Baidaa Al-Bander

Multiscale sequential convolutional neural networks for simultaneous detection of fovea and optic disc

Abstract Detecting the locations of the optic disc and fovea is a crucial task towards developing automatic diagnosis and screening tools for retinal disease. We propose to address this challenging problem by investigating the potential of applying deep learning techniques to this field. In the proposed method, simultaneous detection of the centers of the fovea and the optic disc (OD) from color fundus images is considered as a regression problem. A deep multiscale sequential convolutional neural network (CNN) is designed and trained. The publically available MESSIDOR and Kaggle datasets are used to train the network and evaluate its performance. The centers of the fovea and the OD in each image were marked by expert graders as the ground truth. The proposed method achieves an accuracy of 97%, 96.7% for the detection of the OD center and 96.6%, 95.6% for the detection of the foveal center of the MESSIDOR and Kaggle test sets respectively. Our promising results demonstrate the excellent performance of the proposed CNNs in simultaneously detecting the centers of both the fovea and OD without human intervention or handcrafted features. Moreover, we can localize the landmarks of an image in 0.007s. This approach could be used as a crucial part of automated diagnosis systems for better management of eye disease.

Automated glaucoma diagnosis using deep learning approach

Glaucoma is one of the common causes of blindness worldwide. It leads to deterioration in vision and quality of life if it is not cured early. This paper addresses the feasibility of developing an automatic feature learning technique for detecting glaucoma in colored retinal fundus images using a deep learning method. A fully automated system based on convolutional neural network (CNN) is developed to distinguish between normal and glaucomatous patterns for diagnostic decisions. Unlike traditional methods where the optic disc features are handcrafted, the features are extracted automatically from the raw images by CNN and fed to the SVM classifier to classify the images into normal or abnormal. We demonstrate an accuracy, specificity and sensitivity of 88.2%, 90.8%, and 85%, respectively which compared favorably to the-state-of-the-art but at considerably lower computational cost. The obtained preliminary results clearly demonstrate that the proposed deep learning method is promising in automatic diagnosis of glaucoma.

Automatic Detection and Distinction of Retinal Vessel Bifurcations and Crossings in Colour Fundus Photography

The analysis of retinal blood vessels present in fundus images, and the addressing of problems such as blood clot location, is important to undertake accurate and appropriate treatment of the vessels. Such tasks are hampered by the challenge of accurately tracing back problems along vessels to their source. This is due to the unresolved issue of distinguishing automatically between vessel bifurcations and vessel crossings in colour fundus photographs. In this paper, we present a new technique for addressing this problem using a convolutional neural network approach to firstly locate vessel bifurcations and crossings and then to classifying them as either bifurcations or crossings. Our method achieves high accuracies for junction detection and classification on the DRIVE dataset and we show further validation on an unseen dataset from which no data has been used for training. Combined with work in automated segmentation, this method has the potential to facilitate: reconstruction of vessel topography, classification of veins and arteries and automated localisation of blood clots and other disease symptoms leading to improved management of eye disease.

Improving the Resolution of Retinal OCT with Deep Learning

In medical imaging, high-resolution can be crucial for identifying pathologies and subtle changes in tissue structure. However, in many scenarios, achieving high image resolution can be limited by physics or available technology. In this paper, we aim to develop an automatic and fast approach to increasing the resolution of Optical Coherence Tomography (OCT) images using the data available, without any additional information or repeated scans. We adapt a fully connected deep learning network for the super-resolution task, allowing multi-scale similarity to be considered, and create a training and testing set of more than 40,000 sample patches from retinal OCT data. Testing our model, we achieve an impressive root mean squared error of 5.847 and peak signal-to-noise ratio (PSNR) of 33.28 dB averaged over 8282 samples. This represents a mean improvement in PSNR of 3.2 dB over nearest neighbour and 1.4 dB over bilinear interpolation. The results achieved so far improve over commonly used fast techniques for increasing resolution and are very encouraging for further development towards fast OCT super-resolution. The ability to increase quickly the resolution of OCT as well as other medical images has the potential to impact significantly on medical imaging at point of care, allowing significant small details to be revealed efficiently and accurately for inspection by clinicians and graders and facilitating earlier and more accurate diagnosis of disease.

Fast blur detection and parametric deconvolution of retinal fundus images

Blur is a significant problem in medical imaging which can hinder diagnosis and prevent further automated or manual processing. The problem of restoring an image from blur degradation remains a challenging task in image processing. Semi-blind deblurring is a useful technique which may be developed to restore the underlying sharp image given some assumed or known information about the cause of degradation. Existing models assume that the blur is of a particular type, such as Gaussian, and do not allow for the approximation of images corrupted by other blur types which are not easily incorporated into deblurring frameworks. We present an automated approach to image deconvolution which assumes that the cause of blur belongs to a set of common types. We develop a hierarchical approach with convolutional neural networks (CNNs) to distinguish between blur types, achieving an accuracy of 0.96 across a test set of 900 images, and to determine the blur strength, achieving accuracy of 0.77 across 1500 test images. Given this, we are able to reconstruct the underlying image to mean ISNR of 7.53.