NUST Institutional Repository

AUTOMATIC DETECTION OF PATHOLOGICAL MYOPIA USING DEEP LEARNING

Show simple item record

dc.contributor.author NAMRA RAUF, Supervised By Dr Syed Omer Gilani
dc.date.accessioned 2020-11-02T11:26:27Z
dc.date.available 2020-11-02T11:26:27Z
dc.date.issued 2019
dc.identifier.uri http://10.250.8.41:8080/xmlui/handle/123456789/8428
dc.description.abstract Pathological myopia (PM) is the severe case of myopia i.e. near sightedness. PM is also called degenerative myopia because it ultimately leads to blindness. Certain myopia-specific pathologies occurs at the eye’s posterior in PM i.e. Foster-Fuchs spot, Cystoid degeneration, Liquefaction, Macular degeneration, Vitreous opacities, Weiss’s reflex, Posterior staphyloma etc. Review of the literature revealed that less amount of work is done on the automatic detection of PM as compared to other retinal pathologies i.e. glaucoma. This research project is aimed at developing a deep learning (DL) approach for automatic detection of PM based on fundus images. Two deep learning techniques (subfield of Machine learning) have been utilized; ANN and CNN, with a particular focus on the latter. In initial step, an already available artificial neural network (ANN) model is tested for the given fundus images in Matlab. The average accuracy obtained with ANN is around 50%. In second step, a convolutional neural network (CNN) model is developed in Spyder. The fundus images are first preprocessed and then fed to the designed CNN model. The CNN model automatically extracts the features from the input images and classifies them i.e. normal image or pathological myopic image. The best performing CNN model achieved an AUC score of 0.9845. The best validation loss obtained is 0.15. en_US
dc.language.iso en_US en_US
dc.publisher SMME-NUST en_US
dc.relation.ispartofseries SMME-TH-413;
dc.subject Pathological myopia (PM), Machine learning (ML), Deep learning (DL), ANN, CNN, AUC en_US
dc.title AUTOMATIC DETECTION OF PATHOLOGICAL MYOPIA USING DEEP LEARNING en_US
dc.type Thesis en_US


Files in this item

This item appears in the following Collection(s)

  • MS [368]

Show simple item record

Search DSpace


Advanced Search

Browse

My Account