Abstract

Automatic disease detection and classification have been attracting much interest. High performance is critical in adoption of such systems, which generally rely on training with a wide variety of annotated data. Availability of such varied annotated data in medical imaging is very scarce. Synthetic data generation is a promising solution to address this problem. We propose a novel method, based on generative adversarial networks (GAN), to generate images with lesions such that the overall severity level can be controlled. We demonstrate the reliability of the generated synthetic images independently as well as by training a computer aided diagnosis (CAD) system with the generated data. We showcase this approach for heamorrhage detection in retinal images with 4 levels of severity. Quantitative assessment results show that the generated synthetic images are very close to the real data. Haemorrhage detection was found to improve with inclusion of synthetic data in the training set with improvements in sensitivity ranging from 20% to 27% over training with just expert marked data.

Abstract

The level set based deformable models (LDM) are commonly used for medical image segmentation. However, they rely on a handcrafted curve evolution velocity that needs to be adapted for each segmentation task. The Convolutional Neural Networks (CNN) address this issue by learning robust features in a supervised end-to-end manner. However, CNNs employ millions of network parameters, which require a large amount of data during training to prevent over-fitting and increases the memory requirement and computation time during testing. Moreover, since CNNs pose segmentation as a region-based pixel labeling, they cannot explicitly model the high-level dependencies between the points on the object boundary to preserve its overall shape, smoothness or the regional homogeneity within and outside the boundary. We present a Recurrent Neural Network based solution called the RACE-net to address the above issues. RACE-net models a generalized LDM evolving under a constant and mean curvature velocity. At each time-step, the curve evolution velocities are approximated using a feed-forward architecture inspired by the multiscale image pyramid. RACE-net allows the curve evolution velocities to be learned in an end-to-end manner while minimizing the number of network parameters, computation time, and memory requirements. The RACEnet was validated on three different segmentation tasks: optic disc and cup in color fundus images, cell nuclei in histopathological images, and the left atrium in cardiac MRI volumes. Assessment on public datasets was seen to yield high Dice values between 0.87 and 0.97, which illustrates its utility as a generic, off-the-shelf architecture for biomedical segmentation.

Abstract

Feature-based registration has been popular with a variety of features ranging from voxel intensity to Self-Similarity Context (SSC). In this paper, we examine the question of how features learnt using various Deep Learning (DL) frameworks can be used for deformable registration and whether this feature learning is necessary or not. We investigate the use of features learned by different DL methods in the current state-ofthe-art discrete registration framework and analyze its performance on 2 publicly available datasets. We draw insights about the type of DL framework useful for feature learning. We consider the impact, if any, of the complexity of different DL models and brain parcellation methods on the performance of discrete registration. Our results indicate that the registration performance with DL features and SSC are comparable and stable across datasets whereas this does not hold for low level features. This shows that when handcrafted features are designed based on good insights into the problem at hand, they perform better or are comparable to features learnt using deep learning framework.

Abstract

Background and Objective: Accurate segmentation of the intra-retinal tissue layers in Optical Coherence Tomography (OCT) images plays an important role in the diagnosis and treatment of ocular diseases such as Age-Related Macular Degeneration (AMD) and Diabetic Macular Edema (DME). The existing energy minimization based methods employ multiple, manually handcrafted cost terms and often fail in the presence of pathologies. In this work, we eliminate the need to handcraft the energy by learning it from training images in an end-to-end manner. Our method can be easily adapted to pathologies by re-training it on an appropriate dataset. Methods: We propose a Conditional Random Field (CRF) framework for the joint multi-layer segmentation of OCT B-scans. The appearance of each retinal layer and boundary is modeled by two convolutional filter banks and the shape priors are modeled using Gaussian distributions. The total CRF energy is linearly parameterized to allow a joint, end-to-end training by employing the Structured Support Vector Machine.

Abstract

Purpose: To compare the diagnostic ability of optical coherence tomography angiography (OCT-A) derived radial peripapillary capillary (RPC) measured capillary density (CD) and inside the optic nerve head (ONH) CD measurements to differentiate between the normal and primary open angle glaucoma (POAG) eyes. Methods: AngioVue disc OCT-A images were obtained and assessed in 83 eyes of POAG patients and 74 age matched healthy eyes. RPC CD was quantitatively measured in the peripapillary area within 3.45 mm circle diameter around the ONH and inside the ONH in 8 equally divided sectors, using Bar – Selective Combination of Shifted Filter Responses method after the suppressing large vessels. Area under receiver operating characteristic (AUROC) curve was used to assess the diagnostic accuracy of the two scanning regions of CD to differentiate between the normal and POAG eyes. Results: The mean peripapillary RPC density (0.12 ± 0.03) and mean ONH CD (0.09 ± 0.03) were significantly lower in POAG eyes when compared to the normal eyes (RPC CD: 0.17 ± 0.05, p < 0.0001 and ONH CD 0.11 ± 0.02, p = 0.01 respectively). The POAG patients showed 29% reduction in the RPC CD and 19% reduction in the ONH CD when compared to the normal eyes. The AUROC for discriminating between healthy and glaucomatous eyes was 0.784 for mean RPC CD and 0.743 for the mean ONH CD. Conclusions: Diagnostic ability of OCT-A derived peripapillary CD and ONH CD was moderate for differentiating between the normal and glaucomatous eyes. Diagnostic ability of even the best peripapillary average and inferotemporal sector for RPC CD and average and superonasal sector for the ONH CD was moderate.

