This article is more than 5 years old. Information may no longer be current.
Computer-based image analysis expands diagnostic parameters for ROP
Click Here to Manage Email Alerts
Image-based grading was more accurate than standard classification in diagnosing retinopathy of prematurity, according to a study presented at the American Ophthalmological Society meeting in Colorado Springs, Colorado.
The identification of plus disease, defined as dilation of the veins and tortuosity of the arteries, is the most critical parameter for grading ROP, according to Michael F. Chiang, MD. He discussed a comparison between the International Classification of Retinopathy of Prematurity (ICROP) and computer image analysis in diagnosing the zone, stage and extent of ROP.
Vascular abnormality in ROP is a continuous spectrum, and discrete classification of ROP as plus disease, pre-plus and normal is overly simplistic, Chiang said.
“The definition of plus disease is very subjective, in many ways,” Chiang said in a subsequent interview with Ocular Surgery News. “For something that’s supposed to be so standardized, there’s actually a significant amount of variability.”
The study showed high variability between experts in diagnosing plus disease, Chiang said.
Standard classification
ICROP was first implemented in 1984 as a way to standardize parameters of ROP such as zone, stage and presence of plus disease, Chiang said. Pre-plus disease was added as a category in 2005.
“[ICROP] was actually a really big advance because it tried to standardize everything. Unless everybody uses the same terminology and speaks the same language, you can’t have consistency,” he said.
However, it does not account for other retinal abnormalities that factor into the diagnosis of plus disease; discrete classification of vascular abnormality (plus, pre-plus and normal) is an oversimplification, Chiang said.
“Plus disease is a yes or no thing, but on the other hand, to have tortuous arteries and dilated veins isn’t a yes or no thing. There’s a whole spectrum of abnormalities,” he said. “ICROP is helpful because it standardizes, but it has always seemed to me that the parameters are a real oversimplification because ... you’re forcing ophthalmologists to choose discrete categories, and people often don’t choose the same category.”
Comparison study
In the comparison study, eight experts examined 77 wide-angle RetCam images (Clarity Medical Systems) and were asked to diagnose disease as plus, pre-plus or normal.
“The ICROP definition of plus disease says that ophthalmologists should examine the dilation only of the veins and the tortuosity only of the arteries, and that they should evaluate these only in the posterior pole of the retina,” Chiang said.
Chiang and colleagues developed a computer program to quantify parameters such as tortuosity of the blood vessels.
“This allowed us to quantify only the arteries, only the veins or all of the blood vessels together. We cropped the images to look only at the central retina or a larger part of the retina including the periphery,” Chiang said. “We basically looked at what set of quantitative parameters of the computer system best correlated with the expert diagnosis.”
Computer analysis agreed most strongly with expert diagnosis when a larger area of the retina was assessed, not only the central retina, and when analysis accounted for tortuosity of both the arteries and veins, not only the arteries, Chiang said.
“What we took away from that is that experts are not doing only what this old-time ICROP definition said. No. 1, they’re looking at the peripheral retina, not only the central retina, and No. 2, they’re looking at the abnormality of both the arteries and the veins, not only the tortuosity of the arteries,” he said. “This study matters because it shows that experts are not doing what the definition says, which makes us question, is the definition really what it should be?”
Accuracy of expert diagnosis ranged from 79% to 99%, while accuracy of the computer image-analysis system was 95%. These accuracy values were all based on comparison to a robust external reference standard for plus disease diagnosis. – by Matt Hasson
- References:
- Campbell JP, et al. Ophthalmology. 2016;doi:10.1016/j.ophtha.2016.04.035.
- Chiang M. International classification of ROP. Presented at: American Ophthalmological Society meeting; May 19-22, 2016; Colorado Springs, Colo.
- For more information:
- Michael F. Chiang, MD, can be reached at Oregon Health & Science University, Casey Eye Institute, 3375 SW Terwilliger Blvd., Portland, OR 97239; email: chiangm@ohsu.edu.
Disclosure: Chiang reports he is an unpaid member of the scientific advisory board of Clarity Medical Systems, is a consultant for Novartis, and serves on the steering committee for the RAINBOW Study.