Purchase this article with an account.
Jiang LIU, Feng Shou YIN, Zhuo Zhang, Beng Hai Lee, Carol Y. Cheung, Baskaran Mani, Tin Aung, Tien Y. Wong; AGLAIA: A-Levelset based Automatic Cup-to-Disc Ratio Measurement for Glaucoma Diagnosis from Fundus Image. Invest. Ophthalmol. Vis. Sci. 2012;53(14):647.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To introduce AGLAIA, a system to automatically segment the optic cup and the optic disc, which are used to calculate the cup-to-disc ratio (CDR) from 2D digital fundus images for glaucoma diagnosis.
We propose an A-Levelset algorithm to overcome the limitations of traditional level set-based algorithms in image segmentation. A-Levelset cascades level set with ASM (Active Shape Model) algorithm and uses ASM to fine-tune the outcome of the level set algorithm. A-Levelset algorithm is performed in the following three steps:- First Step: Run level set algorithm over the image to obtain the segmented object;- Second Step: Register the level set segmented object with the ASM mean shape;- Third Step: Run ASM with the registered mean shape as the initial contour toget the fine-tuned segmented object. The AGLAIA system uses the A-Levelset algorithm to segment the optic cup and the optic disc from 2D digital fundus images for glaucoma diagnosis.
2616 non-stereoscopic fundus images (306 glaucoma eyes, 2310 non-glaucoma eyes) collected from various clinical and population-based studies were used to evaluate the AGLAIA system. A previously reported system, level set based approach (Wong et al, ARVO 2010) was used as a comparison. The MSE (Mean Square Error) and MAE (Mean Absolute Error) errors for both approaches are shown in Table 1. Table 1 shows that AGLAIA outperforms level set based approach by reducing the CDR measurement MAE from 0.349 to 0.21 and MSE from 0.156 to 0.07. Furthermore, AGLAIA’s MAE is closer to the inter- and intra-observer variability rates of the glaucoma specialists, which stands at 0.2 and 0.15 respectively.
Using an innovative image processing A-Levelset algorithm, AGLAIA system is able to achieve a good performance in CDR measurement. The testing results on an extensive dataset are promising and demonstrate that the AGLAIA system has the potential to be expanded into an automated screening tool for early glaucoma detection.
This PDF is available to Subscribers Only