Computer-Based Image Analysis for Plus Disease Diagnosis in Retinopathy of Prematurity

Abstract

Presence of plus disease in retinopathy of prematurity (ROP) is an important criterion for identifying treatment-requiring ROP. Plus disease is defined by a standard published photograph selected over 20 years ago by expert consensus. However, diagnosis of plus disease has been shown to be subjective and qualitative. Computer-based image analysis, using quantitative methods, has potential to improve the objectivity of plus disease diagnosis. The objective was to review the published literature involving computer-based image analysis for ROP diagnosis. The PubMed and Cochrane library databases were searched for the keywords “retinopathy of prematurity” AND “image analysis” AND/OR “plus disease.” Reference lists of retrieved articles were searched to identify additional relevant studies. All relevant English-language studies were reviewed. There are four main computer-based systems, ROPtool (AU ROC curve, plus tortuosity 0.95, plus dilation 0.87), RISA (AU ROC curve, arteriolar TI 0.71, venular diameter 0.82), Vessel Map (AU ROC curve, arteriolar dilation 0.75, venular dilation 0.96), and CAIAR (AU ROC curve, arteriole tortuosity 0.92, venular dilation 0.91), attempting to objectively analyze vessel tortuosity and dilation in plus disease in ROP. Some of them show promise for identification of plus disease using quantitative methods. This has potential to improve the diagnosis of plus disease, and may contribute to the management of ROP using both traditional binocular indirect ophthalmoscopy and image-based telemedicine approaches.

Computer-Based Image Analysis for Plus Disease Diagnosis in Retinopathy of Prematurity: Performance of the “i-ROP” System and Image Features Associated With Expert Diagnosis

“…While it is challenging to directly compare the performance of the i-ROP system to prior CBIA systems due to different reference standards, classification levels, and performance measures, no prior system has performed as well as the human experts it was compared against at three level classification. 14 …”

“…[8][9][10][11] Telemedicine has been used to address these challenges in delivery of care, and computer-based image analysis could complement these systems. [12][13][14][15] There have been numerous studies on computeraided quantification of vascular features in ROP. [13][14][15] However, previous studies have had several limitations: (1) Little work to our knowledge has focused on comparative analysis to identify image features that are most important for diagnosis.…”

“…[12][13][14][15] There have been numerous studies on computeraided quantification of vascular features in ROP. [13][14][15] However, previous studies have had several limitations: (1) Little work to our knowledge has focused on comparative analysis to identify image features that are most important for diagnosis. 16,17 (2) Most previous studies have worked on two-level classification (plus versus not plus), and do not address preplus disease.…”

“…16,17 (2) Most previous studies have worked on two-level classification (plus versus not plus), and do not address preplus disease. 13,14 (3) The ''reference standards'' for evaluation of computer systems have been very limited, often consisting of diagnosis by a single clinician. (4) There has been no standardization regarding which specific image sizes and vascular parameters should be measured (Patel SN, et al IOVS.…”

“…Most studies have used regular statistics (e.g., minimum, maximum, and median) of measured parameters (e.g., tortuosity, integrated curvature), 14,15 but these statistics may provide biased estimates about disease severity since an image contains healthy and abnormal vessels. This study was designed to address all of these gaps in knowledge.…”