A fast and flexible algorithm for microstructure reconstruction combining simulated annealing and deep learning

ObjectivesTo evaluate image quality and reconstruction times of a commercial deep learning reconstruction algorithm (DLR) compared to hybrid-iterative reconstruction (Hybrid-IR) and model-based iterative reconstruction (MBIR) algorithms for cerebral non-contrast CT (NCCT).MethodsCerebral NCCT acquisitions of 50 consecutive patients were reconstructed using DLR, Hybrid-IR and MBIR with a clinical CT system. Image quality, in terms of six subjective characteristics (noise, sharpness, grey-white matter differentiation, artefacts, natural appearance and overall image quality), was scored by five observers. As objective metrics of image quality, the noise magnitude and signal-difference-to-noise ratio (SDNR) of the grey and white matter were calculated. Mean values for the image quality characteristics scored by the observers were estimated using a general linear model to account for multiple readers. The estimated means for the reconstruction methods were pairwise compared. Calculated measures were compared using paired t tests.ResultsFor all image quality characteristics, DLR images were scored significantly higher than MBIR images. Compared to Hybrid-IR, perceived noise and grey-white matter differentiation were better with DLR, while no difference was detected for other image quality characteristics. Noise magnitude was lower for DLR compared to Hybrid-IR and MBIR (5.6, 6.4 and 6.2, respectively) and SDNR higher (2.4, 1.9 and 2.0, respectively). Reconstruction times were 27 s, 44 s and 176 s for Hybrid-IR, DLR and MBIR respectively.ConclusionsWith a slight increase in reconstruction time, DLR results in lower noise and improved tissue differentiation compared to Hybrid-IR. Image quality of MBIR is significantly lower compared to DLR with much longer reconstruction times.Key Points• Deep learning reconstruction of cerebral non-contrast CT results in lower noise and improved tissue differentiation compared to hybrid-iterative reconstruction.• Deep learning reconstruction of cerebral non-contrast CT results in better image quality in all aspects evaluated compared to model-based iterative reconstruction.• Deep learning reconstruction only needs a slight increase in reconstruction time compared to hybrid-iterative reconstruction, while model-based iterative reconstruction requires considerably longer processing time.

Deep learning reconstruction (DLR) algorithms have begun replacing iterative reconstruction (IR) in CT. Besides the potential to reduce noise, the application of artificial intelligence can also allow for "super resolution," allowing traditional CT systems to produce images with improved spatial resolution by training against higher-resolution images. This study aims to characterize the image quality produced by a DLR "super resolution" algorithm. This study compares Canon's Precise IQ Engine (PIQE) Cardiac, a DLR algorithm, with the AIDR Enhanced 3D iterative reconstruction (IR) algorithm using quantitative phantom-based image quality measurements. Scans of the American College of Radiology (ACR) Gammex 464 image quality phantom with and without a torso-sized elliptical ring were acquired using CTDIvol of 2.6, 15.7, and 32.9 mGy. Each experimental condition was scanned 20 times. Reconstructions used AIDR with a cardiac kernel and PIQE Cardiac with both 512×512 and 1024×1024 matrices. The task-based modulation transfer function (MTFtask) was calculated for Teflon, acrylic, polyethylene, and air inserts. Contrast-to-noise ratio (CNR) and low-contrast object specific CNR (CNRLO) were calculated for the 25-mm and 6-mm low-contrast disks in module 2 of the phantom. NPS and noise magnitude were calculated from the module's background. Both versions of PIQE showed improvement in MTFtask over AIDR for most frequencies, with PIQE 1024 outperforming PIQE 512. Larger differences were noted at higher contrast levels and with the smaller phantom. NPS curves for PIQE 512 and PIQE 1024 were nearly identical at all doses. AIDR showed a noise magnitude at least 44% higher than PIQE for all scenarios. The NPS shapes differed between AIDR and PIQE, producing noticeably different noise textures that were dose- and phantom size-dependent. CNRLO values were similar between PIQE 512 and PIQE 1024 and were higher than those of AIDR for most scenarios; however, AIDR outperformed PIQE for the 6-mm objects at the medium and high doses. Use of PIQE on the 25-mm disk permits dose reductions in the small phantom without loss of CNRLO. PIQE provides substantial image quality improvements over AIDR in phantom-based quantitative measurements. The potential for dose reduction is also noted. PIQE 1024 can be used to improve the spatial resolution over PIQE 512, with no apparent trade-offs in image quality.

In this work, we propose a deep learning approach for parallel magnetic resonance imaging (MRI) reconstruction, termed a variable splitting network (VS-Net), for an efficient, high-quality reconstruction of undersampled multi-coil MR data. We formulate the generalized parallel compressed sensing reconstruction as an energy minimization problem, for which a variable splitting optimization method is derived. Based on this formulation we propose a novel, end-to-end trainable deep neural network architecture by unrolling the resulting iterative process of such variable splitting scheme. VS-Net is evaluated on complex valued multi-coil knee images for 4-fold and 6-fold acceleration factors. We show that VS-Net outperforms state-of-the-art deep learning reconstruction algorithms, in terms of reconstruction accuracy and perceptual quality. Our code is publicly available at https://github.com/j-duan/VS-Net.

