Correcting Susceptibility Artifacts of MRI Sensors in Brain Scanning: A 3D Anatomy-Guided Deep Learning Approach.

Echo planar imaging (EPI), a fast magnetic resonance imaging technique, is a powerful tool in functional neuroimaging studies. However, susceptibility artifacts, which cause misinterpretations of brain functions, are unavoidable distortions in EPI. This paper proposes an end-to-end deep learning framework, named TS-Net, for susceptibility artifact correction (SAC) in a pair of 3D EPI images with reversed phase-encoding directions. The proposed TS-Net comprises a deep convolutional network to predict a displacement field in three dimensions to overcome the limitation of existing methods, which only estimate the displacement field along the dominant-distortion direction. In the training phase, anatomical T1-weighted images are leveraged to regularize the correction, but they are not required during the inference phase to make TS-Net more flexible for general use. The experimental results show that TS-Net achieves favorable accuracy and speed trade-off when compared with the state-of-the-art SAC methods, i.e., TOPUP, TISAC, and S-Net. The fast inference speed (less than a second) of TS-Net makes real-time SAC during EPI image acquisition feasible and accelerates the medical image-processing pipelines.

Unbalanced Hybrid AOA/RSSI Localization for Simplified Wireless Sensor Networks.

Source positioning using hybrid angle-of-arrival (AOA) estimation and received signal strength indicator (RSSI) is attractive because no synchronization is required among unknown nodes and anchors. Conventionally, hybrid AOA/RSSI localization combines the same number of these measurements to estimate the agents' locations. However, since AOA estimation requires anchors to be equipped with large antenna arrays and complicated signal processing, this conventional combination makes the wireless sensor network (WSN) complicated. This paper proposes an unbalanced integration of the two measurements, called 1AOA/nRSSI, to simplify the WSN. Instead of using many anchors with large antenna arrays, the proposed method only requires one master anchor to provide one AOA estimation, while other anchors are simple single-antenna transceivers. By simply transforming the 1AOA/1RSSI information into two corresponding virtual anchors, the problem of integrating one AOA and N RSSI measurements is solved using the least square and subspace methods. The solutions are then evaluated to characterize the impact of angular and distance measurement errors. Simulation results show that the proposed network achieves the same level of precision as in a fully hybrid nAOA/nRSSI network with a slightly higher number of simple anchors.

An unsupervised deep learning technique for susceptibility artifact correction in reversed phase-encoding EPI images.

Echo planar imaging (EPI) is a fast and non-invasive magnetic resonance imaging technique that supports data acquisition at high spatial and temporal resolutions. However, susceptibility artifacts, which cause the misalignment to the underlying structural image, are unavoidable distortions in EPI. Traditional susceptibility artifact correction (SAC) methods estimate the displacement field by optimizing an objective function that involves one or more pairs of reversed phase-encoding (PE) images. The estimated displacement field is then used to unwarp the distorted images and produce the corrected images. Since this conventional approach is time-consuming, we propose an end-to-end deep learning technique, named S-Net, to correct the susceptibility artifacts the reversed-PE image pair. The proposed S-Net consists of two components: (i) a convolutional neural network to map a reversed-PE image pair to the displacement field; and (ii) a spatial transform unit to unwarp the input images and produce the corrected images. The S-Net is trained using a set of reversed-PE image pairs and an unsupervised loss function, without ground-truth data. For a new image pair of reversed-PE images, the displacement field and corrected images are obtained simultaneously by evaluating the trained S-Net directly. Evaluations on three different datasets demonstrate that S-Net can correct the susceptibility artifacts in the reversed-PE images. Compared with two state-of-the-art SAC methods (TOPUP and TISAC), the proposed S-Net runs significantly faster: 20 times faster than TISAC and 369 times faster than TOPUP, while achieving a similar correction accuracy. Consequently, S-Net accelerates the medical image processing pipelines and makes the real-time correction for MRI scanners feasible. Our proposed technique also opens up a new direction in learning-based SAC.

Hybrid Deep Learning-Gaussian Process Network for Pedestrian Lane Detection in Unstructured Scenes.

Pedestrian lane detection is an important task in many assistive and autonomous navigation systems. This article presents a new approach for pedestrian lane detection in unstructured environments, where the pedestrian lanes can have arbitrary surfaces with no painted markers. In this approach, a hybrid deep learning-Gaussian process (DL-GP) network is proposed to segment a scene image into lane and background regions. The network combines a compact convolutional encoder-decoder net and a powerful nonparametric hierarchical GP classifier. The resulting network with a smaller number of trainable parameters helps mitigate the overfitting problem while maintaining the modeling power. In addition to the segmentation output for each test image, the network also generates a map of uncertainty-a measure that is negatively correlated with the confidence level with which we can trust the segmentation. This measure is important for pedestrian lane-detection applications, since its prediction affects the safety of its users. We also introduce a new data set of 5000 images for training and evaluating the pedestrian lane-detection algorithms. This data set is expected to facilitate research in pedestrian lane detection, especially the application of DL in this area. Evaluated on this data set, the proposed network shows significant performance improvements compared with several existing methods.

