No abstract

In the field of medical image diagnostics, automatic glaucoma classification is a daunting task that requires carefully crafted features (Machine Learning) or tuning hyperparameters over a huge search space (Deep Learning). In this paper, we proposed a wavelet scattering based approach that extracts features from fundus images at different sub-stages, reduces the feature space using principal component analysis (PCA) and represents the classifying features in terms of projection loss along the subspace for each class. This simple yet elegant approach yields fast-paced results with great interpretability, that deep learning-based approaches lack, and excels in most performance metrics when compared to other machine learning based methods.