Diagnosis of the diabetic retinopathy by the computer–supported analysis of the digital fundus photograph.

Abstract

Abstract: : Purpose: Digital fundus photographs were analyzed to develop a computer–assisted automatic staging system for the diabetic retinopathy. Methods: High–resolution (2048 x 1360 pixels) color fundus photographs of the diabetic patients were acquired by the digital camera (D10, Canon Inc., Tokyo, Japan) attached to the conventional fundus camera system (CF–60UVi, Canon Inc.), and archived as JPEG format image. Then, after recruitment, JPEG format was converted to 256–colored DICOM format to optimize the analysing process. Optic disc and macula was segmented by the threshold method to provide the 4–quadrant division of the fundus image. Retinal hemorrhage/microaneurysm and hard exudate were segmented and counted in each 4–quadrants by the threshold, seed–growing and snake method. Each image was classified into the appropriate stage according to the newly proposed staging of diabetic retinopathy by WHO. The results were compared with the reading by an ophthalmologist. Results: The recognition of the retinal hemorrhage/microaneurysm and hard exudate by the computer showed 82.4% of sensitivity and 89.8% of specificity. The staging of the diabetic retinopathy by the computer analysis and that by the ophthalmologist showed good correlation (kappa value = 0.72). Conclusions: Automated analysis of the digital fundus image by the computer may provide a preliminary readings and clues of the diabetic retinopathy screening.

Keywords: image processing • imaging/image analysis: clinical • diabetic retinopathy