Automated Grading of Diabetic Macular Edema by Grid Scanning Optical Coherence Tomography

Abstract

Abstract: : Purpose: To improve the detection of clinically significant diabetic macular edema (DME) using optical coherence tomography (OCT). Methods:The Stratus OCT system was used to scan the macula of patients referred for evaluation of diabetic macular edema (DME). The standard Fast Thickness Map scan pattern (FTM) and a custom designed "Macular Grid 5" (MG5, 768 A–scans in even spaced grid, 5mm diam.) were used on 22 eyes of 12 subjects. OCT macular retinal thickness maps (FTM) from 75 normal subjects were used as reference. Edema on the OCT map is defined as thickening > 3 SD above the normal reference mean. We developed an automated image processing program for the MG5 scan to map retinal thickness, detect areas of edema, compute parameters on areas of retinal thickening (area, distance to fixation) and grade Clinically Significant Macular Edema (CSME). The MG5 gradings are compared with clinical gradings based on biomicroscopy and stereo photography. The following definitions are used for grading (presence v. absence) macular edema: CSME1: retinal edema within 0.5 mm radius of foveal center (fixation). CSME2: retinal edema > 1 disc area (1.5 mm diam. circle) within 1.5 mm of center. DME: any macular edema. The standard STRATUS analysis of FTM was also used to detect CSME1. Results: The automated detection of retinal boundaries was verified visually and found to be correct in 44/47 MG5 scans (our software) and on 19/24 FTM scans (Stratus software). Macular grid 5 (MG5) is scanned twice for each eye. The repeatability (pooled SD) of the total area of edema was 0.81 mm² (CV=8.1%) and for closest distance between edema & fixation was 0.09 mm. CSME1 was detected in 4 eyes by MG5 only, in 1 eyes clinically only, and in 8 eyes by both. CSME2 was detected in 2 eyes by MG5 only, in 3 eyes clinically only, and in 11 eyes by both. DME was detected in 1 eye by MG5 only, in 1 eye clinically only, and in 15 eyes by both. Comparing MG5 and FTM, CSME1 was detected in 3 eyes by MG5 only, in 1 eye by FTM only and in 9 eyes by both. Conclusions: The processing program for the OCT grid scan (MG5) is accurate and repeatable. Automated classification by the new MG5 scan pattern was slightly more likely to detect CSME1 but generally correlated well with clinical grading and standard OCT analysis (FTM). MG5 provides more information in the perifoveal macula than FTM for the diagnosis of CSME2. Automated grading improves the objectivity of CSME diagnosis and may be useful in clinical studies.

Keywords: imaging/image analysis: clinical • diabetic retinopathy • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound)