Retinal image analysis based on mixture models to detect hard exudates.

摘要: Diabetic Retinopathy is one of the leading causes blindness in developed countries. Hard exudates have been found to be most prevalent earliest clinical signs retinopathy. Thus, automatic detection hard from retinal images clinically significant. In this study, an method detect proposed. The algorithm based on mixture models dynamically threshold order separate background. A postprocessing technique, edge detection, applied distinguish cotton wool spots and other artefacts. We prospectively assessed performance using a database 80 with variable colour, brightness, quality. obtained sensitivity 90.2% positive predictive value 96.8% lesion-based criterion. image-based classification accuracy also evaluated obtaining 100% specificity 90%.