Compressive Radar Imaging of Stationary Indoor Targets with Low-rank plus Jointly Sparse and Total Variation Regularizations.

This paper addresses the problem of wall clutter mitigation and image reconstruction for through-wall radar imaging (TWRI) of stationary targets by seeking a model that incorporates low-rank (LR), joint sparsity (JS), and total variation (TV) regularizers. The motivation of the proposed model is that LR regularizer captures the low-dimensional structure of wall clutter; JS guarantees a small fraction of target occupancy and the similarity of sparsity profile among channel images; TV regularizer promotes the spatial continuity of target regions and mitigates background noise. The task of wall clutter mitigation and target image reconstruction is formulated as an optimization problem comprising LR, JS, and TV regularization terms. To handle this problem efficiently, an iterative algorithm based on the forward-backward proximal gradient splitting technique is introduced, which captures wall clutter and yields target images simultaneously. Extensive experiments are conducted on real radar data under compressive sensing scenarios. The results show that the proposed model enhances target localization and clutter mitigation even when radar measurements are significantly reduced.

DeepDSSR: Deep Learning Structure for Human Donor Splice Sites Recognition.

Human genes often, through alternative splicing of pre-messenger RNAs, produce multiple mRNAs and protein isoforms that may have similar or completely different functions. Identification of splice sites is, therefore, crucial to understand the gene structure and variants of mRNA and protein isoforms produced by the primary RNA transcripts. Although many computational methods have been developed to detect the splice sites in humans, this is still substantially a challenging problem and further improvement of the computational model is still foreseeable. Accordingly, we developed DeepDSSR (deep donor splice site recognizer), a novel deep learning based architecture, for predicting human donor splice sites. The proposed method, built upon publicly available and highly imbalanced benchmark dataset, is comparable with the leading deep learning based methods for detecting human donor splice sites. Performance evaluation metrics show that DeepDSSR outperformed the existing deep learning based methods. Future work will improve the predictive capabilities of our model, and we will build a model for the prediction of acceptor splice sites.

Multipolarization Through-Wall Radar Imaging Using Low-Rank and Jointly-Sparse Representations.

Compressed sensing techniques have been applied to through-the-wall radar imaging (TWRI) and multipolarization TWRI for fast data acquisition and enhanced target localization. The studies so far in this area have either assumed effective wall clutter removal prior to image formation or performed signal estimation, wall clutter mitigation, and image formation independently. This paper proposes a low-rank and sparse imaging model for jointly addressing the problem of wall clutter mitigation and image formation in multichannel TWRI. The proposed model exploits two important structures of through-wall radar signals: low-rank structure of the wall reflections and jointly-sparse structure among the different polarization images. The task of removing wall clutter and reconstructing multichannel images of the same scene behind-the-wall is formulated as a regularized least squares problem, where low-rank regularization is enforced for the wall components, and joint-sparsity penalty is imposed on channel images. To solve the optimization problem, an iterative algorithm based on the proximal gradient technique is introduced, which simultaneously estimates the wall interferences and yields multichannel images of the indoor targets. Experiments on real and simulated radar data are conducted under full measurements and compressive sensing scenarios. The results show that the proposed model is very effective at removing unwanted wall clutter and enhancing the stationary targets, even under considerable reduction in measurements.

Probabilistic fuzzy image fusion approach for radar through wall sensing.

This paper addresses the problem of combining multiple radar images of the same scene to produce a more informative composite image. The proposed approach for probabilistic fuzzy logic-based image fusion automatically forms fuzzy membership functions using the Gaussian-Rayleigh mixture distribution. It fuses the input pixel values directly without requiring fuzzification and defuzzification, thereby removing the subjective nature of the existing fuzzy logic methods. In this paper, the proposed approach is applied to through-the-wall radar imaging in urban sensing and evaluated on real multi-view and polarimetric data. Experimental results show that the proposed approach yields improved image contrast and enhances target detection.

Automatic left ventricle detection in echocardiographic images for deformable contour initialization.

The accurate left ventricular boundary detection in echocardiographic images allow cardiologists to study and assess cardiomyopathy in patients. Due to the tedious and time consuming manner of manually tracing the borders, deformable models are generally used for left ventricle segmentations. However, most deformable models require a good initialization, which is usually outlined manually by the user. In this paper, we propose an automated left ventricle detection method for two-dimensional echocardiographic images that could serve as an initialization for deformable models. The proposed approach consists of pre-processing and post-processing stages, coupled with the watershed segmentation. The pre-processing stage enhances the overall contrast and reduces speckle noise, whereas the post-processing enhances the segmented region and avoids the papillary muscles. The performance of the proposed method is evaluated on real data. Experimental results show that it is suitable for automatic contour initialization since no prior assumptions nor human interventions are required. Besides, the computational time taken is also lower compared to an existing method.

