Thesis Project Form

Alzheimer’s Disease (AD) and Mild Cognitive Impairment (MCI) are traditionally diagnosed and monitored through cognitive assessments. However, literature highlights that motor function deficits, such as impairments in fine motor control, coordination and goal-directed reaching movements, often precede severe cognitive decline. Complex motor tasks require precise sensorimotor integration and motor planning, which are frequently compromised early in neurodegenerative diseases. To effectively capture these impairments, immersive Virtual Reality (VR) environments have emerged as a highly engaging, ecologically valid platform, overcoming the limitations of traditional laboratory settings.
This thesis is motivated by the need to explore the electromyographic (EMG) dimension of these complex motor behaviors. By applying advanced data-driven modeling, specifically muscle synergy extraction and Recurrent Neural Networks (RNNs) within a Reservoir Computing framework, this project aims to decode the hidden neural dynamics of motor control.
Uncovering these neuromuscular biomarkers holds the potential to significantly enhance early diagnosis protocols, accurately track disease progression and provide objective metrics to evaluate the efficacy of novel therapeutic or rehabilitative interventions.

The study investigates impairments in sensorimotor coordination and control in patients with Alzheimer's Disease (AD), according to the evidence that motor function deficits are early indicators of AD and Mild Cognitive Impairment (MCI) alongside cognitive decline.
The experimental framework employs immersive Virtual Reality (VR) to simulate catching tasks, with the student granted access to the laboratories of the IRCCS Fondazione Santa Lucia. While data acquisition and curation will encompass both kinematic and electromyographic (EMG) data from patient cohorts and healthy controls, the investigation prioritizes the muscular component. The aim of the analysis will be to fill a specific gap in the current
research framework, exploring the EMG dimension of the dataset.
To achieve a comprehensive understanding of motor control, the data analysis will be approached using two complementary methods. Initially, factorization algorithms will be applied to extract muscle synergies, determining how motor coordination and control dimensionality vary across different stages of disease progression. Subsequently, the use of Recurrent Neural Networks (RNNs) within the Reservoir Computing (RC) framework will be explored to model the experimental data. This computational approach aims to infer patient-specific trajectories associated with changes in muscle activation patterns, allowing to position in a low-dimensional latent space (namely, the subject landscape) the internal dynamics of
the motor control system for each subject and experimental session.
The goal of this research is to identify biomarkers capable of quantifying the patient's sensorimotor state and potentially tracking their response to therapeutic interventions.

On the experimental side, the project involves gaining hands-on experience in managing data acquisition sessions with AD patients. Regarding data analysis, the focus will be on the preprocessing and feature extraction of EMG and kinematic signals. Furthermore, the work requires the use of MATLAB and the Synergy Analyzer toolbox to extract and interpret muscle synergies and assess the modularity of motor control. Finally, computational modeling will be explored by applying Machine Learning paradigms, specifically Recurrent Neural Networks (RNNs) and the Reservoir Computing (RC) framework, to model complex motor dynamics and extract patient-specific latent trajectories.