Background: Migraine is a highly prevalent neurological disorder affecting over one billion people worldwide, yet reliable objective biomarkers remain elusive. High-density electroencephalography (HD-EEG) functional connectivity analysis offers a non-invasive approach to investigate brain network alterations in migraine patients and evaluate treatment response. Objective: This thesis investigated functional connectivity differences in high-frequency episodic migraine patients before and after atogepant treatment, and compared patients with healthy controls, using both conventional statistical methods and interpretable machine learning approaches to identify potential migraine biomarkers. Methods: HD-EEG data (128 channels) were acquired from 37 migraine patients at baseline (T0) and after three months of treatment (T1). Functional connectivity was computed across six brain networks (Default Mode Network, Dorsal Attention Network, Ventral Attention Network, Language Network, Somatomotor Network, and Visual Network) in five frequency bands (delta, theta, alpha, beta, gamma). Statistical analysis employed paired and independent sample tests with False Discovery Rate correction across multiple recording sessions to assess test-retest reliability. Four supervised machine learning algorithms (Random Forest, Logistic Regression, Support Vector Machine, and Naive Bayes) were applied to three binary classification tasks, with model interpretability assessed using SHAP values. Results: Statistical analysis revealed no significant differences in network connectivity between pre-treatment and post-treatment patients across all analyses. However, consistent alterations in intra-network connectivity within the Ventral Attention Network (VAN) in the alpha frequency band distinguished pre-treatment patients from healthy controls across all test-retest iterations. Machine learning classification achieved high accuracy (>90%) in differentiating both pre- and posttreatment patients from controls, with Logistic Regression demonstrating optimal performance. In contrast, T0 versus T1 classification yielded poor results (accuracy 52%), consistent with statistical findings. Feature importance analysis identified connectivity between the Default Mode Network and Vision Network (DMN-VN) as a key discriminative feature between high frequency episodic migraine patients and controls.
Background: L’emicrania è un disturbo neurologico ad alta prevalenza, che colpisce oltre un miliardo di persone in tutto il mondo. Tuttavia, non esistono biomarcatori oggettivi e affidabili. L’analisi della connettività funzionale mediante elettroencefalografia ad alta densità (HD-EEG) rappresenta un approccio non invasivo per indagare le alterazioni delle reti cerebrali nei pazienti emicranici e valutarne la risposta al trattamento. Obiettivo: Questa tesi ha indagato le differenze nella connettività funzionale in pazienti con emicrania episodica ad alta frequenza, prima e dopo il trattamento con Atogepant. Inoltre, sono stati confrontati i pazienti con soggetti di controllo sani, utilizzando sia metodi statistici convenzionali sia approcci interpretabili di machine learning, al fine di identificare potenziali biomarcatori associati al sottotipo di emicrania. Metodi: I dati HD-EEG (128 canali) sono stati acquisiti da 37 pazienti emicranici al basale (T0) e dopo tre mesi di trattamento (T1). La connettività funzionale è stata calcolata in sei reti cerebrali (Default Mode Network, Dorsal Attention Network, Ventral Attention Network, Language Network, Somatomotor Network e Visual Network) e in cinque bande di frequenza (delta, theta, alfa, beta, gamma). L’analisi statistica ha impiegato test per campioni appaiati e indipendenti con correzione FDR, valutando anche l’affidabilità test-retest. Sono stati applicati quattro algoritmi di machine learning supervisionato (Random Forest, Regressione Logistica, Support Vector Machine e Naive Bayes) a tre compiti di classificazione binaria. L’interpretabilità dei modelli è stata valutata tramite SHAP. Risultati: L’analisi statistica non ha evidenziato differenze significative nella connettività di rete tra pazienti pre- e post-trattamento in nessuna delle analisi condotte. Tuttavia, alterazioni consistenti della connettività intra-rete all’interno della Ventral Attention Network (VAN) nella banda di frequenza alfa hanno distinto i pazienti pre-trattamento dai controlli sani in tutte le iterazioni test-retest. La classificazione con machine learning ha raggiunto un’accuratezza di circa il 90% nel differenziare sia i pazienti pre- che post-trattamento dai controlli, mentre la classificazione T0 vs T1 ha prodotto risultati scarsi (accuratezza 52%), coerenti con i risultati statistici. L’analisi con SHAP ha identificato la connettività tra la Default Mode Network e la Visual Network (DMN-VN) come caratteristica discriminante chiave tra i pazienti con emicrania episodica ad alta frequenza e i controlli.
INDAGINE ESPLORATIVA DEI BIOMARCATORI DI RISPOSTA AL TRATTAMENTO IN PAZIENTI CON EMICRANIA EPISODICA AD ALTA FREQUENZA TRATTATI CON ATOGEPANT ATTRAVRESO L’ANALISI DELLA CONNETTIVITÀ FUNZIONALE DA EEG AD ALTA DENSITÀ
FERRERI, LUDOVICO
2024/2025
Abstract
Background: Migraine is a highly prevalent neurological disorder affecting over one billion people worldwide, yet reliable objective biomarkers remain elusive. High-density electroencephalography (HD-EEG) functional connectivity analysis offers a non-invasive approach to investigate brain network alterations in migraine patients and evaluate treatment response. Objective: This thesis investigated functional connectivity differences in high-frequency episodic migraine patients before and after atogepant treatment, and compared patients with healthy controls, using both conventional statistical methods and interpretable machine learning approaches to identify potential migraine biomarkers. Methods: HD-EEG data (128 channels) were acquired from 37 migraine patients at baseline (T0) and after three months of treatment (T1). Functional connectivity was computed across six brain networks (Default Mode Network, Dorsal Attention Network, Ventral Attention Network, Language Network, Somatomotor Network, and Visual Network) in five frequency bands (delta, theta, alpha, beta, gamma). Statistical analysis employed paired and independent sample tests with False Discovery Rate correction across multiple recording sessions to assess test-retest reliability. Four supervised machine learning algorithms (Random Forest, Logistic Regression, Support Vector Machine, and Naive Bayes) were applied to three binary classification tasks, with model interpretability assessed using SHAP values. Results: Statistical analysis revealed no significant differences in network connectivity between pre-treatment and post-treatment patients across all analyses. However, consistent alterations in intra-network connectivity within the Ventral Attention Network (VAN) in the alpha frequency band distinguished pre-treatment patients from healthy controls across all test-retest iterations. Machine learning classification achieved high accuracy (>90%) in differentiating both pre- and posttreatment patients from controls, with Logistic Regression demonstrating optimal performance. In contrast, T0 versus T1 classification yielded poor results (accuracy 52%), consistent with statistical findings. Feature importance analysis identified connectivity between the Default Mode Network and Vision Network (DMN-VN) as a key discriminative feature between high frequency episodic migraine patients and controls.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_Ferreri_nuovo_pdfA.pdf
accesso aperto
Dimensione
12.56 MB
Formato
Adobe PDF
|
12.56 MB | Adobe PDF | Visualizza/Apri |
È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/33653