Machine learning-based ischemic stroke detection and categorization with non-invasive plantar pressure data.

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• Stroke diagnosis relies on expensive imaging and manual assessments, leading to delays and potential errors.
• There is a critical need for accessible, cost-effective, and automated stroke detection methods.

Purpose of the Study:

• To develop and validate a machine learning-based screening algorithm for stroke detection.
• To utilize non-invasive foot pressure signals as a cost-effective alternative to traditional diagnostic methods.

Main Methods:

• A machine learning algorithm was developed using foot pressure signals recorded during walking.
• Empirical Fourier Decomposition and a novel set of biomarkers were used to analyze pressure distribution.
• The ReliefF algorithm, support vector machine, and K-nearest neighbors were employed for feature selection and classification.

Main Results:

• The algorithm was evaluated on 82 subjects (36 stroke patients, 46 controls) using 198 foot plantar sensors.
• An average accuracy rate of 99.19% was achieved, demonstrating high detection performance.
• The method showed robust performance against clinical factors, including stroke side and blood pressure status.

Conclusions:

• The proposed method offers a unique, reliable, and cost-effective automatic stroke detection solution.
• It successfully identifies stroke using a minimal set of biomarkers from toe and finger regions.
• This approach has the potential to significantly improve early stroke diagnosis and patient management.