An Interpretable Data-Driven Score for the Assessment of Fundus Images Quality
| Autor: | Youri Peskine, Farida Cheriet, Marie Carole Boucher |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
business.industry
Computer science Image quality media_common.quotation_subject Pattern recognition 02 engineering and technology Diabetic retinopathy Fundus (eye) medicine.disease 01 natural sciences Data-driven 010309 optics 0103 physical sciences Quality Score 0202 electrical engineering electronic engineering information engineering medicine Structure based 020201 artificial intelligence & image processing Quality (business) Artificial intelligence business Grading (tumors) media_common |
| Zdroj: | Lecture Notes in Computer Science ISBN: 9783030505158 ICIAR (2) |
| DOI: | 10.1007/978-3-030-50516-5_28 |
| Popis: | Fundus images are usually used for the diagnosis of ocular pathologies such as diabetic retinopathy. Image quality need however to be sufficient in order to enable grading of the severity of the condition. In this paper, we propose a new method to evaluate the quality of retinal images by computing a score for each image. Images are classified as gradable or ungradable based on this score. First, we use two different U-Net models to segment the macula and the vessels in the original image. We then extract a patch around the macula in the image containing the vessels. Finally, we compute a quality score based on the presence of small vessels in this patch. The score is interpretable as the method is heavily inspired by the way clinicians assess image quality, according to the Scottish Diabetic Retinopathy Grading Scheme. The performances are evaluated on a validation database labeled by a clinician. This method presented a sensitivity of 95% and a specificity of 100% on this database. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|