JMIR Publications

ABSTRACT

Background:

Artificial Intelligence (AI) technologies for vision-based epilepsy monitoring are advancing rapidly in healthcare. Despite growing research utilizing various video data sources and analytical approaches, no comprehensive framework exists to classify these emerging technologies.

Objective:

The purpose of this study is to develop and validate a comprehensive taxonomy for categorizing AI technologies in vision-based epilepsy monitoring. We aim to provide fundamental insights into AI-based approaches in epilepsy care based on visual data processing. The proposed taxonomy specifies key properties of these technologies. This will contribute to a more robust knowledge base for further developing advanced healthcare solutions in epilepsy care.

Methods:

Guided by an extended taxonomy development framework, our taxonomy was developed within six iterative cycles based on data from theory and practice. We conducted a structured literature review, market analysis and validation using market-ready solutions. We reviewed articles published between January 6, 2025, and October 1, 2025 to identify and synthesize review articles and empirical studies focusing on AI technologies for epilepsy monitoring. Studies were included if they addressed AI- or machine-learning–based monitoring or prediction of epileptic seizures using visual or video data in human populations, while non-English publications, non-epilepsy research, studies solely on EEG or wearables, animal studies and pre-2012 publications were excluded. To ensure validity and practical relevance, 9 domain experts evaluated the taxonomy using a Delphi technique.

Results:

A total of 34 studies and 8 reviews were included. The review analysis revealed 9 dimensions, e.g., data acquisition source, tracking target, image processing, type of classifier and performance metrics, followed by the study analysis with 7 additional dimensions, e.g., environment, seizure classification, data privacy and user interface. The final taxonomy comprises 23 dimensions with 102 defining characteristics. Results show a predominance of detection and classification in stationary settings with limited predictive and real-time feedback. The study revealed a prevalence of deep-learning detection methods, accompanied by inconsistencies in performance reporting and limited patient functionalities. Additionally, privacy safeguards were found to be underreported. The taxonomy converts these patterns into actionable guidance for standardized benchmarking, procurement evaluation, and UI/XAI design to support translation to practice. In addition, the study identifies major limitations of current vision-based systems and synthesizes 6 main findings and 12 implications for future research and practice. Key challenges are highlighted in the areas of standardization, seizure prediction, and real-time applicability.

Conclusions:

By structuring the landscape and clarifying research gaps, this study offers a solid conceptual foundation for vision-based AI technologies for epilepsy monitoring and advances the field by linking application context to system architecture, visual analysis, AI model, performance reporting, and feedback design as a practical decision aid. Furthermore, the specified implications offer a practical outlook to facilitate further research.