Paper Info
Title
Diagnostic Information based Super-Resolution of Retinal Optical Coherence Tomography Images
Abstract
This paper presents a novel diagnostic information based super-resolution (SR) method for the optical coherence tomography (OCT) images of the human retina. First, the edge preserving, guided image filtering is applied on the low resolution (LR) OCT image to remove the speckle noises. Then an appropriate section is selected for observation. The diagnostic relevance of the selected section is evaluated using the proposed features followed by classification. The histogram of oriented gradients (HoG) is computed on the selected OCT section to encode clinically relevant information. A probabilistic framework based on the Gaussian mixture model (GMM) is applied on the HoG features to obtain the Gaussian posteriorgram features. The posteriorgram features minimize the intra-class variability and maximize the inter-class variability. These features are applied to the k-nearest neighbour (k-NN) classifier for section classification. Based on the clinical relevance of the selected section, highly accurate learning based SR method or the simple bicubic interpolation is applied to the selected section. The proposed method achieves a sensitivity, specificity and accuracy of 74.93%, 97.88% and 92.57% during classification. The method obtains a peak signal to noise ratio (PSNR) of 27.47 dB during the SR process.
Year
DOI
Venue
2018
10.1109/SPCOM.2018.8724447
2018 International Conference on Signal Processing and Communications (SPCOM)
Keywords
Field
DocType
Retina,Interpolation,Feature extraction,Spread spectrum communication,Speckle,Training,Dictionaries
Computer vision,Optical coherence tomography,Computer science,Artificial intelligence,Retinal,Superresolution
Conference
ISSN
ISBN
Citations 
2474-9168
978-1-5386-3821-7
0
PageRank 
References 
Authors
0.34
0
3
Name
Order
Citations
PageRank
Vineeta Das152.56
S. Dandapat226128.51
Prabin Kumar Bora352.22