Fatigue is a debilitating symptom of multiple sclerosis (MS), affecting over 80% of patients and significantly impairing their quality of life. Despite its prevalence, the neural mechanisms underlying fatigue in MS remain poorly understood. Functional magnetic resonance imaging (fMRI) offers a powerful tool to investigate the brain’s intrinsic connectivity and dynamic network behavior. This project aims to analyze fMRI time-series data to uncover differences in functional connectivity and network dynamics among MS patients with fatigue, MS patients without fatigue, and healthy controls. By identifying distinct connectivity patterns and their relationships with structural MRI findings, this study seeks to advance our understanding of MS-related fatigue and identify potential neuroimaging biomarkers.

Goals for the BrainHack:

• Primary Objective: To investigate differences in functional connectivity and network dynamics among:
• MS patients with fatigue
• MS patients without fatigue
• Healthy controls
• Secondary Objectives:
• To explore correlations between fMRI connectivity metrics and clinical measures of fatigue severity
• To integrate structural MRI findings (e.g., lesion load, gray matter atrophy) with functional connectivity results to provide a comprehensive view of the neural underpinnings of fatigue in MS

Contributors' skills:

• Python codind
• fMRI
• MRI segmentation

In recent years, multimodal systems based on large language models (LLMs) have been successfully applied to encode and decode information from neurophysiological data. Most applications have focused on tracing and reconstructing images and language in the brain, using pictures, textual reports, and EEG data. While previous work has unveiled plausible spatial neural correlates of dream events, its time aspect remains elusive, In this project, we aim to trace dream events in EEG signals, using pre-trained LLMs and standard interpretability tools from the NLP and AI literature.

Goals for the BrainHack:

• Develop a framework to pre-process EEG data having different formats for the same LLM
• Set up a training regime suitable to the available data
• Train and monitor a first set of models or toy examples
• Prepare analysis pipeline
• Analyse a first set of models

Contributors' skills:

• Basic Python for text mining and analysis
• Python scientific, neuroscience, or ML (e,g., numpy, scikilearn, seabonr, mne, pytorch, hugging face)
• EEG/neuroscience data analysis knowledge
• ML/AI knowledge (even basic)

Arousal, defined as the behavioral state of alertness, is a potential driver of variability in brain activity and functional connectivity, particularly within networks such as the Default Mode Network and the Salience Network. In this project, we aim to investigate the effects of arousal on brain connectivity using a recently published open-access dataset that integrates pre-processed pupillometry, fMRI, and EEG data, enabling a comprehensive multimodal analysis. By combining pupillometry (as a proxy to identify high and low arousal levels), EEG (offering high temporal resolution of brain activity), and fMRI (providing high spatial resolution of functional activity), we can investigate the relationships between arousal states and both functional and effective connectivity across key brain regions.

Goals for the BrainHack:

• Identify arousal-related brain regions through the analysis of the relationships between pupil dynamics and fMRI signals at both voxel and region levels (e.g., via correlation or regression analyses on resting-state data)
• Compare functional and effective connectivity within arousal-related networks across high and low arousal states, using indices derived from both fMRI and EEG resting-state connectivity

As a secondary objective, we aim to replicate these analyses on a visual task dataset to compare the impact of arousal under resting-state and task-based conditions.

Contributors' skills:

• MATLAB programming (basic)
• Python programming (basic)
• Basic fMRI/EEG/pupillometry analysis knowledge

Recent studies have explored the potential of rhythmic pattern extraction for the development of human-computer interfaces able to provide visual or audio feedback - potentially in real-time - with various goals, such as assessing the effectiveness of performing specific motor tasks, facilitating learning and skill acquisition, and enhancing the reproducibility of technical gestures. A recent example is the TickTacking interface from Rocchesso et al. (2023), where a trajectory is drawn on a screen based on rhythms extracted by the bimanual tapping of two buttons.
This project aims to explore the possibility of developing an interface similar to TickTacking in which a piano keyboard replaces the two buttons, and the user is either a beginner or an expert pianist. Onset detection (i.e., the detection of the starting time of each note) is of fundamental importance to achieve this goal. Indeed, its analysis allows one to extract the inter-onset interval provided as an input to the interface. Hence, this project is focused on evaluating the feasibility of the application of various algorithms for onsets and tempo detection in the case of publicly available recordings and live performances obtained using an electronic piano at the disposal of the researchers. The project itself is part of the broader PRIN 2022 project MAHATMA (Multiscale Analysis of Human and Artificially Generated Trajectories, Models and Applications).
Possible developments of this research are:

• classifying automatically audio excerpts executed by beginners or experts
• evaluating the effectiveness of the video/audio feedback generated by the extended interface in the context of a learning process in which a beginner pianist follows a rhythmic (or polyrhythmic) pattern generated by an expert
• generating artificial rhythmic patterns imitating expert behavior using generative adversarial learning techniques

Goals for the BrainHack:

• Evaluating the feasibility of the application of various algorithms for onsets and tempo detection (applied, e.g., to the first prelude of Bach's Well-Tempered Clavier and the first movement of Beethoven's Moonlight Sonata) in the case of publicly available recordings and live performances obtained using an electronic piano at the disposal of the researchers
• Extracting various inter-onset intervals from the available data and performing their statistical analysis

Contributors' skills:

• MATLAB programming
• Python programming
• ML and statistical analysis
• Sound processing (basic)