Adaptive hierarchical architecture for visual recognition.

We propose a new hierarchical architecture for visual pattern classification. The new architecture consists of a set of fixed, directional filters and a set of adaptive filters arranged in a cascade structure. The fixed filters are used to extract primitive features such as orientations and edges that are present in a wide range of objects, whereas the adaptive filters can be trained to find complex features that are specific to a given object. Both types of filter are based on the biological mechanism of shunting inhibition. The proposed architecture is applied to two problems: pedestrian detection and car detection. Evaluation results on benchmark data sets demonstrate that the proposed architecture outperforms several existing ones.

A pyramidal neural network for visual pattern recognition.

In this paper, we propose a new neural architecture for classification of visual patterns that is motivated by the two concepts of image pyramids and local receptive fields. The new architecture, called pyramidal neural network (PyraNet), has a hierarchical structure with two types of processing layers: Pyramidal layers and one-dimensional (1-D) layers. In the new network, nonlinear two-dimensional (2-D) neurons are trained to perform both image feature extraction and dimensionality reduction. We present and analyze five training methods for PyraNet [gradient descent (GD), gradient descent with momentum, resilient back-propagation (RPROP), Polak-Ribiere conjugate gradient (CG), and Levenberg-Marquadrt (LM)] and two choices of error functions [mean-square-error (mse) and cross-entropy (CE)]. In this paper, we apply PyraNet to determine gender from a facial image, and compare its performance on the standard facial recognition technology (FERET) database with three classifiers: The convolutional neural network (NN), the k-nearest neighbor (k-NN), and the support vector machine (SVM).

Efficient training algorithms for a class of shunting inhibitory convolutional neural networks.

This article presents some efficient training algorithms, based on first-order, second-order, and conjugate gradient optimization methods, for a class of convolutional neural networks (CoNNs), known as shunting inhibitory convolution neural networks. Furthermore, a new hybrid method is proposed, which is derived from the principles of Quickprop, Rprop, SuperSAB, and least squares (LS). Experimental results show that the new hybrid method can perform as well as the Levenberg-Marquardt (LM) algorithm, but at a much lower computational cost and less memory storage. For comparison sake, the visual pattern recognition task of face/nonface discrimination is chosen as a classification problem to evaluate the performance of the training algorithms. Sixteen training algorithms are implemented for the three different variants of the proposed CoNN architecture: binary-, Toeplitz- and fully connected architectures. All implemented algorithms can train the three network architectures successfully, but their convergence speed vary markedly. In particular, the combination of LS with the new hybrid method and LS with the LM method achieve the best convergence rates in terms of number of training epochs. In addition, the classification accuracies of all three architectures are assessed using ten-fold cross validation. The results show that the binary- and Toeplitz-connected architectures outperform slightly the fully connected architecture: the lowest error rates across all training algorithms are 1.95% for Toeplitz-connected, 2.10% for the binary-connected, and 2.20% for the fully connected network. In general, the modified Broyden-Fletcher-Goldfarb-Shanno (BFGS) methods, the three variants of LM algorithm, and the new hybrid/LS method perform consistently well, achieving error rates of less than 3% averaged across all three architectures.

Skin segmentation using color pixel classification: analysis and comparison.

This paper presents a study of three important issues of the color pixel classification approach to skin segmentation: color representation, color quantization, and classification algorithm. Our analysis of several representative color spaces using the Bayesian classifier with the histogram technique shows that skin segmentation based on color pixel classification is largely unaffected by the choice of the color space. However, segmentation performance degrades when only chrominance channels are used in classification. Furthermore, we find that color quantization can be as low as 64 bins per channel, although higher histogram sizes give better segmentation performance. The Bayesian classifier with the histogram technique and the multilayer perceptron classifier are found to perform better compared to other tested classifiers, including three piecewise linear classifiers, three unimodal Gaussian classifiers, and a Gaussian mixture classifier.

A generalized feedforward neural network architecture for classification and regression.

This article presents a new generalized feedforward neural network (GFNN) architecture for pattern classification and regression. The GFNN architecture uses as the basic computing unit a generalized shunting neuron (GSN) model, which includes as special cases the perceptron and the shunting inhibitory neuron. GSNs are capable of forming complex, nonlinear decision boundaries. This allows the GFNN architecture to easily learn some complex pattern classification problems. In this article the GFNNs are applied to several benchmark classification problems, and their performance is compared to the performances of SIANNs and multilayer perceptrons. Experimental results show that a single GSN can outperform both the SIANN and MLP networks.