MEMO: dataset and methods for robust multimodal retinal image registration with large or small vessel density...

Area of Science:

• Ophthalmology
• Medical Imaging
• Biomedical Engineering

Background:

• Accurate measurement of retinal blood flow (RBF) is crucial for early diagnosis and treatment of ocular diseases.
• Current methods lack the precision to determine capillary flowrates effectively.
• Combining erythrocyte-mediated angiography (EMA) for flow measurement and optical coherence tomography angiography (OCTA) for structural imaging offers potential for precise RBF assessment.

Purpose of the Study:

• To address the unexplored challenge of multimodal retinal image registration between EMA and OCTA.
• To introduce MEMO, the first public multimodal EMA and OCTA retinal image dataset.
• To develop a robust method for multimodal retinal image registration despite significant differences in vessel density (VD).

Main Methods:

• Proposed VDD-Reg, a segmentation-based deep learning framework comprising a vessel segmentation module and a registration module.
• Developed LVD-Seg, a two-stage semi-supervised learning framework for training the vessel segmentation module using supervised and unsupervised losses.
• Utilized the newly established MEMO dataset and the CF-FA dataset for evaluation.

Main Results:

• VDD-Reg demonstrated superior quantitative and qualitative performance compared to existing methods.
• The method achieved robust registration results even with large differences in vessel density, as shown on the MEMO dataset.
• VDD-Reg maintained accuracy with as few as three annotated vessel segmentation masks, indicating its practical feasibility.

Conclusions:

• The developed VDD-Reg framework effectively addresses the challenge of multimodal EMA-OCTA retinal image registration.
• The MEMO dataset provides a valuable resource for advancing research in this area.
• VDD-Reg shows significant potential for improving the early diagnosis and management of ocular diseases through precise RBF measurement.