University of Patras
Somatosensory evoked potentials as a rehabilitation assessment tool for spinal cord injury patients, with the contribution of artificial intelligence
Abstract
dc:descriptionBackground: Spinal Cord Injury (SCI) is among the most severe neurological disorders, often resulting in irreversible motor and sensory deficits. Current clinical tools, such as the ASIA score, provide essential diagnostic insight but fall short in predicting individualized recovery outcomes with high accuracy and objectivity. There is a growing need for scalable, data-driven solutions to improve neuro-rehabilitation planning and patient management. Objective: This doctoral thesis investigates the prognostic value of Somatosensory Evoked Potentials (SSEPs) in SCI recovery. It aims to develop and validate artificial intelligence (AI) models for predicting outcomes, design a clinical decision support system, and evaluate the generalizability of this approach to Multiple Sclerosis (MS). The work is structured around three core pillars: a scoping review to map the existing evidence, a primary machine learning study on a multicenter cohort, and a meta-analysis to quantify the predictive performance of evoked potential-based AI models across neurological disorders. Methods: The research utilized multiple methodological approaches. First, a scoping review was conducted to synthesize existing literature. Then, using a high-quality retrospective dataset from the European Multicenter Study about Spinal Cord Injury (EMSCI), we analyzed 123 records after filtering 748 patients’ clinical and electrophysiological data. We applied supervised ML algorithms—including Decision Trees, Neural Networks, and ensemble models—to identify key predictors of SCI recovery, with a focus on the prognostic value of SSEPs compared to other inputs like MEPs, NCS, and sensory scores. Additionally, an external MS dataset was used to test the generalizability of the models in predicting Expanded Disability Status Scale (EDSS) progression. Lastly, a meta-analysis was performed to pool predictive accuracy data for EP-based ML models in both SCI and MS. Results: SSEPs emerged as the most predictive electrophysiological parameter, achieving up to 91.9% accuracy in forecasting motor recovery. A Decision Support System based solely on sensory electrophysiology reached 86% accuracy, proving to be a viable, cost-effective, and scalable alternative to conventional assessments. Feature importance analyses confirmed the dominant role of SSEPs in SCI outcome prediction. Moreover, the same AI methodology applied to MS data showed that SSEPs and multimodal evoked potentials hold predictive value in other neurological diseases. The meta-analysis confirmed the strong pooled predictive performance of EP-based ML models (accuracy: 79.2%, AUC: 0.82) across diseases. Conclusion: This thesis validates the integration of SSEP-based biomarkers with AI-driven analytics for SCI prognosis, highlighting their superiority in recovery prediction and clinical decision support. The successful extension of this framework to MS confirms the general utility of electrophysiological modeling in broader neurorehabilitation contexts. Finally, the study advocates for the adoption of digital health tools—including portable SSEPs and AI-based decision systems—into routine neurorehabilitation to facilitate real-time, patient-specific treatment optimization.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Chrysanthakopoulou, Dionysia
- Χρυσανθακοπούλου, Διονυσία
Subjects
dc:subject × 20- Κάκωση νωτιαίου μυελού
- Σωματοαισθητικά προκλητά δυναμικά
- Τεχνητή νοημοσύνη
- Ηλεκτροφυσιολογικοί Βιοδείκτες
- Νευροαποκατάσταση
- Spinal cord injury
- Somatosensory evoked potentials
- Artificial intelligence
- Electrophysiological Biomarkers
- Neurorehabilitation
- Επιστήμες Υγείας
- Ιατρική και Επιστήμες Υγείας
- Αποκατάσταση
- Βασική Ιατρική
- Νευροεπιστήμες
- Health Sciences
- Medical and Health Sciences
- Rehabilitation
- Basic Medicine
- Neurosciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/59848