An Evaluation for Automated Cup-Disc-Ratio Assessment System for Digital Fundus Images

Abstract

Purpose: : To evaluate the performance of an automatic system for cup-to-disc ratio (CDR) measurement from non-stereoscopic digital fundus images

Methods: : We developed an enhanced version of our previously reported automatic system (Wong et al, ARVO 2009). The vertical optic cup and disc heights were detected from points in the inferior (5 o'clock to 7 o'clock) and superior zones (11 o'clock to 1 o'clock). The system is tested on a study sample of 71 retinal fundus images (3072 x 2048, Field 2) from the Singapore Malay Eye Study (SiMES) study. Clinical CDR was defined by an ophthalmologist.

Results: : The mean clinical CDR was 0.61 and the mean CDR from the automated system was 0.63. Compared with the clinical reference, the automatic system had a vertical disc error of 118µm (95% CI , 93.8 - 143.1) and a vertical cup error of 146µm (95% CI, 117.5 - 174.4). The CDR of the automatic system had an unsigned error of 0.08 CDR units (95% CI, 0.061 - 0.090), with 94.4% of the images having a CDR error of less then 0.2 CDR units.

Conclusions: : The enhanced automated system has reduced CDR errors due to the presence of blood vessels within the optic nerve head. This system reports a CDR error within the intra-observer variability of 0.2 CDR in 94.4% of the images. The results are promising for further testing to be done on an expanded data set to assess its performance as an automated screening tool for early glaucoma detection.

Keywords: image processing • imaging/image analysis: clinical