Browsing by Author "Hana Mekonen"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Glaucoma Detection Using Macular OCT Images Based on Deep Convolutional Neural Networks
    (Addis Ababa University, 2023-03) Hana Mekonen; Dawit Assefa (PhD); Tesfaye Tadesse (PhD) Clinical Advisor
    Glaucoma is a major public health problem as it is the second leading cause of blindness after cataract. Since vision loss due to glaucoma can't be recovered, an early, reliable diagnosis is desirable. Although complete eye examination is recommended for assessment of both structural and functional states of the disease, glaucomatous structural changes precede functional changes. For instance, many studies reported that 25-30% of ganglion cell loss precedes the manifestation of visual field defect and loss of retinal nerve fiber layers (RNFL) occurs approximately six years before any detectable visual field defect. Therefore, the early diagnosis of glaucoma relies on the detection of these structural changes. Recently, classifying glaucomatous images taken from different modalities based on Deep Learning (DL) is increasingly being studied. Most of the researchers, however, relied on images generated from a fundus camera and others on OCT scans taken from the optic nerve head (ONH). Various others relied on specific information derived from the OCT machine itself including thickness and deviation maps of macular and ONH scans, and en-face images. However, the glaucomatous eye can be more effectively detected by analyzing the degeneration of the ganglion cell complex (GCCs) by using original OCT complete scans as input. The current thesis study used deep segmentation models to extract the GCC region which is composed of the retinal nerve fiber layer and ganglion cells with the inner plexiform layer. The study also used Convolutional Neural Network (CNN) based classifiers for detecting glaucomatous pathologies by paying attention to the GCC region of the macula Spectral Domain Optical Coherence Tomograpgy (SD-OCT) scans. The data set utilized for training and validation of the models composed of 1,262 locally acquired macula SD-OCT B-scans (431 non-glaucomatous and 830 glaucomatous) from four different regions of the macula: superior outside, inferior outside, inferior inside and central macula regions. Transfer learning was employed for segmentation as well as classifying the dataset. Deep segmentation models, SegNet, PSPNet, and RAG−Netv2, were employed for segmentation and CNN models namely VGG16, VGG19, and ResNet50 were used for classification purpose. SegNet showed the best performance for retinal layer segmentation with 97.89% accuracy, 87% recall, 87.5% f1-score, 88% precision, 89% mean dice coefficient, and 81% mean_IOU. In terms of classification of glaucomatous and normal images, the best accuracy of 94.3% was obtained using VGG16 computed on the superior outside macula region, with 93.3% precision, 91.7% recall, 91.8% f1-score and 91.7% AUC. The study has demonstrated that using GCC aware deep learning model based on macula B-scans show great promises in accurate screening of glaucoma and suggested that incorporating DL technology into macula SD-OCT for glaucoma assessment has the potential to fill some gaps in current practices and clinical workflow.