Abstract

Diabetic retinopathy (DR) is a disease which is widely diagnosed using (colour fundus) images. Efficiency and accuracy are critical in diagnosing DR as lack of timely intervention can lead to irreversible visual impairment. In this paper, we examine strategies for scrutinizing images which affect diagnostic performance of medical practitioners via an eye-tracking study. A total of 56 subjects with 0 to 18 years of experience participated in the study. Every subject was asked to detect DR from 40 images. The findings indicate that practitioners use mainly two types of strategies characterized by either higher dwell duration or longer track length. The main findings of the study are that higher dwell-based strategy led to higher average accuracy (> 85%) in diagnosis, irrespective of the expertise of practitioner; whereas, the average obtained accuracy with a long-track length-based strategy was dependent on the expertise of the practitioner. In the second part of the paper, we use the experimental findings to recommend a scanning strategy for fast and accurate diagnosis of DR that can be potentially used by image readers. This is derived by combining the eyetracking gaze maps of medical experts in a novel manner based on a set of rules. This strategy requires scrutiny of images in a manner which is consistent with spatial preferences found in human perception in general and in the domain of fundus images in particular. The Levenshtein distance-based assessment of gaze patterns also establish the effectiveness of the derived scanning pattern and is thus recommended for image readers.

Abstract

Annotations are critical for machine learning and developing computer aided diagnosis (CAD) algorithms. Good performance of CAD is critical to their adoption, which generally rely on training with a wide variety of annotated data. However, a vast amount of medical data is either unlabeled or annotated only at the image-level. This poses a problem for exploring data driven approaches like deep learning for CAD. In this paper, we propose a novel crowdsourcing and synthetic image generation for training deep neural net-based lesion detection. The noisy nature of crowdsourced annotations is overcome by assigning a reliability factor for crowd subjects based on their performance and requiring region of interest markings from the crowd. A generative adversarial network-based solution is proposed to generate synthetic images with lesions to control the overall severity level of the disease. We demonstrate the reliability of the crowdsourced annotations and synthetic images by presenting a solution for training the deep neural network (DNN) with data drawn from a heterogeneous mixture of annotations. Experimental results obtained for hard exudate detection from retinal images show that training with refined crowdsourced data/synthetic images is effective as detection performance in terms of sensitivity improves by 25%/27% over training with just expert-markings.

Abstract

Optical coherence tomography (OCT) images are corrupted by speckle noise due to the underlying coherence-based strategy. Speckle suppression/removal in OCT images plays a significant role in both manual and automatic detection of diseases, especially in early clinical diagnosis. In this paper, we propose a new method for denoising OCT images based on Convolutional Neural Network by learning common features from unpaired noisy and clean OCT images in an unsupervised, end-to-end manner. The proposed method consists of a combination of two autoencoders with shared encoder layers, which we call as Shared Encoder (SE) architecture. The SE is trained to reconstruct noisy and clean OCT images with respective autoencoders. The denoised OCT image is obtained using a cross-model prediction. The proposed method can be used for denoising OCT images with or without pathology from any scanner. The SE architecture was assessed using public datasets and found to perform better than baseline methods exhibiting a good balance of retaining anatomical integrity and speckle reduction.

Abstract

Domain Specific Search Engines are expected to provide relevant search results. Availability of enormous number of URLs across subdomains improves relevance of domain specific search engines. The current methods for seed URLs can be systematic ensuring representation of subdomains. We propose a fine grained approach for automatic extraction of seed URLs at subdomain level using Wikipedia and Twitter as repositories. A SeedRel metric and a Diversity Index for seed URL relevance are proposed to measure subdomain coverage. We implemented our approach for 'Security - Information and Cyber' domain and identified 34,007 Seed URLs and 400,726 URLs across subdomains. The measured Diversity index value of 2.10 conforms that all subdomains are represented, hence, a relevant 'Security Search Engine' can be built. Our approach also extracted more URLs (seed and child) as compared to existing approaches for URL extraction.

Abstract

Translation Quality Estimation (QE) aims to estimate the quality of an automated machine translation (MT) output without any human intervention or reference translation. With the increasing use of MT systems in various cross-lingual applications, the need and applicability of QE systems is increasing. We study existing approaches and propose multiple neural network approaches for sentence-level QE, with a focus on MT outputs in Indian languages. For this, we also introduce five new datasets for four language pairs: two for English–Gujarati, and one each for English–Hindi, English–Telugu and English–Bengali, which includes one manually post-edited dataset for English– Gujarati. These Indian languages are spoken by around 689M speakers world-wide. We compare results obtained using our proposed models with multiple state-ofthe-art systems including the winning system in the WMT17 shared task on QE and show that our proposed neural model which combines the discriminative power of carefully chosen features with Siamese Convolutional Neural Networks (CNNs) works best for all Indian language datasets.