Skip to main content
. 2013 May 24;54(5):3546–3559. doi: 10.1167/iovs.12-10347

Table 1.

ARIA Literature on DR at a Glance

Reference No.
DR Task
Method
Public Dataset?
No. of images
Image Type
Validation Level
Sensitivity
Specificity
2 MA, dot HM Pixel clustering, kNN No 16,670 Color fundus Image 47.7% 90%
51 MA Radon transform, wavelet preprocessing Yes 100 Color fundus Lesion 50% >10 false positive per image
49 MA Wavelet transform Yes 100 Color fundus Lesion 60% 8 false positive per image
52 MA Double-ring filter Yes 100 Color fundus Lesion 65% 27 false positive per image
53 MA Wavelet, genetic algorithm No 1115 Color fundus + angiogram Lesion 90.24% 89.75%
54 MA Diameter closing, feature extraction, classification No 115 Color fundus Lesion 88.50% 2.13 false positive per image
55 MA Edge inference No 49 Color fundus Lesion 68% >40 false positive per image
48 Exudates Machine learning No 430 Color fundus Lesion 95% 88%
43 MA, dot HM Combination of methods No 15,473 Color fundus Image 97.90% 67.40%
56 MA Watershed contrast normalization No 1677 Color fundus Image 85.40% 83.10%
57 MA Wavelet template matching No 995 Green channel fundus Lesion 87.90% 96.20%
58 MA Generalized eigenvectors No 70 Color fundus Region 93% NA
59 MA 2-D adaptive filtering, region growing No 11 Angiogram Region 90.72% 82.35%
50 MA, dot HM Pixel classification No 240 Color fundus Region 100% 87%
60 MA, dot HM RetinaLyze (proprietary) No 400 Color fundus Image 96.70% 71.40%
61 Drusen Histogram-adaptive local thresholding No 23 Color fundus Region 98.80% 99.31%
62 MA, HM Morphological, region growing, neural network No 142 Color fundus Region 77.50% 88.70%
63 Exudates Dynamic clustering with domain knowledge No 543 Color fundus Image 100% 74%
64 MA Morphological, region growing No 46 Color fundus Image 90% 80%
65 Exudates Statistical classification with local window-based verification No 200 Color fundus Image 100% 70%
66 MA Manual rule-based classifier No 3885 Red-free Image 85% 76%
46 MA, HM, exudates, cotton wool Estimate background intensity, extract candidate regions for classification No 268 Color fundus Lesion 94% 69%
67 MA, dot HM Pattern recognition No 400 35-mm color slides Lesion NA NA
68 MA Manual rule-based classifier No 88 Angiogram Lesion 82% 84%
69 HM, exudates Neural network No 480 Red-free Region 73.80% 73.80%
70 MA Matched filter No 6 Angiogram Lesion 45% >150 false positive per image

MA, microaneurysm; HM, haemorrhage; kNN, k-nearest neighbor.