| Abstrakt: |
This paper presents early-automated glaucoma detection algorithm by extracting early diagnostic parameters, namely, parapapillary atrophy and Cup to Disc ratio using DIP. Glaucoma is optic neuropathy chronic eye diseases, which is progressed as intraocular pressure inside eyeball increases. Proposed method mainly consists of five stages, namely ROI extraction, pre-processing, diagnostic parameter extraction, feature extraction, and classification. First, optic disc region is extracted from fundus image because diagnostic parameters lie around OD. ROI is extracted using the horizontal and vertical edge of vascular tree in OD region. Diagnostic parameters such as CDR and PPA are extracted from ROI region. CDR is calculated using K-means clustering and polar transform. PPA is earliest sign of glaucoma. In many cases, occurrence of PPA before the increased CDR value gives indication of early glaucoma. PPA detection and extraction are accomplished using polar transform method and direct least-square fitting algorithm of an ellipse. Optimal categorization of PPA area and residue area is done by K-means clustering. Features of diagnostic parameters such as CDR and area of PPA are extracted for classification model. Features of fundus images are taken as input and labels as output for training parameters of SVM. SVM model is trained with five-fold validation on 141 images, which consists of healthy and glaucoma due to PPA and CDR. Thirty-nine images are used for testing of SVM model and sensitivity, specificity, and accuracy values are 100%, 88.2%, and 97.3% respectively. Diagnostic results obtained by proposed algorithm, about early glaucoma presence is further validated by ophthalmologist. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink