Abstract
Purpose: :
Telemedicine and computer assisted analysis can aid in the early detection of diabetes and other diseases in large patient populations. We present an algorithm to automatically detect exudate-like structures in a single fundus image and diagnose clinically significant macular edema (CSME) using numerical measurements of the detections. The algorithm was developed for a telemedicine network of primary care clinics based in Memphis, TN, North Carolina, and Mississippi.
Methods: :
The exudation detection involves Field of View detection, vessel segmentation, a novel image normalization method, wavelet edge analysis and evaluation of the global probability of exudation. The main contributions to the field are the normalization approach, which is fast and consistent across the dataset employed, and the use of a limited set of features to diagnose CSME.
Results: :
We assessed the algorithm performance using a dataset of 169 fundus images collected from the telemedicine network with a diverse ethnic background (59% African-American, 28% Caucasian, 10% Hispanic and 3% other). The algorithm detected on average 58% of the exudates per image with CSME, and detected lesions on 100% of the images with lesions. The sensitivity and positive predictive value of the exudates segmentation was measured as 0.81/0.40 on an image-by-image basis. The method was able to detect CSME with area under ROC curve of 0.81. The time required for the complete diagnosis is ~4 seconds per image.
Conclusions: :
Our approach is a promising method to automatically diagnose CSME quickly and efficiently. This method has the advantage of not needing a long training phase which is typical of pattern recognition approaches which can be error prone given the difficulty in accurately ground-truthing lesions. In a broad-based screening environment, this approach should yield a great reduction in the number of patients who require review by an ophthalmologist with a very low risk of classifying a CSME patient as healthy.
Keywords: image processing • diabetic retinopathy • imaging/image analysis: non-clinical