Improvement of image quality in diffuse optical tomography by use of full time-resolved data

In the field of diffuse optical tomography (DOT), it is widely accepted that time-resolved (TR) measurement can provide the richest information on photon migration in a turbid medium, such as biological tissue. However, the currently available image reconstruction algorithms for TR DOT are based mostly on the cw component or some featured data types of original temporal profiles, which are related to the solution of a time-independent diffusion equation. Although this methodology can greatly simplify the reconstruction process, it suffers from low spatial resolution and poor quantitativeness owing to the limitation of effectively applicable data types. To improve image quality, it has been argued that exploiting the full TR data is essential. We propose implementation of a DOT algorithm by using full TR data and furthermore a variant algorithm with time slices of TR data to alleviate the computational complexity and enhance noise robustness. Compared with those algorithms where the featured data types are used, our evaluations on the spatial resolution and quantitativeness show that a significant improvement in imaging quality can be achieved when full TR data are used, which convinces the DOT community of the potential advantage of the TR domain over cw and frequency domains.     

Improved fuzzy entropy clustering algorithm for MRI brain image segmentation

Magnetic resonance imaging (MRI) brain image segmentation is essential at preliminary stage in the neuroscience research and computer‐aided diagnosis. However, presence of noise and intensity inhomogeneity in MRI brain images leads to improper segmentation. The fuzzy entropy clustering (FEC) is often used to deal with noisy data. One major disadvantage of the FEC algorithm is that it does not consider the local spatial information. In this article, we have proposed an improved fuzzy entropy clustering (IFEC) algorithm by introducing a new fuzzy factor, which incorporates both local spatial and gray‐level information. The IFEC algorithm is insensitive to noise, preserves the image detail during clustering, and is free of parameter selection. The efficacy of IFEC algorithm is demonstrated by comparing it quantitatively with the state‐of‐the‐art segmentation approaches in terms of similarity index on publically available real and simulated MRI brain images.     

Improved accuracy of reconstructed diffuse optical tomographic images by means of spatial deconvolution: two-dimensional quantitative characterization

Systematic characterization studies are presented, relating to a previously reported spatial deconvolution operation that seeks to compensate for the information-blurring property of first-order perturbation algorithms for diffuse optical tomography (DOT) image reconstruction. In simulation results that are presented, this deconvolution operation has been applied to two-dimensional DOT images reconstructed by solving a first-order perturbation equation. Under study was the effect on algorithm performance of control parameters in the measurement (number and spatial distribution of sources and detectors, presence of noise, and presence of systematic error), target (medium shape; and number, location, size, and contrast of inclusions), and computational (number of finite-element-method mesh nodes, length of filter-generating linear system, among others) parameter spaces associated with computation and the use of the deconvolution operators. Substantial improvements in reconstructed image quality, in terms of recovered inclusion location, size, and contrast, are found in all cases. A finding of practical importance is that the method is robust to appreciable differences between the optical coefficients of the media used for filter generation and those of the target media to which the filters are subsequently applied.     

Image correction scheme applied to functional diffuse optical tomography scattering images

We have extended our investigation on the use of a linear algorithm for enhancing the accuracy of diffuse optical tomography (DOT) images, to include spatial maps of the diffusion coefficient. The results show that the corrected images are markedly improved in terms of estimated size, spatial resolution, two-object resolving power, and quantitative accuracy. These image-enhancing effects are significant at expected levels of diffusion-coefficient contrast in tissue and noise levels typical of experimental DOT data. Overall, the types and magnitudes of image-enhancing effects obtained here are qualitatively similar to those seen in previous studies on mu(a) perturbations. The implications for practical implementations of DOT time-series imaging are discussed.     

Optimal linear inverse solution with multiple priors in diffuse optical tomography

A general framework for incorporating single and multiple priors in diffuse optical tomography is described. We explore the use of this framework for simultaneously utilizing spatial and spectral priors in the context of imaging breast cancer. The utilization of magnetic resonance images of water and lipid content as a statistical spatial prior for the diffuse optical image reconstructions is also discussed. Simulations are performed to demonstrate the significant improvement in image quality afforded by combining spatial and spectral priors.     

