Abstract
The study of retinal hemodynamics plays an important role to understand the onset and progression of diabetic retinopathy. In this work, we developed an interactive retinal analysis tool to quantitatively measure the blood flow velocity (BFV) and blood flow rate (BFR) in the macular region using the Retinal Function Imager (RFI). By employing a high definition stroboscopic fundus camera, the RFI device is able to assess retinal blood flow characteristics in vivo. However, the measurements of BFV using a user-guided vessel segmentation tool may induce significant inter-observer differences and BFR is not provided in the built-in software. In this work, we have developed an interactive tool to assess the retinal BFV and BFR in the macular region. Optical coherence tomography data was registered with the RFI image to locate the fovea accurately. The boundaries of the vessels were delineated on a motion contrast enhanced image and BFV was computed by maximizing the cross-correlation of pixel intensities in a ratio video. Furthermore, we were able to calculate the BFR in absolute values (μl/s). Experiments were conducted on 122 vessels from 5 healthy and 5 mild non-proliferative diabetic retinopathy (NPDR) subjects. The Pearson's correlation of the vessel diameter measurements between our method and manual labeling on 40 vessels was 0.984. The intraclass correlation (ICC) of BFV between our proposed method and built-in software was 0.924 and 0.830 for vessels from healthy and NPDR subjects, respectively. The coefficient of variation between repeated sessions was reduced significantly from 22.5% to 15.9% in our proposed method (p<0.001).
•An interactive tool to assess the retinal blood flow value and blood follow rate in the macular region was developed;•Optical coherence tomography data was registered with Retinal function imager image to locate the fovea accurately;•The vessel boundaries were delineated on a motion contrast enhanced image and BFV was computed by maximizing the cross-correlation of intensities;•The coefficient of variation between repeated sessions was reduced significantly from 22.5% in the RFI built-in software to 15.9% in our proposed method (p<0.001).
