Near-real-time Mueller polarimetric image processing for neurosurgical intervention

Area of Science:

• Biomedical Optics
• Medical Imaging
• Neurosurgery

Background:

• Wide-field imaging Mueller polarimetry offers label-free, non-invasive visualization of white matter fiber orientation for neurosurgery.
• Conventional methods require long acquisition times, hindering real-time clinical application.
• Current polarimetric data processing limits translation to clinical practice.

Purpose of the Study:

• To develop a learning-based denoising framework for fast, single-shot, noisy polarimetric acquisitions.
• To devise performance-optimized image processing tools for deriving clinically relevant parameters.
• To enable near-real-time performance for polarimetric parameter estimation in neurosurgery.

Main Methods:

• A deep learning framework was tailored for denoising single-shot, noisy polarimetric intensity measurements.
• Performance-optimized image processing tools were developed to derive key polarimetric parameters.
• The framework was trained and validated on human brain samples, including tumor-affected tissues.

Main Results:

• The denoising framework achieved significant improvements in accuracy and image quality compared to state-of-the-art methods.
• Near-real-time processing was achieved for a localized field of view (approximately 6.5 mm²).
• Denoised data revealed clear directional patterns of neuronal fiber tracts, preserving information on directional disruption in neoplastic lesions.

Conclusions:

• The developed end-to-end image processing achieved real-time performance for neurosurgical applications.
• The method provides high-quality, label-free visualization of white matter architecture.
• These advancements facilitate the clinical translation of Mueller polarimetry for oncological neurosurgery.