Showing papers by "Adrian Basarab published in 2019"

Deep Learning-Based Super-Resolution Applied to Dental Computed Tomography

Abstract: The resolution of dental computed tomography (CT) images is limited by detector geometry, sensitivity, patient movement, the reconstruction technique and the need to minimize radiation dose. Recently, the use of convolutional neural network (CNN) architectures has shown promise as a resolution enhancement method. In the current work, two CNN architectures—a subpixel network and the so called U-net—have been considered for the resolution enhancement of 2-D cone-beam CT image slices of ex vivo teeth. To do so, a training set of 5680 cross-sectional slices of 13 teeth and a test set of 1824 slices of 4 structurally different teeth were used. Two existing reconstruction-based super-resolution methods using $\boldsymbol {\ell _{2}}$ -norm and total variation regularization were used for comparison. The results were evaluated with different metrics (peak signal-to-noise ratio, structure similarity index, and other objective measures estimating human perception) and subsequent image-segmentation-based analysis. In the evaluation, micro-CT images were used as ground truth. The results suggest the superiority of the proposed CNN-based approaches over reconstruction-based methods in the case of dental CT images, allowing better detection of medically salient features, such as the size, shape, or curvature of the root canal.

A Tensor Factorization Method for 3-D Super Resolution With Application to Dental CT

Abstract: Available super-resolution techniques for 3-D images are either computationally inefficient prior-knowledge-based iterative techniques or deep learning methods which require a large database of known low-resolution and high-resolution image pairs. A recently introduced tensor-factorization-based approach offers a fast solution without the use of known image pairs or strict prior assumptions. In this paper, this factorization framework is investigated for single image resolution enhancement with an offline estimate of the system point spread function. The technique is applied to 3-D cone beam computed tomography for dental image resolution enhancement. To demonstrate the efficiency of our method, it is compared to a recent state-of-the-art iterative technique using low-rank and total variation regularizations. In contrast to this comparative technique, the proposed reconstruction technique gives a 2-order-of-magnitude improvement in running time—2 min compared to 2 h for a dental volume of $282\times 266\times392$ voxels. Furthermore, it also offers slightly improved quantitative results (peak signal-to-noise ratio and segmentation quality). Another advantage of the presented technique is the low number of hyperparameters. As demonstrated in this paper, the framework is not sensitive to small changes in its parameters, proposing an ease of use.

Motion Compensated Dynamic MRI Reconstruction With Local Affine Optical Flow Estimation

Abstract: This paper proposes a novel framework to reconstruct dynamic magnetic resonance imaging (DMRI) with motion compensation (MC). Specifically, by combining the intensity-based optical flow constraint with the traditional compressed sensing scheme, we are able to jointly reconstruct the DMRI sequences and estimate the interframe motion vectors. Then, the DMRI reconstruction can be refined through MC with the estimated motion field. By employing the coarse-to-fine multi-scale resolution strategy, we are able to update the motion field in different spatial scales. The estimated motion vectors need to be interpolated to the finest resolution scale to compensate the DMRI reconstruction. Moreover, the proposed framework is capable of handling a wide class of prior information (regularizations) for DMRI reconstruction, such as sparsity, low rank, and total variation. The formulated optimization problem is solved by a primal–dual algorithm with linesearch due to its efficiency when dealing with non-differentiable problems. Experiments on various DMRI datasets validate the reconstruction quality improvement using the proposed scheme in comparison to several state-of-the-art algorithms.

Robust Optical Flow Estimation in Cardiac Ultrasound Images Using a Sparse Representation

Abstract: This paper introduces a robust 2-D cardiac motion estimation method. The problem is formulated as an energy minimization with an optical flow-based data fidelity term and two regularization terms imposing spatial smoothness and the sparsity of the motion field in an appropriate cardiac motion dictionary. Robustness to outliers, such as imaging artefacts and anatomical motion boundaries, is introduced using robust weighting functions for the data fidelity term as well as for the spatial and sparse regularizations. The motion fields and the weights are computed jointly using an iteratively re-weighted minimization strategy. The proposed robust approach is evaluated on synthetic data and realistic simulation sequences with available ground-truth by comparing the performance with state-of-the-art algorithms. Finally, the proposed method is validated using two sequences of in vivo images. The obtained results show the interest of the proposed approach for 2-D cardiac ultrasound imaging.

