Speckle pattern analysis with deep learning for low-cost stroke detection: a phantom-based feasibility study.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Stroke is a major cause of global disability, requiring prompt diagnosis to prevent irreversible brain damage.
• Current advanced imaging techniques are often inaccessible in resource-limited settings due to cost and logistical challenges.

Purpose of the Study:

• To evaluate a laser speckle-based imaging technique integrated with deep learning for detecting simulated stroke conditions.
• To assess the feasibility of differentiating healthy blood flow from restricted flow states using speckle patterns in artificial arteries.

Main Methods:

• Artificial arteries (3-6 mm diameter) were embedded in chicken tissue at depths of 0-10 mm to simulate various blood flow scenarios.
• A high-speed camera captured laser-induced speckle patterns, which were then analyzed by a 3D convolutional neural network (X3D_M) for flow condition classification.

Main Results:

• The technique achieved high classification accuracy, ranging from 95% to 100% for superficial, larger vessels.
• Robust detection was maintained for smaller or deeper arteries, indicating the reliability of spatiotemporal speckle pattern analysis.

Conclusions:

• Combining speckle imaging with deep learning presents a potential solution for accessible and rapid stroke detection.
• Future research will focus on in vivo experiments, addressing challenges like skull optical properties and near-infrared illumination for intracranial applications.