Fractal analysis of Optical Coherence Tomography images for the classification of diabetes-induced retinal damage

Abstract

Purpose: : Differences in optical properties and irregularity measures of normal and abnormal retinal tissue may provide additional information of retinopathy development in diabetic eyes. We evaluated the sensitivity of optical coherence tomography images to sample diabetic morphology using fractal analysis.

Methods: : A total of 74 eyes of healthy subjects (34±12 yrs,), 38 eyes with diabetes mellitus with no retinopathy (DM, 35±10 yrs) and 43 eyes with mild diabetic retinopathy (MDR, 43±17 yrs) on biomicroscopy were enrolled. Optical coherence tomography (OCT) imaging of the macula was performed by Stratus OCT (Carl Zeiss Meditec, Dublin, CA, USA), and OCTRIMA software was used to extract 7 intraretinal layers. Then, fractal dimension was calculated using a 1D wavelet algorithm. ANOVA followed by Newman-Keuls post-hoc analyses were used to test for differences between groups. The level of significance was set at p<0.001. Receiver operating characteristic (ROC) curves were calculated to assess the discriminating power between DM and MDR eye. Area under the receiver operating characteristic curve (AUROC) was used to compare diagnostic power.

Results: : Fractal dimension increased for all the layers except the ganglion cell layer and inner plexiform layer complex (GCL+IPL, Control vs. DM vs. MDR, 1.68±0.01 vs. 1.62±0.06 vs. 1.57±0.05, respectively, p<0.001). The highest AUROC values estimated for the fractal dimension was observed for GCL+IPL when comparing MDR with DM eyes (0.77±0.05). The maximum discrimination value of 0.80 for FD (standard error =0.05) for the GCL+IPL complex was obtained at a FD ≤ 1.56 (cut off point, asymptotic 95% CI: lower-upper bound=0.67-0.87). At this value, the sensitivity for the GCL+IPL complex is 65.1% with a specificity of 76.3%.

Conclusions: : Our results suggest that diabetic eyes with and without early retinopathy may be discriminated by analyzing the OCT signal using fractal analysis. The decrease of fractal dimension of the GCL+IPL layer may be due to apoptosis and the consequent cellular structural changes. The use of OCT fractal analysis for the classification of diabetes-induced retinal damage could potentially provide an additional diagnostic information for the early detection of DR and the follow-up of its progression.

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