Abstract
Purpose :
To retrieve flow signal from under-sampled low-quality 6×6-mm optical coherence tomography angiography (OCTA) of the superficial vascular complex (SVC) of the retina using deep learning.
Methods :
6×6-mmand 3×3-mm OCTA scans of the macula were acquired with scanning density of 304×304 lines on one or both eyes of 178 participants (141 diabetic retinopathy (DR), 37 healthy) using a 70-kHz commercial OCT system (Optovue, Inc.). SVC angiograms were generated by maximum projection of the OCTA signal in the slab including nerve fiber layer and ganglion cell layer. A deep residual learning network [Fig. 1] was trained to generate high-resolution 6×6-mm SVC angiograms using the cropped 3×3-mm section of the low-resolution 6×6-mmangiograms as input, and the registered 3×3-mm SVC angiograms of the same eye as ground truth. The loss function used in the learning stage was a linear combination of the mean square error and the structural similarity. On randomly selected 29 DR eyes, we evaluated the high-resolution 6×6-mmangiogram produced by the network for noise intensity in foveal avascular zone, contrast, vascular connectivity within the 3×3-mm section, and false flow signal on the reconstructed images.
Results :
Compared to the original 6×6-mm angiograms, the noise intensity of reconstructed images was significantly reduced (29.76 ± 30.57 vs. 0.35 ± 1.55, t-test, p <0.001); the image contrast was significantly improved (0.19 ± 0.00 vs. 0.22 ± 0.00, t-test, p <0.001). Compared to the original 3×3-mm angiogram with proper sampling density,vascular connectivity of the reconstructed cropped 3×3-mm section of 6×6-mm was significantly enhanced (0.90 ± 0.01 vs. 0.98 ± 0.00, t-test, p <0.001) [Fig. 2]. In a noise simulation experiment, we found our algorithm did not generate false flow signal when the noise intensity was under 500, which is far above the noise intensity measured in original 3×3-mm (117.15 ± 97.40) and 6×6-mm (29.76 ± 30.57) angiograms.
Conclusions :
Our method can reconstruct high-resolution angiogram from the low-definition 6×6-mm en face OCTA. The enhanced 6×6-mm angiograms may improve the accuracy of disease biomarkers such as non-perfusion area and vessel density.
This is a 2020 ARVO Annual Meeting abstract.