Related Experiment Video
Updated: Sep 7, 2025
06:37
Electroencephalography Network Indices as Biomarkers of Upper Limb Impairment in Chronic Stroke
Published on: July 14, 2023
965
Stroke recovery phenotyping through network trajectory approaches and graph neural networks
Sanjukta Krishnagopal1, Keith Lohse2, Robynne Braun3
1Gatsby Computational Neuroscience Unit, University College London, London, W1T 4JG, UK. s.krishnagopal@ucl.ac.uk.
Brain Informatics
|June 19, 2022
Summary
This study introduces network science and machine learning to identify distinct stroke recovery patterns. The novel approach aids in predicting patient recovery trajectories and understanding complex symptom interactions post-stroke.
Area of Science:
- Neurology
- Computational Neuroscience
- Data Science
Background:
- Stroke causes significant neurological injury affecting multiple domains.
- Clinical recovery tracking uses diverse measures, challenging integrated analysis.
- Existing methods limit understanding of complex, time-evolving symptom interactions.
Purpose of the Study:
- To apply network science and machine learning for identifying stroke recovery patterns.
- To develop a method for predicting recovery trajectories and symptom progression.
- To enhance the clinical application of computational approaches in stroke research.
Main Methods:
- Utilized Trajectory Profile Clustering (TPC) on NINDS tPA trial data.
- Analyzed 11 neurological domains across 5 time points.
- Employed graph neural networks for predictive validation.
Main Results:
- Identified 3 distinct stroke recovery trajectory profiles.
- Profiles align with clinically relevant stroke syndromes and symptom severity.
- Demonstrated effective prediction of patient stratification into recovery profiles.
Conclusions:
- Trajectory Profile Clustering effectively identifies clinically relevant recovery subtypes.
- The approach enables early prediction of symptom progression subtypes.
- This work pioneers network trajectory approaches for stroke recovery phenotyping.
