Abstract
Purpose::
To assess the effect of high-quality JPEG compression on the accuracy of grading digital fundus images in exudative Age-related Macular Degeneration.
Methods::
5 angiograms from 5 different patients that were acquired on a Topcon TRC 50IX system were selected and stored as uncompressed TIFF images (1.25MB per image). The native TIFF angiograms (Group 1) were imported into specialised grading software provided by Digital Healthcare Ltd (Cambridge, UK). The same angiograms were also compressed on import using the GDI+ .NET framework 1.1 JPEG codecs (Group 2) and saved as separate files. A QFactor of 100 was used to compress the images which gives a ratio of about 4:1 for colour images and 3:1 for FA images (circa 300KB per colour image and circa 500KB per fluorescein image). Angiograms were presented for grading in a masked and random fashion to four trained photograders in a Reading Centre at the Central Angiographic Resource Facility. Concise measurements for exudative AMD lesion components were taken on fluorescein angiograms including area of classic or occult, blood, elevated blocked fluorescence (EBF), total lesion and serous pigment epithelial detachment (SPED). Two graders assessed the angiograms in Group 1 and two graders assessed the angiograms in Group 2, and after a period of one week, the groups were switched and were re-graded.
Results::
There were no significant differences in the identification or area measurements of classic CNV, blood, EBF and SPED for compressed versus uncompressed images. Uncompressed and compressed images were similarly graded for percentage of classic by all 4 graders (p > 0.1 paired t-test) indicating that lesion area measurement was also unaffected by compression.
Conclusions::
These results suggest that a JPEG compression set to maximum quality has no effect on identification or area measurements of lesion components in the grading of exudative lesions in a reading centre setting. As none of the angiograms had an occult component, the effect of compression on grading occult CNV could not be determined. Confirmation of these findings in a larger sample with a wider spectrum of exudative AMD lesion subtypes has begun. If found to be robust, these findings have important implications for image transfer, retrieval and storage at Reading Centres.
Keywords: imaging/image analysis: clinical • age-related macular degeneration • image processing