Weakly Supervised Deep Learning-Based Optical Coherence Tomography Angiography.

Area of Science:

• Biomedical Imaging
• Medical Technology
• Ophthalmology

Background:

• Optical coherence tomography angiography (OCTA) is vital for microvasculature imaging.
• Deep learning excels in OCTA reconstruction but requires high-quality, difficult-to-obtain labels.
• Existing methods face challenges due to imaging hardware and data acquisition constraints.

Purpose of the Study:

• To develop and evaluate a weakly supervised deep learning pipeline for OCTA reconstruction.
• To overcome the dependency on high-quality training labels in OCTA imaging.
• To assess the feasibility of this new approach for clinical applications.

Main Methods:

• Proposed an unprecedented weakly supervised deep learning pipeline for OCTA reconstruction.
• Utilized a cross-validation strategy for evaluation.
• Tested the pipeline on both in vivo animal and human eye datasets.

Main Results:

• The weakly supervised approach demonstrated comparable or superior performance to supervised learning methods.
• The pipeline successfully reconstructed OCTA images without high-quality labels.
• The method proved effective on diverse datasets, including human eye data.

Conclusions:

• Weakly supervised learning is a viable strategy for OCTA reconstruction.
• The proposed pipeline offers a promising alternative to label-dependent methods.
• This approach holds potential for advancing clinical OCTA applications.