Brain Stroke Classification via Machine Learning Algorithms Trained with a Linearized Scattering Operator.

Area of Science:

• Medical Imaging
• Machine Learning
• Biomedical Engineering

Background:

• Accurate and timely brain stroke classification is critical for patient outcomes.
• Existing methods for brain stroke detection using microwave imaging require large datasets for machine learning model training.
• Generating these datasets is computationally intensive and time-consuming.

Purpose of the Study:

• To propose an efficient and fast method for generating large datasets for machine learning-based brain stroke classification.
• To apply and validate this method using a realistic microwave imaging system model.
• To compare the performance of different machine learning algorithms for stroke classification.

Main Methods:

• Utilized the distorted Born approximation and linearization of the scattering operator to accelerate dataset generation.
• Developed a microwave imaging system model with 24 conformal antennas operating at 1 GHz on a 3D anthropomorphic head model.
• Trained and evaluated Support Vector Machine (SVM), Multilayer Perceptron (MLP), and K-Nearest Neighbors (KNN) algorithms.

Main Results:

• The proposed method significantly reduces the time required for dataset generation.
• All tested machine learning algorithms successfully classified stroke presence, type (hemorrhagic/ischemic), and location.
• Performance was validated using full-wave simulations, including variations in antenna design and amplitude-only data acquisition.

Conclusions:

• The developed method is efficient and fast for creating essential datasets for brain stroke classification.
• Machine learning algorithms show promise for real-time stroke detection using microwave imaging.
• Further research can explore real-time applications and system optimization.