Time-resolved diffuse optical tomographic imaging for the provision of both anatomical and functional information about biological tissue

We present in vivo images of near-infrared (NIR) diffuse optical tomography (DOT) of human lower legs and forearm to validate the dual functions of a time-resolved (TR) NIR DOT in clinical diagnosis, i.e., to provide anatomical and functional information simultaneously. The NIR DOT system is composed of time-correlated single-photon-counting channels, and the image reconstruction algorithm is based on the modified generalized pulsed spectral technique, which effectively incorporates the TR data with reasonable computation time. The reconstructed scattering images of both the lower legs and the forearm revealed their anatomies, in which the bones were clearly distinguished from the muscles. In the absorption images, some of the blood vessels were observable. In the functional imaging, a subject was requested to do handgripping exercise to stimulate physiological changes in the forearm tissue. The images of oxyhemoglobin, deoxyhemoglobin, and total hemoglobin concentration changes in the forearm were obtained from the differential images of the absorption at three wavelengths between the exercise and the rest states, which were reconstructed with a differential imaging scheme. These images showed increases in both blood volume and oxyhemoglobin concentration in the arteries and simultaneously showed hypoxia in the corresponding muscles. All the results have demonstrated the capability of TR NIR DOT by reconstruction of the absolute images of the scattering and the absorption with a high spatial resolution that finally provided both the anatomical and functional information inside bulky biological tissues.     

Semi-three-dimensional algorithm for time-resolved diffuse optical tomography by use of the generalized pulse spectrum technique

Although a foil three-dimensional (3-D) reconstruction with both 3-D forward and inverse models provide, the optimal solution for diffuse optical tomography (DOT), because of the 3-D nature of photon diffusion in tissue, it is computationally costly for both memory requirement and execution time in a conventional computing environment. Thus in practice there is motivation to develop an image reconstruction algorithm with dimensional reduction based on some modeling approximations. Here we have implemented a semi-3-D modified generalized pulse spectrum technique for time-resolved DOT, where a two-dimensional (2-D) distribution of optical properties is approximately assumed, while we retain 3-D distribution of photon migration in tissue. We have validated the proposed algorithm by reconstructing 3-D structural test objects from both numerically simulated and experimental date. We demonstrate our algorithm by comparing it with the calibrated 2-D reconstruction that is in widespread use as a shortcut to 3-D imaging and proving that the semi-3-D algorithm outperforms the calibrated 2-D algorithm.     

The clique potential of Markov random field in a random experiment for estimation of noise levels in 2D brain MRI

Effective performance of many image processing and image analysis algorithms is strongly dependent on accurate estimation of noise level. We exploit the simplicity and similarity of statistics of human anatomy among different subjects to develop new noise level estimation algorithm for magnetic resonance images of brain. Objects of the experiment are noise‐free 3D brain MRI of 422 subjects. There are 21 slices for each subject. For each slice, total clique potential (TCP) of Markov random field, computed from local clique potential, is indexed by 200 different levels of noise. The sample space is the set of TCP‐noise level data of each slice. The random variable is the set of indices of noise level of TCP in each element of sample space that is closest in numerical value to TCP measured from a test MRI slice. Noise level is estimated from the mean and variance of the random variable. We also report the formulation of a generalized mathematical model describing relationship between TCP and Rician noise level in brain MRI images. Our proposal can operate in the absence of signals in the background and significantly reduce modeling errors inherent in strong parametric assumptions adopted by some of the current algorithms. © 2013 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 23, 304–413, 2013     

Developments with maximum-likelihood x-ray computed tomography: initial testing with real data

