Automatic Detection of Hard Exudates Shadow Region within Retinal Layers of OCT Images

Table 1

Some reported studies for the detection of hard exudates in DME.

Author	Techniques used	Limitation
Long et al. [11]	Developed an automated detection module using fundus image, the algorithm uses dynamic threshold and fuzzy C-means clustering for hard exudates detection.	The algorithm has few drawbacks due to the poor quality of an image; the detection of the hard exudates includes the bright cotton wool spots, and small hard exudates were ignored.
Srinivasan et al. [9]	The algorithm attempted to classify retinal diseases from OCT images using histograms of oriented gradient (HOG) descriptors consisting of a total of 45 subjects.	The method is limited to classifying and detecting early-stage retinal diseases such as diabetic retinopathy and glaucoma; it needs improvement.
Davoudi et al. [12]	The author used the color fundus camera and OCT images and discussed the characteristics of macula edema and hard exudates using African American patients with type 2 diabetes. In addition, the regression model was used to find an association between serum lipid levels.	There can be misclassification of hyper-reflective foci (as micro hard exudates) in comparison with other retina pathologies.
Lammer et al. [13]	The detection of hard exudates was performed with the help of a fundus image and PS-OCT in patients. The pixel-to-pixel analysis of hard exudates in fundus images was done, and the result was compared with PS-OCT generated report.	The limited dataset was used, and the applied segmentation algorithm needed improvement.
Niu et al. [14]	The findings were established on nonproliferative diabetic retinopathy (NPDR) and proliferative diabetic retinopathy (PDR) patients, where the association of hyper-reflective foci and the presence of hard exudates using SD-OCT image are done.	The limitation of this study includes errors in the segmentation of hard exudates and hyper-reflective foci, resulting in possible inaccuracies in analysis.
Maurya et al. [15]	The automated method used to detect cystoid macula edema and serous retinal detachment in OCT image using gradient information-based segmentation of the retinal boundaries.	The study detected only three types of DME, and results have to be compared using color fundus images for improvement inaccuracy.
Girard et al. [16]	The proposed method improves the quality of an optical coherence tomography (OCT) image. It removes the blood vessel shadow and enhances the contrast of an OCT image by using the techniques of exponentiation and compensation.	In the study, the posterior boundaries of the tissues are still not detected.
Camino et al. [17]	The case study includes twenty healthy volunteers and proposes an algorithm for the detection of shadow from vitreous floaters that recovers the vessel information in the area where the shadow is not severe.	The size of the population is small, and vignetting artifacts at the corner of OCTA images is the concern.