Embedding computational neurorehabilitation in clinical practice using a modular intelligent health system

TL;DR

This paper tackles the fragmentation of data in neurorehabilitation and the need for real-time, patient-specific optimization via computational neurorehabilitation. It proposes an embedded i-health system with a three-layer architecture—Data Collection, Computational, and Clinical Ingestion—that interoperates with home and clinical data streams and supports real-time decision support. The demonstrator in a stroke rehabilitation clinic shows successful data capture (>1000 patient-days), integrated dashboards, and a pathway to trajectory, causal, and subpopulation analyses using tools like Ehrapy and Grafana, while highlighting usability and organizational adoption challenges. The work provides a practical blueprint for translating compNR from theory to clinical practice, with potential improvements in efficiency, personalization, and outcome prediction.

Abstract

A significant and rising proportion of the global population suffer from non-communicable diseases, such as neurological disorders. Neurorehabilitation aims to restore function and independence of neurological patients through providing interdisciplinary therapeutic interventions. Computational neurorehabilitation, an automated simulation approach to dynamically optimize treatment effectivity, is a promising tool to ensure that each patient has the best therapy for their current status. However, computational neurorehabilitation relies on integrated data flows between clinical assessments, predictive models, and healthcare professionals. Current neurorehabilitation practice is limited by low levels of digitalization and low data interoperability. We here propose and demonstrate an embedded intelligent health system that enables detailed digital data collection in a modular fashion, real-time data flows between patients, models, and clinicians, clinical integration, and multi-context capacities, as required for computational neurorehabilitation approaches. We give an outlook on how modern exploratory data analysis tools can be integrated to facilitate model development and knowledge inference from secondary use of observational data this system collects. With this blueprint, we contribute towards the development of integrated computational neurorehabilitation workflows for clinical practice.

Figures (3)

• Figure 1: Schematic overview of the proposed i-health system depicting the three-layer model. (a) Data Collection shows individual examples for clinical and home data sources to capture continuous health information. (b) Computation shows a schematic representation of computational neurorehabilitation from data to prediction of the optimal intervention. (c) Clinical Ingestion depicts data visualization modes and interoperability interfaces to clinical information systems and data analysis tools.
• Figure 2: Example visualization of an integrated dashboard embedded via https in a clinical data collection system. The dashboard is in clinician perspective and features representations that were developed iteratively together with the clinical team. Each of the representations allows deeper exploration by selecting it to show more detailed data
• Figure 3: Schematic overview of the temporal representation learning workflow and downstream tasks for time-informed population-level analyses. (a) Multi-modal rehabilitation data, originating from digital assessments, medical records, and interventional data collected across multiple clinical cohorts. (b) Downstream analysis, which depends on the research question and may include causal effect inference, trajectory modeling, or subpopulation analysis.