Search

Microaneurysms (MAs) are the early indications of diabetic retinopathy (DR), which may result in total visual loss. MAs detection is an exciting work due to its small, darkish color and subtle nature. The automatic detection and categorization of MAs in retinal fundus images using a multi‐scale approach based on Deep Neural Networks and Neighborhood Analysis is proposed in this research article. MAs segmentation, classification, and preprocessing comprise the three steps of the proposed technique. To extract the multi‐scale MA features, the auto‐weight dilated convolutional unit (AD) is specifically used for convolutional feature maps. To fuse convolutional feature maps in encoding layers, the AD unit used a learnable set of parameters. An efficient architecture for feature extraction during the encoding step is incorporated into the AD unit. We integrated the AD unit with the benefits of the U‐Net network for deep and shallow features. Additionally, in order to improve the suggested model and produce the final derivation, we developed a novel optimization approach. After that, neighborhood analysis is performed to name the Micro‐aneurysm because the lesion is actually a collection of independent little images rather than the entire image. The classification accuracy of the proposed method for the three different data sets such as MESSIDOR, online retinal fundus image database for glaucoma analysis (ORIGA), and RIM‐ONE‐R1, is 99.28%, 98.95%, and 98.76% respectively. The results show a good performance of the proposed model against the other analyzed procedure. [ABSTRACT FROM AUTHOR]

This article analyzes the sensitivity of deep learning methods for ocular fundus segmentation. We use an empirical methodology based on non‐adversarial perturbed datasets. The research is motivated by mass screening and self‐administered tests in which AI methods are needed and may be given images with focus issues. These substandard pictures are simulated using blurring algorithms of varying designs and kernel sizes which are subjected to a test of inter‐network and inter‐dataset sensitivity using Gaussian noise as a control. We test for simulated defocusing, motion blur, and Gaussian noise. The DRIVE, CHASE, and STARE datasets were used to generate 441 synthetic datasets for testing. Analysis of the resultant n = 37,888 sample has identified anomalies illuminating failure modes of state‐of‐the‐art ocular segmentation models. Data analysis led us to conclude that accuracy is a poor measure of sensitivity to input change and that even moderate levels of blur have dramatic effects on said sensitivity. We further concluded that architectures show larger variations in sensitivity to blur which do not match either their reported or measured performance on unblurred test datasets. Our analysis has attributed at least a part of the problem to overfitting non‐essential input dataset features and resolution sensitivity. [ABSTRACT FROM AUTHOR]

A novel two‐branched deep convolutional (TWEEC) network is developed for computer‐aided glaucoma diagnosis. The TWEEC network is designed to simultaneously extract anatomical information related to the optic disc and surrounding blood vessels which are the retinal structures most affected by glaucoma progression. The spatial retinal images and wavelet approximation subbands are compared as inputs to the proposed network. TWEEC's performance is compared to three implemented convolutional networks, one of which employs transfer learning. Experiments showed that the introduced TWEEC network achieved accuracies of 98.78% and 96.34% for the spatial and wavelet inputs, respectively, by that outperforming the other three deep networks by 8‐15%. This work paves the way for the development of efficient deep learning based computer‐aided glaucoma diagnosis tools. Moreover, the present study illustrates that considering specific wavelet subbands for the training of convolutional networks can result in reliable performance with the advantage of reduced overall network training time. [ABSTRACT FROM AUTHOR]

Automatic detection of maculopathy disease is a very important step to achieve high‐accuracy results for the early discovery of the disease to help ophthalmologists to treat patients. Manual detection of diabetic maculopathy needs much effort and time from ophthalmologists. Detection of exudates from retinal images is applied for the maculopathy disease diagnosis. The first proposed framework in this paper for retinal image classification begins with fuzzy preprocessing in order to improve the original image to enhance the contrast between the objects and the background. After that, image segmentation is performed through binarization of the image to extract both blood vessels and the optic disc and then remove them from the original image. A gradient process is performed on the retinal image after this removal process for discrimination between normal and abnormal cases. Histogram of the gradients is estimated, and consequently the cumulative histogram of gradients is obtained and compared with a threshold cumulative histogram at certain bins. To determine the threshold cumulative histogram, cumulative histograms of images with exudates and images without exudates are obtained and averaged for each type, and the threshold cumulative histogram is set as the average of both cumulative histograms. Certain histogram bins are selected and thresholded according to the estimated threshold cumulative histogram, and the results are used for retinal image classification. In the second framework in this paper, a Convolutional Neural Network (CNN) is utilized to classify normal and abnormal cases. [ABSTRACT FROM AUTHOR]

Limited labelled data is a challenge in the field of medical imaging and the need for a large number of them is paramount for the training of machine learning algorithms, as well as measuring the performance of image processing algorithms. The purpose of this study is to construct synthetic and labelled optical coherence tomography (OCT) data to solve the problems of having access to accurately labelled data and evaluating the processing algorithms. In this study, a modified active shape model is used which considers the anatomical features of available images such as the number and thickness of the layers as well as their associated brightness, the location of retinal blood vessels and shadow information with respect to speckle noise. The algorithm is also able to provide different data sets with the varying noise level. The validity of the proposed method for the synthesis of retinal images is measured by two methods (qualitative assessment and quantitative analysis). [ABSTRACT FROM AUTHOR]