We investigate the potential and present limitations of a maximum-likelihood (ML) approach to x-ray computed tomography that utilizes Poisson modeling and an iterative gradient-based algorithm. This model and algorithm incorporate the finite width of the x-ray beam, and they were extended from an approach originally proposed by Lange et al. [IEEE Trans. Med. Imaging MI-6, 106-114 (1987)]. Lowcount data, obtained from an industrial computed-tomography scanner, are used to reconstruct an image of a concrete cube with metal reinforcing bars. We utilize both ML and filtered backprojection to reconstruct a cross section of the internal structure of the cube. In this initial evaluation with low-count data the images reconstructed by ML show several potential advantages over those reconstructed by filtered backprojection. The advantages shown are the following: (1) there are significantly reduced noise and streak artifacts in the ML image; (2) some of the known structural detail is more apparent in the ML image; (3) there is a closer quantitative fit, based on log-likelihood and residual calculations, between the ML image and the observed data; (4) the ML approach shows the potential to achieve finer spatial resolution than filtered backprojection. We observe two present, yet addressable, limitations of the ML approach. First, the ML image currently has a peripheral smoothing artifact that seems to disappear gradually with increasing iteration numbers. This smoothing is possibly caused by the slow rate of convergence of the algorithm and may be addressed by future acceleration strategies. Second, the finer spatial resolution achieved with the ML approach currently occurs at the expense of noise and edge artifacts. This limitation may be addressed by a number of extended ML and maximum a posteriori approaches that are currently under investigation in other modalities of imaging to address similar noise and edge artifacts.     

Simultaneous tomographic reconstruction and segmentation with class priors

We consider tomographic imaging problems where the goal is to obtain both a reconstructed image and a corresponding segmentation. A classical approach is to first reconstruct and then segment the image; more recent approaches use a discrete tomography approach where reconstruction and segmentation are combined to produce a reconstruction that is identical to the segmentation. We consider instead a hybrid approach that simultaneously produces both a reconstructed image and segmentation. We incorporate priors about the desired classes of the segmentation through a Hidden Markov Measure Field Model, and we impose a regularization term for the spatial variation of the classes across neighbouring pixels. We also present an efficient implementation of our algorithm based on state-of-the-art numerical optimization algorithms. Simulation experiments with artificial and real data demonstrate that our combined approach can produce better results than the classical two-step approach.     

Correction of systematic spatial noise in push-broom hyperspectral sensors: application to CHRIS/PROBA images

Hyperspectral remote sensing images are affected by different types of noise. In addition to typical random noise, nonperiodic partially deterministic disturbance patterns generally appear in the data. These patterns, which are intrinsic to the image formation process, are characterized by a high degree of spatial and spectral coherence. We present a new technique that faces the problem of removing the spatially coherent noise known as vertical striping, usually found in images acquired by push-broom sensors. The developed methodology is tested on data acquired by the Compact High Resolution Imaging Spectrometer (CHRIS) onboard the Project for On-board Autonomy (PROBA) orbital platform, which is a typical example of a push-broom instrument exhibiting a relatively high noise component. The proposed correction method is based on the hypothesis that the vertical disturbance presents higher spatial frequencies than the surface radiance. A technique to exclude the contribution of the spatial high frequencies of the surface from the destriping process is introduced. First, the performance of the proposed algorithm is tested on a set of realistic synthetic images with added modeled noise in order to quantify the noise reduction and the noise estimation accuracy. Then, algorithm robustness is tested on more than 350 real CHRIS images from different sites, several acquisition modes (different spatial and spectral resolutions), and covering the full range of possible sensor temperatures. The proposed algorithm is benchmarked against the CHRIS reference algorithm. Results show excellent rejection of the noise pattern with respect to the original CHRIS images, especially improving the removal in those scenes with a natural high contrast. However, some low-frequency components still remain. In addition, the developed correction model captures and corrects the dependency of the noise patterns on sensor temperature, which confirms the robustness of the presented approach.     

Noise-cancellation-based nonuniformity correction algorithm for infrared focal-plane arrays

The spatial fixed-pattern noise (FPN) inherently generated in infrared (IR) imaging systems compromises severely the quality of the acquired imagery, even making such images inappropriate for some applications. The FPN refers to the inability of the photodetectors in the focal-plane array to render a uniform output image when a uniform-intensity scene is being imaged. We present a noise-cancellation-based algorithm that compensates for the additive component of the FPN. The proposed method relies on the assumption that a source of noise correlated to the additive FPN is available to the IR camera. An important feature of the algorithm is that all the calculations are reduced to a simple equation, which allows for the bias compensation of the raw imagery. The algorithm performance is tested using real IR image sequences and is compared to some classical methodologies.     

