Automatic Diagnosis of Type of Glaucoma Using Order-One 2D-FBSE-EWT

This paper presents a framework for the automatic classification of Primary Angle-Closure Glaucoma (PACG), Pri-mary Open-Angle Glaucoma (POAG), and secondary Glaucoma from a normal subject. Order-one two-dimensional-Fourier-Bessel series expansion-empirical wavelet transform (2D-FBSE-EWT) based fusion ensemble ResNet-50 model is used in this work. Order-one 2D-FBSE-EWT decomposes the fundus images into sub-images. Subsequently, each sub-image is fed to the ResNet-50 model for extraction of deep features. Thereafter, deep features from each sub-images are ensembled. The ensembled features are then reduced using principal component analysis, and finally the reduced features are fed to a Softmax classifier for classification. Besides this approach, 4-channel, 3-channel (diagonal-wise grouping), and 2-channel (diagonal-wise grouping and neglecting diagonal detail component) sub-image groupings are also compared at 5-fold and 10-fold cross-validation. The 3-channel order-one 2D-FBSE-EWT based fusion ensemble ResNet-50 model provided an accuracy of 93% for the balanced database whereas it was limited to an accuracy of 78.3% for the unbalanced database at 10-fold cross-validation.