Strategia di selezione del modello nel dynamic causal modeling per la risonanza magnetica funzionale: applicazione a un compito di osservazione dell'azione

Magnetic Resonance Imaging (MRI) techniques enable the investigation of the human brain under multiple perspectives, ranging from its anatomy (e.g., the morphologic characterisation of different brain regions), microstructure (e.g., the anatomical connections between different regions) and its functions (e.g., the functional connections between different regions). Functional MRI (fMRI) measures blood flow as an indirect marker of neural activity. During the performance of a task, neural activity increases, leading to a higher consumption of oxygen. This is followed by a vascular response that overcompensates the oxygen demand, resulting in changes in the balance between oxyhemoglobin and deoxyhemoglobin, which in turn affects the BOLD signal. Therefore, changes in blood flow reflect the underlying metabolic demand rather than neural activity directly. The correlation between the BOLD signals of pairs of regions defines functional connectivity, i.e., the interaction between distributed brain regions, which is essential to understand how functional brain networks operate. Despite functional connectivity being a fundamental feature to understand brain dynamics, it lacks causal information. Indeed, functional connectivity measures the correlation between regional signals without considering the mechanisms that drive these interactions. Causal interactions in a brain network can be measured through effective connectivity. Dynamic Causal Modelling (DCM) is an established framework that models causal interactions between brain regions, estimating both the direction and the influence of these interactions using a generative model. The aim of this thesis was to examine effective connectivity within a visuomotor network, defined as a network integrating visual processing and motor planning, including the Visual Cortex, Primary Motor Cortex, Supplementary Motor Area/Premotor Cortex, Cingulate Cortex, Superior Parietal Lobule and the Cerebellum. In particular, two experimental conditions were compared: Action Observation (AO), where subjects observe an action, and Action Observation with visual cue (AO-BAR), where observation is combined with an additional visual cue. This additional visual information is used to guide the task. The analysis was performed using two fMRI datasets specific for AO and AO-BAR. The pipeline included preprocessing steps, followed by the extraction of regions embedded in the visuomotor network and the estimation of representative time series. Connectivity modelling was then performed using DCM, and different models of network organisation, including both fully connected and hypothesis-driven reduced models informed by prior literature (e.g. Lorenzi et al., 2025), were compared at both subject and group levels. The results showed that the AO condition exhibited a stable and well-supported effective connectivity pattern, with multiple connections such as from the Visual Cortex to motor regions and to the Cerebellum, reflecting the integration between visual processing, motor planning and coordination within the visuomotor network. In contrast, the AO-BAR condition did not show strongly supported connections and presented greater variability across subjects. Model 4, representing a distributed architecture with interactions between visual, motor and cerebellar regions, was identified as the most plausible architecture for the AO dataset. These findings suggest that the visuomotor network is coherently engaged during action observation, whereas the addition of a visual cue does not improve connectivity and may instead increase cognitive demand, reducing the stability of the network. Overall, the results highlight how connectivity strongly depends on task design, and how even small experimental changes can significantly affect brain network organisation.

Le tecniche di Risonanza Magnetica (MRI) consentono di studiare il cervello umano da diverse prospettive, tra cui anatomia, microstruttura e funzione. La risonanza magnetica funzionale (fMRI) misura il flusso sanguigno come marcatore indiretto dell’attività neurale, attraverso variazioni del segnale BOLD legate alla risposta vascolare alla richiesta di ossigeno. La correlazione tra segnali BOLD definisce la connettività funzionale, che descrive l’interazione tra regioni cerebrali, ma non fornisce informazioni causali. Tali interazioni possono essere studiate tramite la connettività effettiva. Il Dynamic Causal Modelling (DCM) è un framework che modella le interazioni causali tra regioni cerebrali, stimandone direzione e influenza. L’obiettivo di questa tesi è stato analizzare la connettività effettiva in una rete visuomotoria, comprendente Corteccia Visiva, Corteccia Motoria Primaria, Area Supplementare Motoria/Corteccia Premotoria, Corteccia Cingolata, Lobulo Parietale Superiore e Cervelletto. Sono state confrontate due condizioni: Action Observation (AO) e Action Observation con stimolo visivo aggiuntivo (AO-BAR). L’analisi è stata condotta su dataset fMRI utilizzando MATLAB e SPM12. La pipeline ha incluso preprocessing, estrazione delle regioni e delle serie temporali, e modellazione DCM con confronto tra modelli completamente connessi e modelli ridotti. I risultati mostrano che la condizione AO presenta una connettività effettiva stabile e ben supportata, con integrazione tra regioni visive, motorie e cerebellari. Al contrario, AO-BAR non mostra connessioni fortemente supportate e presenta maggiore variabilità. Il Modello 4 è risultato il più plausibile per AO. Questi risultati suggeriscono che la rete visuomotoria è attivata in modo coerente durante l’osservazione di un’azione, mentre l’aggiunta di un segnale visivo può aumentare il carico cognitivo e ridurre la stabilità della rete. Nel complesso, evidenziano come la connettività dipenda dal design del compito e come anche piccole modifiche possano influenzare l’organizzazione delle reti cerebrali.