Automatic Diagnosis of Different Grades of Diabetic Retinopathy and Diabetic Macular Edema Using 2-D-FBSE-FAWT

Diabetic retinopathy (DR) and diabetic macular edema (DME) are the two most common causes of blindness. The proposed work uses the order-zero and order-one 2-D Fourier–Bessel series expansion-based flexible analytic wavelet transform (2-D-FBSE-FAWT) methods to diagnose different DR and DME grades. The comparison of different subband grouping methods is studied in this article based on the classification performance. Statistical features of the local binary pattern and its rotation-invariant variance are used in the study. For classification purposes, random forest,
$k$
-nearest neighbors, and support vector machine are used. The Indian diabetic retinopathy image dataset (IDRiD) and Messidor database are used to evaluate the proposed method. Average accuracy (
$\text {ACY}_{\mathrm{ avg}}$
) measures have been considered for performance comparison. For IDRiD,
$\text {ACY}_{\mathrm{ avg}}$
obtained for DR and DME is 0.955 and 0.965, respectively. For the Messidor database,
$\text {ACY}_{\mathrm{ avg}}$
is 0.975 for DR and 0.985 for DME. For diagnosis of different grades of DR and DME, the best tuned parameters for designing order-zero 2-D-FBSE-FAWT filter banks are level of decomposition = 3,
$Q$
-factor = 3, dilation factor = 0.5, and redundancy = 1.