An automated system for the detection of AMD-related drusen in retinal fundus images

Abstract

 

Purpose

 

To assess the performance of an automatic system for the detection of the presence of drusen in early age-related macular degeneration (AMD) in retinal fundus images.

 

Methods

 

We tested the performance of a proposed system shown in Fig. 1 for the automatic detection of the presence of drusen in early AMD using a sample of images from the Singapore Malay Eye Study. The proposed system first performs detection of the optic disk, which is used as a reference point for the macula. Automatic macula centre localization is carried out using a seeded mode tracking approach. A 2 optic disk diameter radius around the detected macula is then extracted. This region is subject to dense sampling and semantic feature extraction to form a hierarchical representation. The presence of drusen in the representation is classified using a support vector machine.

 

Results

 

The sample of images consisted of 253 normal eyes and 94 eyes with drusen. The drusen images had been clinically verified for early AMD. Using our proposed system, the macula centre was successfully automatically located in 345 images. The area under the receiver operating characteristic curve for our proposed system was calculated to be 0.84, with an average running time of 30 seconds per image.

 

Conclusions

 

An automatic system to detect the presence of early AMD-related drusen was tested. Experimental results are promising and encouraging for the further evaluation the system as an tool for the early screening and detection of AMD.

 

 

Fig 1. Framework for the proposed automatic drusen detection system

 

Keywords: 549 image processing • 550 imaging/image analysis: clinical • 585 macula/fovea