Deep learning-enabled speckle reduction for cleared-sample coherent scattering tomography.

Area of Science:

• Neuroimaging
• Biomedical Optics
• Machine Learning

Background:

• Clearing Assisted Scattering Tomography (CAST) enables whole-brain imaging for visualizing fine-scale brain connectivity.
• CAST, as a coherent optical tomography method, suffers from inherent speckle noise, degrading image quality and hindering quantitative analysis.
• Existing speckle reduction methods for optical coherence tomography (OCT) are not directly applicable to CAST due to differing noise and sample statistics.

Purpose of the Study:

• To develop a specialized speckle reduction method for CAST whole-brain images.
• To effectively suppress speckle noise while preserving fine structural details in CAST neuroimaging data.
• To evaluate the performance and generalizability of the proposed method.

Main Methods:

• Introduced CLEAR Net, a learning-based network designed for speckle reduction in cleared-sample CAST imaging.
• Trained and validated CLEAR Net on whole-brain white matter CAST datasets.
• Benchmarked CLEAR Net against existing speckle reduction algorithms and evaluated its performance on ophthalmic OCT datasets.

Main Results:

• CLEAR Net effectively suppresses speckle noise in whole-brain CAST images.
• The method successfully preserves fine structural details crucial for connectivity analysis.
• CLEAR Net demonstrates generalizability, showing effectiveness on diverse imaging datasets.

Conclusions:

• CLEAR Net offers a robust solution for speckle noise reduction in CAST neuroimaging.
• This advancement improves the quality and quantitative potential of whole-brain connectivity mapping.
• The developed network has potential applications beyond CAST, including other coherent optical imaging modalities.