Feature Based Differentiator for Fluid Filled Retinal Abnormalities in Retina Using Optical Coherence Tomography Images

Abstract

Introduction: The Optical Coherence Tomography is widely used in ophthalmic imaging to assess the condition of retina. It serves as an effective tool in diagnosing various fluid related abnormalities in retina, which are prior stages to vision loss. The overall pattern of fluid collection is a vital tool in disease identification. Aim: To classify various fluid filled retinal abnormalities, like, Cystoid Macular Edema (CME), Choroidal Neo Vascular Membrane (CNVM), and Macular Hole (MH) based on various features. Materials and Methods: A total of 114 images were acquired using TOPCONN and ZEISS OCT devices. The obtained images were converted to grayscale and subjected to pre-processing technique. Homomorphic Wiener filter was used to remove the speckle noises. The Region of Interest was then identified by basic edge detection algorithm, for which, various features were extracted and utilised for classification. Outputs were cross verified with a medical expert and the performance of the proposed system was evaluated. Results: Based on the proposed system of classification of various fluid filled retinal disorders, the input image was classified as Class 0—Normal, Class 1—CME, Class 2—CNVM, and Class 3—MH based on the pattern of fluid accumulation. The overall performance was compared and evaluated and it was identified that the system exhibited 91.65% accuracy, 90.36% Sensitivity, 92.95% Specificity and Youden's Index value of 0.83. Conclusion: As OCT serves an important tool for pre-screening of blindness, automations in these areas remains of higher potential. The classification is based on the features used, which showed a significant difference between the various classes as classified. The overall performance seems to be satisfactory and would thus help in early detection of fluid related abnormalities and also can be used as an expert tool to analyse the efficiency of the therapies.