Iterative Reconstruction of Medical Ultrasound Images Using Spectrally Constrained Phase Updates

Abstract: Image deconvolution is a standard numerical procedure used in medical ultrasound imaging for improving the resolution and contrast of diagnostic sonograms However, due to the intrinsic bandlimitedness of ultrasound scanners and the adverse effect of measurement noises, image deconvolution is known to be exceedingly sensitive to the errors incurred during inference of the point spread function (PSF) that characterizes the imaging system in use In this case, even the slightest errors in specification of the PSF are likely to result in significant artifacts, rendering the reconstructed images worthless To address the aforementioned problem, this paper describes a new method for blind deconvolution of ultrasound images, in which the errors due to inaccuracies in specification of the PSF are eliminated concurrently with estimation of tissue reflectivity directly from its associated radio-frequency data A principal derivation and justification of the proposed method are supported by experimental results which demonstrate the effectiveness and viability of the new technique

High-resolution and high-sensitivity blood flow estimation using optimization approaches with application to vascularization imaging

Abstract: In this paper, we address the problem of high-resolution flow estimation in medical ultrasound images. Imaging methods based on ultrafast sequences associated with adaptive spatiotemporal SVD clutter filtering have recently improved blood flow detection. Herein, we investigate a new way of addressing the clutter filtering problem in order to obtain a high-resolution flow estimation, through solving an inverse problem corresponding to both deconvolution and robust principal component analysis. Applied to tissue vascularization imaging via power Doppler images, the proposed method highlights finer details on experimental data compared to existing approaches.

On Multifractal Tissue Characterization in Ultrasound Imaging

Abstract: Tissue characterization based on ultrasound (US) images is an extensively explored research field. Most of the existing techniques are focused on the estimation of statistical or acoustic parameters from the backscattered radio-frequency signals, thus complementing the visual inspection of the conventional B-mode images. Additionally, a few studies show the interest of analyzing the fractal or multifractal behavior of human tissues, in particular of tumors. While biological experiments sustain such multifractal behaviors, the observations on US images are rather empirical. To our knowledge, there is no theoretical or practical study relating the fractal or multifractal parameters extracted from US images to those of the imaged tissues. The aim of this paper is to investigate how multifractal properties of a tissue correlate with the ones estimated from a simulated US image for the same tissue. To this end, an original simulation pipeline of multifractal tissues and their corresponding US images is proposed. Simulation results are compared to those in an in vivo experiment.

On Single-Image Super-Resolution in 3D Brain Magnetic Resonance Imaging

Abstract: The objective of this work is to apply 3D super resolution (SR) techniques to brain magnetic resonance (MR) image restoration. Two 3D SR methods are considered following different trends: one recently proposed tensor-based approach and one inverse problem algorithm based on total variation and low rank regularization. The evaluation of their effectiveness is assessed through the segmentation of brain compartments: gray matter, white matter and cerebrospinal fluid. The two algorithms are qualitatively and quantitatively evaluated on simulated images with ground truth available and on experimental data. The originality of this work is to consider the SR methods as an initial step towards the final segmentation task. The results show the ability of both methods to overcome the loss of spatial resolution and to facilitate the segmentation of brain structures with improved accuracy compared to native low-resolution MR images. Both algorithms achieved almost equivalent results with a highly reduced computational time cost for the tensor-based approach.

On the Design of a Pelvic Phantom for Magentic Resononance and Utltrasound Image Fusion

Abstract: The purpose of this paper is to introduce a customized multi- modality phantom designed to facilitate the proof-of-concept of MRI/ultrasound fusion approaches. Phantom experiments are often required before in vivo validation, giving access to more challenging data than numerical simulations. Neverthe- less, manufactured phantoms are expensive and usually lack of flexibility. In contrast, the proposed model was inexpensive and accurately designed to overcome multimodal registration issues.

Robust Cardiac Motion Estimation With Dictionary Learning and Temporal Regularization for Ultrasound Imaging

Abstract: Estimating the cardiac motion from ultrasound (US) images is an ill-posed problem that requires regularization. In a recent study, it was shown that constraining the cardiac motion fields to be patch-wise sparse in a learnt overcomplete motion dictionary is more accurate than local parametric models (affine) or global functions (B-splines, total variation). In this work, we extend this method by incorporating temporal smoothness in a multi-frame optical-flow (OF) strategy. An efficient optimization strategy using the constrained split augmented Lagrangian shrinkage algorithm (C-SALSA) is proposed. The performance is evaluated on a realistic simulated cardiac dataset with available ground-truth. A comparison with the pairwise approach shows the interest of the proposed temporal regularization and multi-frame strategy in terms of accuracy and computational time.

Fusion Of Magnetic Resonance And Ultrasound Images: A Preliminary Study On Simulated Data

Abstract: We propose a new fusion method for magnetic resonance imaging (MRI) and ultrasound (US) data combining two inverse problems: MRI reconstruction using super-resolution and US image despeckling, using a model relating the two modalities through an unknown polynomial function. We demonstrate the accuracy of the proposed fusion algorithm by quantitative and qualitative evaluation using synthetic data. The resulting fused image is shown to have an improved signal to noise ratio and spatial resolution compared to the native MRI and US images.

