Related Experiment Video
Updated: Jul 3, 2026
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
Decoding functional changes in the brain following ischemic stroke: a multimodal feature approach integrating fNIRS
Hong Xu1, Cunyuan Luan2, Jiawen Yin3
1Neurology Department, Beijing Hospital, National Center for Gerontology, National Clinical Research Center for Gerontology, The Key Laboratory of Geriatrics of NHC, Institute of Geriatric Medicine, Chinese Academy of Medical Sciences, China; School of Medicine, University of Chinese Academy of Sciences, Beijing 100049, China.
Functional near-infrared spectroscopy (fNIRS) combined with deep learning effectively detects neurovascular deficits in ischemic stroke (IS) patients. This approach shows promise for non-invasive post-stroke assessment and rehabilitation planning.
Area of Science:
- Neuroscience
- Medical Imaging
- Machine Learning
Background:
- Ischemic stroke (IS) causes significant brain network remodeling.
- Functional near-infrared spectroscopy (fNIRS) monitors cerebral hemodynamics but its potential for complex dysfunction is underutilized.
Purpose of the Study:
- Investigate task-evoked and resting-state fNIRS responses to decode neurovascular deficits in IS patients.
- Compare the performance of traditional machine learning classifiers with a Transformer-based deep learning model for IS detection.
Main Methods:
- 160 participants (115 IS patients, 45 controls) underwent fNIRS during resting state, verbal fluency task (VFT), and right-hand gripping task (RHGT).
- Hemodynamic features (HbO, HbR, HbT) were analyzed for functional connectivity and cortical activation.
- Machine learning models, including a Transformer, were used for IS classification, with SHAP analysis for feature importance.
Main Results:
- IS patients showed reduced resting-state functional connectivity in sensorimotor regions.
- VFT revealed hemodynamic abnormalities in prefrontal areas (DLPFC, BA45).
- The Transformer model achieved superior classification performance (AUC=0.86, accuracy=86%) compared to traditional classifiers, with sensorimotor and prefrontal regions identified as key discriminative features.
Conclusions:
- Distinct hemodynamic abnormalities were identified in IS patients across different fNIRS paradigms.
- fNIRS combined with deep learning offers a promising non-invasive method for post-stroke functional assessment.
- Sensorimotor and prefrontal regions are crucial for distinguishing IS patients from controls using fNIRS data.