*, 2023, pp. 1-24. doi.org/10.1007/s00607-023-01193-7

International Conferences