Summary: Image segmentation plays a crucial role in many image processing and understanding applications. Despite the huge number of proposed image segmentation techniques, accurate segmentation remains a significant challenge in image analysis. This article investigates the viability of using echo state network (ESN), a biologically inspired recurrent neural network, as features extractor for efficient color image segmentation. First, an ensemble of initial pixel features is extracted from the original images and injected into the ESN reservoir. Second, the internal activations of the reservoir neurons are used as new pixel features. Third, the new features are classified using a feed forward neural network as a readout layer for the ESN. The quality of the pixel features produced by the ESN is evaluated through extensive series of experiments conducted on real world image datasets. The optimal operating range of different ESN setup parameters for producing competitive quality features is identified. The performance of the proposed ESN‐based framework is also evaluated on a domain‐specific application, namely, blood vessel segmentation in retinal images where experiments are conducted on the widely used digital retinal images for vessel extraction (DRIVE) dataset. The obtained results demonstrate that the proposed method outperforms state‐of‐the‐art general segmentation techniques in terms of performance with an F‐score of 0.92 ± 0.003 on the segmentation evaluation dataset. In addition, the proposed method achieves a comparable segmentation accuracy (0.9470) comparing with reported techniques of segmentation of blood vessels in images of retina and outperform them in terms of processing time. The average time required by our technique to segment one retinal image from DRIVE dataset is 8 seconds. Furthermore, empirically derived guidelines are proposed for adequately setting the ESN parameters for effective color image segmentation. [ABSTRACT FROM AUTHOR]

As a weakly supervised learning technique, multiple instance learning (MIL) has shown an advantage over supervised learning methods for automatic detection of diabetic retinopathy (DR): only the image‐level annotation is needed to achieve both detection of DR images and DR lesions, making more graded and de‐identified retinal images available for learning. However, the performance of existing studies on this technique is limited by the use of handcrafted features. The authors propose a deep MIL method for DR detection, which jointly learns features and classifiers from data and achieves a significant improvement on detecting DR images and their inside lesions. Specifically, a pre‐trained convolutional neural network is adapted to achieve the patch‐level DR estimation, and then global aggregation is used to make the classification of DR images. Further, the authors propose an end‐to‐end multi‐scale scheme to better deal with the irregular DR lesions. For detection of DR images, they achieve an area under the ROC curve of 0.925 on a subset of a Kaggle dataset, and 0.960 on Messidor. For detection of DR lesions, they achieve an F1‐score of 0.924 with sensitivity 0.995 and precision 0.863 on DIARETDB1 using the connected component‐level validation. [ABSTRACT FROM AUTHOR]

Precise localisation of an optic disk (OD) in the retinal images is one of the most important problems in the ophthalmic image processing. Although a considerable progress has been made towards a computerised solution of the problem, the numerical algorithms often fail on retinal images characterised by poor quality. Therefore, the authors propose a new method suitable for low‐quality images based on exploiting the convergence of the blood vessels to the OD. The novelty of the proposed techniques includes clustering the vessels endowed with a novel correction procedure and the vessel transform (VT) which measures the distance to the main clusters. The algorithm is integrated into the scale‐space (SS) analysis to detect the boundary of the OD. The integrated method is called SS algorithm with VT (SSVT). SSVT has been tested on retinal images from two databases with fair and poor images against the fuzzy convergence (FC) method and a modification of the circular transform proposed by Lu. The absolute improvement on sensitivity of SSVT against FC and Lu's are up to 12.37% and 8.18%. Bigger improvements of SSVT in terms of positive predictive value are up to 37.46% and 30.84% against FC and Lu's, respectively. [ABSTRACT FROM AUTHOR]

Purpose In this paper, the aim is to demonstrate enhanced processing of sequences of fundus images obtained using a commercial AO flood illumination system. The purpose of the work is to (1) correct for uneven illumination at the retina (2) automatically select the best quality images and (3) precisely register the best images. Methods Adaptive optics corrected retinal images are pre-processed to correct uneven illumination using different methods; subtracting or dividing by the average filtered image, homomorphic filtering and a wavelet based approach. These images are evaluated to measure the image quality using various parameters, including sharpness, variance, power spectrum kurtosis and contrast. We have carried out the registration in two stages; a coarse stage using cross-correlation followed by fine registration using two approaches; parabolic interpolation on the peak of the cross-correlation and maximum-likelihood estimation. The angle of rotation of the images is measured using a combination of peak tracking and Procrustes transformation. Results We have found that a wavelet approach ( Daubechies 4 wavelet at 6th level decomposition) provides good illumination correction with clear improvement in image sharpness and contrast. The assessment of image quality using a ' Designer metric' works well when compared to visual evaluation, although it is highly correlated with other metrics. In image registration, sub-pixel translation measured using parabolic interpolation on the peak of the cross-correlation function and maximum-likelihood estimation are found to give very similar results ( RMS difference 0.047 pixels). We have confirmed that correcting rotation of the images provides a significant improvement, especially at the edges of the image. We observed that selecting the better quality frames (e.g. best 75% images) for image registration gives improved resolution, at the expense of poorer signal-to-noise. The sharpness map of the registered and de-rotated images shows increased sharpness over most of the field of view. Conclusion Adaptive optics assisted images of the cone photoreceptors can be better pre-processed using a wavelet approach. These images can be assessed for image quality using a ' Designer Metric'. Two-stage image registration including correcting for rotation significantly improves the final image contrast and sharpness. [ABSTRACT FROM AUTHOR]

In this paper, a partial supervision strategy for a recently developed clustering algorithm, the nearest neighbour clustering algorithm (NNCA), is proposed. The proposed method (NNCA-PS) offers classification capability with a smaller amount of a priori knowledge, where a small number of data objects from the entire data set are used as labelled objects to guide the clustering process towards a better search space. Experimental results show that NNCA-PS gives promising results of 89% sensitivity at 95% specificity when used to segment retinal blood vessels, and a maximum classification accuracy of 99.5% with 97.2% average accuracy when applied to a breast cancer data set. Comparisons with other methods indicate the robustness of the proposed method in classification. Additionally, experiments on parallel environments indicate the suitability and scalability of NNCA-PS in handling larger data sets. [ABSTRACT FROM AUTHOR]