Improved speed and accuracy of plus disease quantification using image fusion methodology

Abstract

IntroductionThe diagnosis of plus disease in Retinopathy of Prematurity (ROP) largely determines the need for treatment; however, this diagnosis is subjective and prone to error. ROPtool is a semiautomated computer program that quantifies vascular tortuosity and dilation. A previous study showed that more than half of video indirect ophthalmoscopy still images had insufficient quality to permit analysis by ROPtool. We evaluated the ability of an image fusion methodology (robust mosaicing) to increase traceability and improve efficiency of posterior pole disease analysis using ROPtool.MethodsWe reviewed video indirect ophthalmoscopy images from routine ROP examinations of 39 right eyes of 39 infants (September 2010-March 2011). We selected the best unenhanced still image from the video for each infant. Robust mosaicing, a multi-frame image fusion algorithm, created an enhanced still image from the same video for each eye. We evaluated the time required and ROPtool’s ability to analyze two major vessels per quadrant on the enhanced vs unenhanced still images.ResultsMean (range) gestational age was 27 weeks (24-35); birthweight 885 g (540-1660). Of 156 quadrants available for analysis, 10 (6%) had plus and 11 (7%) had pre-plus disease. ROPtool analysis was faster (124 vs 153 seconds; P = 0.017) and able to trace more quadrants (143/156, 92% vs 115/156, 74%; P < 0.0001) using enhanced vs unenhanced still images, respectively.DiscussionEnhancing images with robust mosaicing increases traceability and decreases time to analyze posterior pole vessels by ROPtool.ConclusionsRetinal image enhancement using robust mosaicing advances efforts to automate the grading of posterior pole disease in ROP. IntroductionThe diagnosis of plus disease in Retinopathy of Prematurity (ROP) largely determines the need for treatment; however, this diagnosis is subjective and prone to error. ROPtool is a semiautomated computer program that quantifies vascular tortuosity and dilation. A previous study showed that more than half of video indirect ophthalmoscopy still images had insufficient quality to permit analysis by ROPtool. We evaluated the ability of an image fusion methodology (robust mosaicing) to increase traceability and improve efficiency of posterior pole disease analysis using ROPtool. The diagnosis of plus disease in Retinopathy of Prematurity (ROP) largely determines the need for treatment; however, this diagnosis is subjective and prone to error. ROPtool is a semiautomated computer program that quantifies vascular tortuosity and dilation. A previous study showed that more than half of video indirect ophthalmoscopy still images had insufficient quality to permit analysis by ROPtool. We evaluated the ability of an image fusion methodology (robust mosaicing) to increase traceability and improve efficiency of posterior pole disease analysis using ROPtool. MethodsWe reviewed video indirect ophthalmoscopy images from routine ROP examinations of 39 right eyes of 39 infants (September 2010-March 2011). We selected the best unenhanced still image from the video for each infant. Robust mosaicing, a multi-frame image fusion algorithm, created an enhanced still image from the same video for each eye. We evaluated the time required and ROPtool’s ability to analyze two major vessels per quadrant on the enhanced vs unenhanced still images. We reviewed video indirect ophthalmoscopy images from routine ROP examinations of 39 right eyes of 39 infants (September 2010-March 2011). We selected the best unenhanced still image from the video for each infant. Robust mosaicing, a multi-frame image fusion algorithm, created an enhanced still image from the same video for each eye. We evaluated the time required and ROPtool’s ability to analyze two major vessels per quadrant on the enhanced vs unenhanced still images. ResultsMean (range) gestational age was 27 weeks (24-35); birthweight 885 g (540-1660). Of 156 quadrants available for analysis, 10 (6%) had plus and 11 (7%) had pre-plus disease. ROPtool analysis was faster (124 vs 153 seconds; P = 0.017) and able to trace more quadrants (143/156, 92% vs 115/156, 74%; P < 0.0001) using enhanced vs unenhanced still images, respectively. Mean (range) gestational age was 27 weeks (24-35); birthweight 885 g (540-1660). Of 156 quadrants available for analysis, 10 (6%) had plus and 11 (7%) had pre-plus disease. ROPtool analysis was faster (124 vs 153 seconds; P = 0.017) and able to trace more quadrants (143/156, 92% vs 115/156, 74%; P < 0.0001) using enhanced vs unenhanced still images, respectively. DiscussionEnhancing images with robust mosaicing increases traceability and decreases time to analyze posterior pole vessels by ROPtool. Enhancing images with robust mosaicing increases traceability and decreases time to analyze posterior pole vessels by ROPtool. ConclusionsRetinal image enhancement using robust mosaicing advances efforts to automate the grading of posterior pole disease in ROP. Retinal image enhancement using robust mosaicing advances efforts to automate the grading of posterior pole disease in ROP.