Purpose
To introduce an automated system that assesses the image quality of the optic disc region and to test its performance in retinal fundus images.
Methods
We developed an Automated Retinal Interest Estimator System to automatically assess the quality of input images as a preprocessing step before passing the images for subsequent analysis. In this way, our system will evaluate the quality of input for computer-aided diagnosis (CAD). The system assesses the quality of an image in three steps. Firstly, a retinal image identification step is used to classify whether the input image is a retinal fundus image. Secondly, a re-evaluation step is performed on non-retinal images for confirmation. Lastly, the optic disc region of interest is located and high level features are extracted to perform quality assessment. The entire process is illustrated in Fig. 1.
Results
The system was tested on 35342 images for retinal image identification step and 370 retinal fundus images for the quality assessment step, using 6200 images and 370 retinal fundus images for training respectively. For retinal image identification, the system achieved an accuracy of 99.54% for the testing data. The area under the operating characteristics curve (AUC) was calculated to be 0.987 for the image quality assessment step.
Conclusions
An automatic system to identify retinal images and assess the image quality of focal region of interest is tested. Experimental result on a large database is promising, showing good potential for the system to be used as a preprocessing tool in computer-aided diagnosis.
Keywords: 549 image processing