Abstract:Background and objective: Diabetic retinopathy (DR), which is generally diagnosed by the presence of hemorrhages and hard exudates, is one of the most prevalent causes of visual impairment and blindness. Early detection of hard exudates (HEs) in color fundus photographs can help in preventing such destructive damage. However, this is a challenging task due to high intra-class diversity and high similarity with other structures in the fundus images. Most of the existing methods for detecting HEs are based on characterizing HEs using hand crafted features (HCFs) only, which can not characterize HEs accurately. Deep learning methods are scarce in this domain because they require large-scale sample sets for training which are not generally available for most routine medical imaging research. Methods: To address these challenges, we propose a novel methodology for HE detection using deep convolutional neural network (DCNN) and multi-feature joint representation. Specifically, we present a new optimized mathematical morphological approach that first segments HE candidates accurately. Then, each candidate is characterized using combined features based on deep features with HCFs incorporated, which is implemented by a ridge regression-based feature fusion. This method employs multi-space-based intensity features, geometric features, a gray-level co-occurrence matrix (GLCM)-based texture descriptor, a gray-level size zone matrix (GLSZM)-based texture descriptor to construct HCFs, and a DCNN to automatically learn the deep information of HE. Finally, a random forest is employed to identify the true HEs among candidates. Results: The proposed method is evaluated on two benchmark databases. It obtains an F-score of 0.8929 with an area under curve (AUC) of 0.9644 on the e-optha database and an F-score of 0.9326 with an AUC of 0.9323 on the HEI-MED database. These results demonstrate that our approach outperforms state-of-the-art methods. Our model also proves to be suitable for clinical applications based on private clinical images from a local hospital. Conclusions: This newly proposed method integrates the traditional HCFs and deep features learned from DCNN for detecting HEs. It achieves a new state-of-the-art in both detecting HEs and DR screening. Furthermore, the proposed feature selection and fusion strategy reduces feature dimension and improves HE detection performance. (C) 2020 Published by Elsevier B.V.

Hyper-reflective Foci Segmentation in SD-OCT Retinal Images with Diabetic Retinopathy Using Deep Convolutional Neural Networks

Fast and Automated Hyperreflective Foci Segmentation Based on Image Enhancement and Improved 3D U-Net in SD-OCT Volumes with Diabetic Retinopathy

Fully Automated Segmentation of Hyper-Reflective Foci in OCT Images Using a U-shape Network

Automatic Segmentation of Hyperreflective Foci in OCT Images Based on Lightweight DBR Network.

Automated Segmentation of Hyperreflective Foci in Spectral Domain Optical Coherence Tomography with Diabetic Retinopathy

Multi-Scale and Multi-Branch Convolutional Neural Network for Retinal Image Segmentation

Automatic Segmentation of Hyperreflective Dots Via Focal Priors and Visual Saliency

Automated Quantification of Hyperreflective Foci in SD-OCT With Diabetic Retinopathy

SANet: a Self-Adaptive Network for Hyperreflective Foci Segmentation in Retinal OCT Images

Joint Segmentation of Multi-Class Hyper-Reflective Foci in Retinal Optical Coherence Tomography Images

Fully Automated Segmentation of Hyperreflective Foci in Optical Coherence Tomography Images

Multimodality Analysis of Hyper-reflective Foci and Hard Exudates in Patients with Diabetic Retinopathy

Hyperreflective Foci Enhancement in a Combined Spatial-Transform Domain for SD-OCT Images

FRCNN: A Combination of Fuzzy-rough-set-based Feature Discretization and Convolutional Neural Network for Segmenting Subretinal Fluid Lesions

CLC-Net: Contextual and Local Collaborative Network for Lesion Segmentation in Diabetic Retinopathy Images

Hard Exudate Detection Based on Deep Model Learned Information and Multi-Feature Joint Representation for Diabetic Retinopathy Screening

HACDR-Net: Heterogeneous-Aware Convolutional Network for Diabetic Retinopathy Multi-Lesion Segmentation

Diagnosing Diabetic Retinopathy in OCTA Images Based on Multilevel Information Fusion Using a Deep Learning Framework

Deep-OCTA: Ensemble Deep Learning Approaches for Diabetic Retinopathy Analysis on OCTA Images

Diabetic Retinopathy Detection Based on Deep Convolutional Neural Networks for Localization of Discriminative Regions

Diabetic Retinopathy Lesion Segmentation Using Deep Multi-Scale Framework