Rician noise removal in magnitude MRI images using efficient anisotropic diffusion filtering

In this article, a new methodology for denoising of Rician noise in Magnetic Resonance Images (MRI) is presented. MRI imaging creates a distinctive view into the interior of a human body and has become an essential tool of clinical diagnosis. However, Rician noise is a type of artifact inherent to the acquisition process of the magnitude MRI image, making diagnosis difficult. We proposed a moment‐based Rician noise reduction technique in anisotropic diffusion filtering. We extend the work of the classical anisotropic diffusion filter and have customized it to remove Rician noise in the magnitude MRI image in 3D domain space. Our proposed scheme shows better results against various quality measures in terms of noise removal and edge preservation while retaining fine textures.     

Dynamic estimation of the modeling error statistics in Diffuse Optical Tomography

Diffuse Optical Tomography (DOT) is a non-invasive imaging technique that suffers from a typical large-scale and ill-posed inverse problem with low spatial resolution. In DOT, the inverse problem is computationally intensive and decreasing the computation complexity and making it well-posed is the one of the most challenging research areas. More precisely, one of the well-known complexity reduction techniques is defined as applying modelling error originated from discretization of forward problem. Applying the discretization error in Bayesian inference has already been discussed; the method in which the likelihood is modified by an off-line prior density estimation. This paper implements a new method to enhance the modelling error approach using an iterative scheme to update statistical parameters of modelling discrepancy in DOT. The algorithm is very similar to Ensemble Kalman Filter. Moreover, the reconstruction process in the applied method is conducted by a small sample size rather than off-line method. Hence, the computation complexity is decreased and the algorithm converges in few iterations. The efficiency of the proposed method is illustrated by simulations.     

Fuzzy clustering-based image segmentation techniques used to segment magnetic resonance imaging/computed tomography scan brain tissues: Comparative analysis

Medical images are obtained with computer-aided diagnosis using electronic devices such as CT scanners and MRI machines. The captured computed tomography (CT)/magnetic resonance imaging (MRI) images typically have limited spatial resolution, low contrast, noise and nonuniform variability in intensity due to environmental effects. Therefore, the distinctions of the objects are blurred, distorted and the meanings of the objects are not quite precise. Fuzzy sets and fuzzy logic are best suited for addressing vagueness and ambiguity. Fuzzy clustering technique has been commonly used for segmentation of images throughout the last decade. This study presents a comparative study of 14 fuzzy-clustered image segmentation algorithms used in the CT scan and MRI brain image segments. This study used 17 data sets including 4 synthetic data sets, namely, Bensaid, Diamond, Square, and its noisy version, 5 real-world digital images, and 8 CT scan/MRI brain images to analyze the algorithms. Ground truth images are used for qualitative analysis. Apart from the qualitative analysis, the study also quantitatively evaluated the methods using three validity metrics, namely, partition coefficient, partition entropy, and Fukuyama-Sugeno. After a thorough and careful review of the results, it is observed that extension of the fuzzy C-means (EFCM) outperformed every other image segmentation algorithm, even in a noisy environment, followed by kernel-based FCM σ, the output of which is also very good after EFCM.     

Binaural Speech Separation Algorithm Based on Long and Short Time Memory Networks

Speaker separation in complex acoustic environment is one of challenging tasks in speech separation. In practice, speakers are very often unmoving or moving slowly in normal communication. In this case, the spatial features among the consecutive speech frames become highly correlated such that it is helpful for speaker separation by providing additional spatial information. To fully exploit this information, we design a separation system on Recurrent Neural Network (RNN) with long short-term memory (LSTM) which effectively learns the temporal dynamics of spatial features. In detail, a LSTM-based speaker separation algorithm is proposed to extract the spatial features in each time-frequency (TF) unit and form the corresponding feature vector. Then, we treat speaker separation as a supervised learning problem, where a modified ideal ratio mask (IRM) is defined as the training function during LSTM learning. Simulations show that the proposed system achieves attractive separation performance in noisy and reverberant environments. Specifically, during the untrained acoustic test with limited priors, e.g., unmatched signal to noise ratio (SNR) and reverberation, the proposed LSTM based algorithm can still outperforms the existing DNN based method in the measures of PESQ and STOI. It indicates our method is more robust in untrained conditions.     

