Wavelet Image Scattering Based Glaucoma Detection

Abstract The ever-growing need for cheap, simple, fast, and reliable healthcare solutions spurred a lot of research activities that are aimed at the reliable deployment of artificial intelligence in the medical fields. However, this has proved to be a daunting task especially when looking to make automated diagnoses using biomedical image data. Biomedical image data have complex patterns which human experts find very hard to comprehend. This paper studies the efficacy of wavelet scattering features from retinal fundus images for automatic glaucoma diagnosis/detection. The influence of wavelet image scattering network parameter settings as well as 2-D channel image representation type on the detection correctness is also examined. The wavelet image scattering network developed in the Matlab environment was used on the RIM-ONE DL image dataset to execute the scattering decomposition and obtain the scattering coefficients. Features from the coefficients were then used to build simple classification algorithms. Maximum detection correctness of 98% was achieved on the held-out test set. Results showed that detection correctness is highly sensitive to scattering network parameter settings and 2-D channel representation type. A superficial comparison of the classification results obtained from the proposed method and those obtained using a convolutional neural network underscores the potentiality of our method.