Biomarcadores predictivos de la eficacia de la estimulación del nervio vago en epilepsia refractaria mediante el análisis longitudinal de conectividad funcional en electroencefalograma.
Loading...
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
La estimulación del nervio vago (VNS, por sus siglas en inglés) es un tratamiento paliativo utilizado en pacientes con epilepsia fármaco resistente (DRE, en inglés); sin embargo, su eficacia clínica es marcadamente heterogénea y carece de biomarcadores predictivos validados. Esta tesis propone un marco metodológico híbrido que integra el análisis de conectividad funcional, la teoría de grafos y el aprendizaje supervisado explicable para predecir y caracterizar la respuesta terapéutica.
A partir de un estudio longitudinal de señales de electroencefalograma (EEG) en 12 pacientes (5 Respondedores y 7 No-Respondedores), capturadas en cuatro puntos temporales (pre-VNS y a los 1, 3 y 6 meses post-VNS), se implementaron métricas de sincronización de fase mediante el índice de desfase ponderado (wPLI) e interacciones no lineales mediante la información mutua (MI). El análisis topológico se realizó mediante métricas de eficiencia global (GE) y modularidad (Q), procesadas con algoritmos de clasificación (SVM, regresión logística y random forest) optimizados mediante RFE.
Los resultados demuestran que la organización topológica previa a la implantación del VNS en la banda alfa constituye un biomarcador predictivo robusto, alcanzando una exactitud del 83% y un AUROC de 0.83. Se identificó que los pacientes Respondedores exhiben un ‘Health Status’ prequirúrgico caracterizado por una alta segregación modular y una baja integración global en alfa, lo que sugiere una arquitectura con mayor elasticidad funcional para la terapia de neuromodulación.
En la dimensión temporal, el análisis mediante modelos lineales mixtos (LMMs) y clasificadores de trayectorias reveló una evolución bifásica de la red en los respondedores: una fase aguda de integración en alfa al primer mes, seguida de una estabilización crónica en theta a los seis meses. En contraste, los no-respondedores presentaron una tendencia al alza hacia la hipersincronía patológica, evidenciada por un aumento continuo de la eficiencia global en la banda theta.
Esta investigación contribuye con un análisis exploratorio al sugerir que la respuesta al VNS no es un evento estático, sino una reorganización neuroplástica cuantificable. Los hallazgos proporcionan evidencia hacia una herramienta de soporte a la decisión médica con el potencial de optimizar la selección de candidatos y personalizar el manejo clínico de la epilepsia refractaria.
Vagus Nerve Stimulation (VNS) is a palliative treatment used for patients with Drug-Resistant Epilepsy (DRE); however, its clinical efficacy is markedly heterogeneous and lacks validated predictive biomarkers. This thesis proposes a hybrid methodological framework that integrates functional connectivity analysis, graph theory, and explainable supervised learning to predict and characterize therapeutic response. Based on a longitudinal electroencephalogram (EEG) study of 12 patients (5 Responders and 7 Non-Responders) captured at four time points (pre-VNS and at 1, 3, and 6 months post-VNS), phase synchronization metrics were implemented using the weighted Phase Lag Index (wPLI) and non-linear interactions via Mutual Information (MI). Topological analysis was performed using Global Efficiency (GE) and Modularity (Q) metrics, processed with classification algorithms (SVM, logistic regression, and random forest) optimized through Recursive Feature Elimination (RFE). The results demonstrate that pre-VNS topological organization in the alpha band constitutes a robust predictive biomarker, achieving an accuracy of 83% and an AUROC of 0.83. It was identified that responders patients exhibit a pre-surgical "Health Status" characterized by high modular segregation and low global integration in the alpha band, suggesting an architecture with greater functional elasticity for neuromodulation. Regarding the temporal dimension, analysis using Linear Mixed Models (LMMs) and trajectory classifiers revealed a biphasic network evolution in responders: an acute integration phase in alpha at one month, followed by chronic stabilization in theta at six months. In contrast, non-responders patients presented a drift toward pathological hypersynchrony, evidenced by a continuous increase in global efficiency. This research contributes an exploratory analysis by suggesting that VNS response is not a static event but a quantifiable neuroplastic reorganization. The findings provide evidence towards implementation of a medical decision support tool with the potential to optimize candidate selection and personalize the clinical management of refractory epilepsy.
Vagus Nerve Stimulation (VNS) is a palliative treatment used for patients with Drug-Resistant Epilepsy (DRE); however, its clinical efficacy is markedly heterogeneous and lacks validated predictive biomarkers. This thesis proposes a hybrid methodological framework that integrates functional connectivity analysis, graph theory, and explainable supervised learning to predict and characterize therapeutic response. Based on a longitudinal electroencephalogram (EEG) study of 12 patients (5 Responders and 7 Non-Responders) captured at four time points (pre-VNS and at 1, 3, and 6 months post-VNS), phase synchronization metrics were implemented using the weighted Phase Lag Index (wPLI) and non-linear interactions via Mutual Information (MI). Topological analysis was performed using Global Efficiency (GE) and Modularity (Q) metrics, processed with classification algorithms (SVM, logistic regression, and random forest) optimized through Recursive Feature Elimination (RFE). The results demonstrate that pre-VNS topological organization in the alpha band constitutes a robust predictive biomarker, achieving an accuracy of 83% and an AUROC of 0.83. It was identified that responders patients exhibit a pre-surgical "Health Status" characterized by high modular segregation and low global integration in the alpha band, suggesting an architecture with greater functional elasticity for neuromodulation. Regarding the temporal dimension, analysis using Linear Mixed Models (LMMs) and trajectory classifiers revealed a biphasic network evolution in responders: an acute integration phase in alpha at one month, followed by chronic stabilization in theta at six months. In contrast, non-responders patients presented a drift toward pathological hypersynchrony, evidenced by a continuous increase in global efficiency. This research contributes an exploratory analysis by suggesting that VNS response is not a static event but a quantifiable neuroplastic reorganization. The findings provide evidence towards implementation of a medical decision support tool with the potential to optimize candidate selection and personalize the clinical management of refractory epilepsy.
Description
Tesis presentada para optar al grado de Magíster en Ciencias de la Ingeniería con mención en Ingeniería Eléctrica.
Keywords
Nervio Vago, Epilepsia, Electrofisiología