Retinal Disease Screening through Learning Dictionaries

Project facts

Project promoter:
Polytechnic University of Valencia
Project Number:
ES07-0032
Target groups
Researchers or scientists
Status:
Completed
Initial project cost:
€2,879
Final project cost:
€2,879
From EEA Grants:
€ 2,879
The project is carried out in:
Spain

Description

Retinal vasculature is able to indicate the status of other vessels of the human body. The main goal of this project is to develop a system that can detect, through image processing, the specific patterns of some diseases that are among the most common causes of blindness in current society. Therefore, the detection of these lesions is crucial for early diagnosis of the pathologies under study. We believe that all of them can be detected by means of texture analysis in which the host group has a significant experience. The biomedical data analysis group of the Department of Computer Science and Electrical Engineering (IDE) at University of Stavanger (UiS) has long experience in texture analysis, both image textures in general and for classifying different biomedical images. This expertise can be useful for analyzing and classifying the different possible diseases that can be seen from fundus images. For UiS this collaboration will provide insight into interesting new applications, and for the Polytechnic University of Valencia it will provide insight into new techniques.

Summary of project results

The main goal of the project was to develop a system that could detect, through image processing, the specific patterns of some diseases that are among the most common causes of blindness in current society: diabetic retinopathy (DR) and age-related macular degeneration (AMD), in order to be able to carry out an automatic screening of these diseases. Image analysis algorithms were studied and implemented that maximize detection probability and minimize false positives. Texture descriptors and classifiers were tested to investigate the discrimination capabilities in the texture of fundus images, the main focus lying in exploring Local Binary Patterns as a descriptor for fundus images. The system was validated using databases of healthy and pathological patients analysed by expert ophtalmologists. Experiments made obtained excellent results in discriminating AMD pathologies, and between good and excellent those detecting DR. Project results were presented to groups of the Department of Computer Science and Electrical Engineering of the University of Stavanger, and a paper to be sent to a peer reviewed journal is in preparation. Thanks to have provided insight into a a new technique, different applications, where this technique could be applied, were identified, for example to detect the optic cup or to determinate the age through the analysis of facial images.

Summary of bilateral results

The donor partner, University of Stavanger, has a long experience in texture analysis, both image textures in general and for classifying different biomedical images. This expertise was very useful for analysing and classifying the different possible diseases that can be seen from the fundus images. Joint cooperation led to reliable experiments and results. Senior researcher from Technical University of Valencia met several experts and groups of the department at host institution who are involved with Kjersti Engan, partner of the activity, in a project whose goal is the detection of anomalies in newborns related with the Spanish researcher’s background in motion detection and motion analysis. A paper has been prepared by three researchers including host and Spanish researcher, on “Inpainting of retinal images”, and meetings have been held with department director in order to establish possibilities of interchange for master and PhD students and other colleagues from the Spanish university.