Abstract
Purpose:
To demonstrate the sensitivity and specificity of the human retina’s optical properties to multiple sclerosis (MS).
Methods:
Our research group has previously established the possibility to statistically classify and identify eyes belonging to diabetic patients from healthy controls, based on the optical properties of the retina. In the current work, we resort to the same approach, aiming to identify specific changes associated to MS. High-definition optical coherence tomography (HD-OCT, Carl Zeiss Meditec, Dubin, CA, USA) was used to image eyes from patients diagnosed with MS (35 eyes from 19 patients) and eyes from a control group (35 eyes from 22 patients). In addition, we collected OCT scans of diabetic retinopathy (DR) patients (ETDRS levels 10 to 35) from our institutional database (35 eyes from 22 patients). In group selection, age-matching was optimized to prevent the possible influence of ageing factors, as previously demonstrated (MS: 38.4 (6.3); controls: 37.0 (7.5); DR: 48.1 (2.3)) (mean (SD), years). OCT scans were segmented to obtain the layers between the inner limiting membrane and the retinal pigment epithelium and the probability density function (PDF) of the resulting data was computed. From each PDF, a set of parameters was determined by capturing the shape of the distribution and used as input to a supervised classification process (support vector machine). To improve the performance of the classification, greedy backward elimination and forward selection routines were ran to identify an optimal set of features and a 10-fold cross-validation process was applied to determine system performance.
Results:
The classification process between the MS and the control group achieved an accuracy of 95.7% with sensitivity of 100% and specificity of 91.4%. To further evaluate the specificity of the process, a classification into three groups (MS, controls and DR) was performed. The system achieved an accuracy of 85.7%, demonstrating the specificity of the information embedded in the optical properties of the retina for the healthy status, diabetes (a multifactorial disease) and MS (a central nervous system disorder).
Conclusions:
These results make evident the possibility to use the retina as a window into brain diseases and systemic conditions, such as diabetes. Additionally, it proves that the retina’s inherent optical information is biologically specific to each of the conditions tested.
Keywords: 688 retina •
550 imaging/image analysis: clinical •
549 image processing