Cardiac Motion Estimation Using Convolutional Sparse Coding

Abstract: This paper studies a new motion estimation method based on convolutional sparse coding. The motion estimation problem is formulated as the minimization of a cost function composed of a data fidelity term, a spatial smoothness constraint, and a regularization based on convolution sparse coding. We study the potential interest of using a convolutional dictionary instead of a standard dictionary using specific examples. Moreover, the proposed method is evaluated in terms of motion estimation accuracy and compared with state-of-the-art algorithms, showing its interest for cardiac motion estimation.

Tissue Reflectivity Function Restoration from Fundamental and Harmonic Ultrasound Images

Abstract: This paper addresses the problem of ultrasound (US) image restoration. In contrast to most of the existing approaches that only take into account fundamental radiofrequency (RF) data, the proposed method also considers harmonic US images. An algorithm based on the alternating direction of multipliers method (ADMM) is proposed to solve the joint deconvolution problem. Simulation results show the interest of the proposed approach when compared to classical US image restoration schemes based only on fundamental data.

Brain Segmentation from Super-Resolved Magnetic Resonance Images

Abstract: The objective of this work is to investigate the ability of a 2D super resolution (SR) technique in 3D restoration and enhancement of brain magnetic resonance images to facilitate the study of cerebral aging bio-markers. The SR method exploits the joint properties of the system point spread function and sub-sampling operators to derive a fast algorithm. Brain images of the common marmoset, Callithrix jacchus, acquired at different ages are used in this study. The evaluation of the final outcome of our method is done by computing the intracranial volume from the segmentation of the brain compartments: gray matter, white matter and cerebrospinal fluid. Results show that the deblurring of the images improves the segmentation process with respect to the ground truth. However, super resolution leads to the best quantification of the intracranial volume when compared to the deblurred and the original images. Therefore, despite its sub-optimality, the 2D SR method provides reliable results for improving the quality of the images used in the study of aging in terms of precision of reconstruction and computational time.

Joint Deconvolution of Fundamental and Harmonic Ultrasound Images

Abstract: This paper studies the interest of using harmonic ultrasound (US) images in the process of tissue reflectivity function restoration from RF data. To this end, two direct models (one for fundamental and another for harmonique images) derived from the equation of US wave propagation are proposed. In particular, an axially varying attenuation matrix is used within the harmonic image model in order to account for the attenuation of harmonic echoes. Based on these two image formation models, a joint deconvolution problem is investigated. The solution of this problem is obtained by minimizing a cost function composed of two data fidelity terms representing the linear and non-linear model components,regularized by an l 1 -norm regularization. The tissue reflectivity function minimizing this function is finally determined using an alternating direction method of multipliers. The performance of the proposed algorithm is quantitatively and qualitatively evaluated on synthetic data, and compared with a classical restoration method used for US images.

Tensor-Factorization-Based 3d Single Image Super-Resolution with Semi-Blind Point Spread Function Estimation

Abstract: A volumetric non-blind single image super-resolution technique using tensor factorization has been recently introduced by our group. That method allowed a 2-order-of-magnitude faster high-resolution image reconstruction with equivalent image quality compared to state-of-the-art algorithms. In this work a joint alternating recovery of the high-resolution image and of the unknown point spread function parameters is proposed. The method is evaluated on dental computed tomography images. The algorithm was compared to an existing 3D super-resolution method using low-rank and total variation regularization, combined with the same alternating PSF-optimization. The two algorithms have shown similar improvement in PSNR, but our method converged roughly 40 times faster, under 6 minutes both in simulation and on experimental dental computed tomography data.

Magnetic Resonance and Ultrasound Image Fusion Using a PALM Algorithm

Abstract: This paper studies a new fusion algorithm for magnetic resonance (MR) and ultrasound (US) images combining two inverse problems for MR image super-resolution and US image despeckling. A polynomial function is used to link the gray levels of the two imaging modalities. Qualitative and quantitative evaluations on experimental phantom data show the interest of the proposed algorithm. The fused image is shown to take advantage of both the good contrast and high signal to noise ratio of the MR image and the good spatial resolution of the US image.

Adaptive Coded Aperture Design by Motion Estimation using Convolutional Sparse Coding in Compressive Spectral Video Sensing

Abstract: This paper proposes a new motion estimation method based on convolutional sparse coding to adaptively design the colored-coded apertures in static and dynamic spectral videos. The motion in a spectral video is estimated from a low-resolution reconstruction of the datacube by training a convolutional dictionary per spectral band and solving a minimization problem. Simulations show improvements in terms of peak signal-to-noise ratio (of up to 2 dB) of the reconstructed videos by using the proposed approach, compared with state-of-art non-adaptive coded apertures.