Effects of Image Compression and Degradation on an Automatic Diabetic Retinopathy Screening Algorithm

Diabetic retinopathy (DR) is one of the leading causes of blindness among adult Americans. Automatic methods for detection of the disease have been developed in recent years, most of them depending on segmentation of bright and red lesions. In this paper we present an automatic DR screening system that does not require segmentation of features and it is only based on textural features obtained using multiscale Amplitude Modulation-Frequency Modulation (AM-FM) decompositions. From the AM-FM decompositions we extract the instantaneous amplitude, instantaneous frequency magnitude, and relative angle for each of the combinations of scales. These features are the inputs of a classification algorithm that uses k-means and partial least squares (PLS). The algorithm achieves an accuracy of detection of 0.88 area under the ROC curve (AROC) for a set of 280 images from the Messidor database. Testing is done using the cross-validation methods, with 168 images being used for training and 112 for testing. The original algorithm is then used to analyze the effects of image compression and degradation, which are expected to be present in an eventual implementation of the system on a clinical setting. Results show that the algorithm is insensitive to illumination variations, but that high rate of compression and large blurring effects degrade its performance.