A stroke detection and discrimination framework using broadband microwave scattering on stochastic models with deep...

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Artificial Intelligence

Background:

• Stroke is a leading cause of death and disability globally.
• Current diagnostic methods are often limited by time constraints and accessibility.
• There is a significant need for rapid, mobile, safe, and low-cost stroke detection technologies.

Purpose of the Study:

• To investigate the use of microwave scattering perturbations for stroke detection.
• To develop deep neural networks for classifying stroke and characterizing hemorrhage.
• To create a novel intelligent diagnostic approach for stroke using microwave technology.

Main Methods:

• Utilized ultra-wideband antenna arrays to capture microwave scattering data.
• Developed two deep neural networks: a classification network for stroke detection and a discrimination network for hemorrhage characterization.
• Trained and validated models on a simulated cohort of 666 subjects using 2D stochastic head models.

Main Results:

• The classification network achieved a stroke detection accuracy greater than 94% with an AUC of 0.996.
• The discrimination network demonstrated a mean squared error of less than 0.004 cm for localization and 0.02 cm for size estimation.
• Successfully learned dielectric signatures of disease from microwave scattering data.

Conclusions:

• The developed microwave-based approach offers a novel method for intelligent stroke diagnostics.
• This technology has the potential to circumvent conventional imaging techniques for faster stroke assessment.
• The findings highlight a promising solution for the unmet clinical need in rapid stroke detection.