The retina of the human eye has the potential to reveal crucial information about several diseases such as diabetes. Several signs such as microaneurysms (MA) manifest themselves as early indicators of Diabetic Retinopathy (DR). Detection of these early signs is important from a clinical perspective in order to suggest appropriate treatment for DR patients. This work aims to improve the detection accuracy of MAs in colour fundus images. While it is expected that multiple images per eye are available in a clinical setup, proposed segmentation algorithms in the literature do not make use of these multiple images. This work introduces a novel MA detection algorithm and a framework for combining spatial and temporal images. A new MA detection method has been proposed which uses a Gaussian matched filter and an ensemble classifier with 70 features for the detection of candidates. The proposed method was evaluated on three public datasets (171 images in total) and has shown improvement in performance for two of the sets when compared to a state-of-the-art method. For lesion-based performance, the proposed method has achieved Retinopathy Online Challenge (ROC) scores of 0.3923, 2109 and 0.1523 in the MESSIDOR, DIARETDB1 and ROC datasets respectively. Based on the ensemble algorithm, a framework for the information combination is developed and consists of image alignment, detecting candidates with likelihood scores, matching candidates from aligned images, and finally fusing the scores from the aligned image pairs. This framework is used to combine information both spatially and temporally. A dataset of 320 images that consists of both spatial and temporal pairs was used for the evaluation. An improvement of performance by 2% is shown after combining spatial information. The framework is applied to temporal image pairs and the results of combining temporal information are analyzed and discussed.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:733455 |
| Date | January 2017 |
| Creators | Habib, Mohamed Mustafa Sayed Ahmed |
| Contributors | Barman, Sarah |
| Publisher | Kingston University |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://eprints.kingston.ac.uk/40455/ |
Page generated in 0.0016 seconds