Application Research Of OCTA Image Enhancement And Quantification Method Based On Deep Learning

In recent years,optical coherence tomography angiography(OCTA)has drawn tremendous attention in ophthalmology as a novel imaging modality to complement traditional fluorescein angiography(FA)and indocyanine green angiography(ICGA).OCTA is non-invasive thus avoids the risk of dye.Compared with the overlapping depth information of vasculature in 2D imaging like FA and ICGA,OCTA is capable of resolving the vessels and capillaries in-depth direction with a high axial resolution of 5?10?m.It has been widely used in studies of various ocular diseases,such as glaucoma,age-related macular degradation,and retinopathy of prematurity.Also,the OCTA imaging of human retina was found to be able to indicate neurodegenerative disorders,such as mild cognitive impairment and Alzheimer's disease.OCTA requires high transverse sampling rates for visualizing retinal and choroidal capillaries,which impedes the popularization of the OCTA technique due to the high cost of speedy acquisition systems.On the other hand,current wide-field OCTA using low transverse sampling causes the underestimation of vascular biomarkers in quantitative analysis.In this paper,we propose to use deep learning to repair the resolution degeneration induced by the low transverse sampling.We conducted intensive experiments on converting the centrally cropped 3×3 mm~2 field of view(FOV)of the 8×8 mm~2 foveal OCTA images(a sampling rate of 22.9?m)to the native3×3 mm~2 en face OCTA images(a sampling rate of 12.2?m).We employed a deep learning approach by using a SCGAN(structural-consistent adversarial networks)which is based on cycle-consistent adversarial network architecture in this conversion.Qualitative results show the resolutions of superficial vascular plexus(SVP)and deep capillary plexus(DCP)are significantly improved by this deep learning image processing method.We also quantitatively compared the representative vascular biomarkers of the SVP and DCP,such as vessel area density,vessel diameter index,and vessel perimeter index,and found that our method could significantly decrease their discrepancy caused by the high and low sampling rates.We further applied our method to enhance diseased cases and calculate vascular biomarkers,which demonstrates its generalization performance and clinical perspective.