Robust camera pose estimation from unknown or known line correspondences

We address the model-to-image registration problem with line features in the following two ways. (a) We present a robust solution to simultaneously recover the camera pose and the three-dimensional-to-two-dimensional line correspondences. With weak pose priors, our approach progressively verifies the pose guesses with a Kalman filter by using a subset of recursively found match hypotheses. Experiments show our method is robust to occlusions and clutter. (b) We propose a new line feature based pose estimation algorithm, which iteratively optimizes the objective function in the object space. Experiments show that the algorithm has strong robustness to noise and outliers and that it can attain very accurate results efficiently.     

Poisson noise reduction from X-ray images by region classification and response median filtering

Medical imaging is perturbed with inherent noise such as speckle noise in ultrasound, Poisson noise in X-ray and Rician noise in MRI imaging. This paper focuses on X-ray image denoising problem. X-ray image quality could be improved by increasing dose value; however, this may result in cell death or similar kinds of issues. Therefore, image processing techniques are developed to minimise noise instead of increasing dose value for patient safety. In this paper, usage of modified Harris corner point detector to predict noisy pixels and responsive median filtering in spatial domain is proposed. Experimentation proved that the proposed work performs better than simple median filter and moving average (MA) filter. The results are very close to non-local means Poisson noise filter which is one of the current state-of-the-art methods. Benefits of the proposed work are simple noise prediction mechanism, good visual quality and less execution time.     

Interventional magnetic resonance imaging: an alternative to image guidance with ionising radiation

At present, interventional procedures, such as stent placement, are performed under X-ray image guidance. Unfortunately with X-ray imaging, both patient and interventionalist are exposed to ionising radiation. Furthermore, X-ray imaging is lacking soft tissue contrast and is not capable of true 3-D displays of either interventional device or tissue morphology. Magnetic resonance imaging (MRI) offers excellent soft tissue contrast, 3-D acquisition techniques, as well as rapid image acquisition and reconstruction. Despite these advantages, MR-guided interventions are challenging owing to the limited access to the patient, strong magnetic and radio-frequency fields that require special interventional devices, inferior image frame rates and spatial resolution, and high MRI scanner noise. For MR-guided intravascular interventions, where access to the target organ is achieved through catheters, dedicated hardware and automated image slice positioning techniques have been developed. We illustrate that MR-guided renal embolisations can be performed in closed-bore high-field MR scanners.     

Detection improvement in spatially filtered x-ray fluoroscopy image sequences

The effect of spatial noise-reduction filtering on human observer detection of stationary cylinders mimicking arteries, catheters, and guide wires in x-ray fluoroscopy was investigated in both single image frames and image sequences. Ideal edge-preserving spatial filtering was simulated by filtering of the noise before addition of the target cylinder. This allowed us to separate the effect of edge blurring from those of noise reduction and spatial noise correlation. We used three different center-weighted averagers that reduced pixel noise variance by factors of 0.75, 0.50, and 0.25. As compared with no filtering, the effect of filtering on detection in single images was statistically insignificant. This indicated an adverse effect of spatial noise correlation on detection that countered the effect of noise reduction. By comparison, spatial filtering significantly improved detection in image sequences and yielded potential x-ray dose savings of 26-34%. Comparison of results with two observer models suggested that human observers have an improved detection efficiency in spatially filtered image sequences as compared with white-noise sequences. Pixel noise reduction, a measure commonly used to assess filter performance, overestimated the effect of filtering on detection and was not a good indicator of image quality. We conclude that edge-preserving spatial filtering is more effective in sequences than in single images and that such filtering can be used to improve image quality in noisy image sequences such as x-ray fluoroscopy.