A Localized Feature Description Means Assisting Diabetic Macular Edema Detection and Classification

Heretofore schemes highly rely on complex feature characterization modules assisting the detection and classification of Diabetic Macular Edema (DME). DME encompasses intrinsic local variations that need to be analyzed for treatment. Accordingly, this paper delivers a localized fundus image feature characterization scheme supporting acute abnormality detection and classification of diverse DME abnormalities. This is accomplished by deploying the localized Triangulated Feature Descriptor (TFD) that solely operates on fundus images to capture appropriate features that are deemed essential for DME detection and classification. TFD is operated in three modes for extracting the vital features and is then coupled with simple image representation schemes to deliver precise feature descriptors concerned with diverse abnormalities. The resultant features are then classified by a supervised learning machine for grading the diverse DME ailments. The novel mechanism is elaborately investigated on benchmarked databases namely DRIVE, DIARETDB1, and MESSIDOR using Receiver Operating Characteristics parameters. The assessments reveal the superiority of the proposed approach over its counterparts. Also, the minimal computational operations involved in capitulating the essential image characteristics is another quality that makes this scheme amicable for real-time DR diagnosis.