An ultra-simple and miniaturized spectrometer using an arbitrarily shaped pinhole as diffraction element reconstructs a broadband spectrum from the information of diffraction of monochromatic radiation by clever computational reconstruction algorithms. This circumvents complex calibration procedures and paves the way to cost-effective on-chip spectrometers combining fast acquisition without significant loss in spectral resolution.

Metallic samples are widely applied in modern industrial production. Due to non-uniformities in the stress load, such samples may become damaged and produce defects, which can cause unnecessary economic losses. In this paper, an online defect detection method is proposed for the quality monitoring of metallic plates. The research involves the design and optimisation of an electromagnetic tomography (EMT) sensor and the development of a fast tomography algorithm. Specifically, a planar array eddy current sensor is designed for in-situ structural health monitoring of metallic specimens. The parameters of the sensor are optimised using an orthogonal methodology and a response surface methodology to improve the uniformity of the sensitivity field. In addition, a second-order iterative Bregman reconstruction algorithm is investigated to reconstruct the defect image, which can improve the reconstruction speed for this ill-posed problem. Simulation and experimental results indicate that the proposed method can be applied to effectively evaluate the locations and sizes of defects in metallic specimens. The correlation coefficients of the reconstructed images using the proposed method are larger than 0.8. Compared with traditional reconstruction algorithms, the method proposed in this paper shows fast convergence speed and smaller estimation errors.

We introduce fast image reconstruction algorithms for emission tomography, which provide not only edge-preserved reconstructions, but also their edge maps simultaneously. To explicitly model the existence of edges, we use the binary line-process model, which is incorporated as a Gibbs prior in the context of a Bayesian maximum a posteriori framework. To efficiently handle the problem of mixed continuous and binary variable objectives, we use a deterministic annealing (DA) algorithm. Since the DA algorithm is computer-intensive and requires many iterations to converge, we apply a block-iterative method derived from the well-known ordered-subset principle. The block-iterative DA algorithm processes the data in blocks within each iteration, thereby accelerating the convergence speed of the standard DA algorithm by a factor proportional to the number of blocks. Our experimental results indicate that, with moderate numbers of blocks and properly chosen hyperparameters, the accelerated DA algorithm provides good reconstructions as well as edge maps with only a few iterations.

This paper describes an implementation of reception and real-time decoding of Asynchronous Sigma-Delta modulator (ASDM) encoded and wirelessly (On-off keying) transmitted signals. By using Fast reconstruction algorithm it is possible to reconstruct the original signal ≈ 25 times faster than by Classical reconstruction algorithm, if the signal lenght is 1 second, and ≈ 61, ≈ 130 and ≈ 228 times faster, if the length of the signal is 2, 4 and 8 seconds, respectively. Developed overall system consists of 3 parts - superheterodyne receiver, digitizer, PC with adaptive data reconstruction algorithm. Real-time decoding algorithm, which consist of data acquistion, filtering, processing, reconstruction and vizualization is implemented in LabView as a virtual instrument with graphical user interface (GUI).

Third generation CTs which acquire projection data using wide fan‐beam X‐rays that cover the cubic surface of the test object have two widely used tomogram reconstruction algorithms: i.e., the direct method and the arrangement method. The direct method applies fan‐like back projection directly in Cartesian coordinates. The arrangement method rearranges data to transform them into parallel fan beams. This paper proposes a new fast reconstruction algorithm in which polar coordinates are established that are spatially synchronous with the measurement geometry, reconstructed images are obtained by performing back projection in the polar coordinates and coordinate transformation is achieved from the polar coordinates to Cartesian coordinates. The features of this method are (1) computation is fast because the results of the many calculations are tabulated; (2) the remaining calculations are only product sums; and (3) it is easy to implement this method in hardware. The quality of images produced by the direct method is compared with that of the arrangement method and the present method using both computer simulation data and clinical data. The comparison shows that (1) spatial resolution of images produced by this method is better than that of the arrangement method, but almost equal to that of the direct method; and (2) the number of artifacts generated by calculations is small.

Through the reconstruction of the fluorescent probe distributions, fluorescence molecular tomography (FMT) can three-dimensionally resolve the molecular processes in small animals in vivo. In this paper, we propose an FMT reconstruction algorithm based on the iterated shrinkage method. By incorporating a surrogate function, the original optimization problem can be decoupled, which enables us to use the general sparsity regularization. Due to the sparsity characteristic of the fluorescent sources, the performance of this method can be greatly enhanced, which leads to a fast reconstruction algorithm. Numerical simulations and physical experiments were conducted. Compared to Newton method with Tikhonov regularization, the iterated shrinkage based algorithm can obtain more accurate results, even with very limited measurement data.

Depth and spectral imaging are essential technologies for a myriad of applications but have been conventionally studied as individual problems. Recent efforts have been made to optically encode spectral-depth (SD) information jointly in a single image sensor measurement, subsequently decoded by a computational algorithm. The performance of single snapshot SD imaging systems mainly depends on the optical modulation function, referred to as codification, and the computational methods used to recover the SD information from the coded measurement. The optical modulation has been conventionally realized using coded apertures (CAs), phase masks, prisms or gratings, active illumination, and many others. In this work, we propose an optical modulation (codification) strategy that employs a color-coded aperture (CCA) in conjunction with a time-varying phase-coded aperture and a spatially-varying pixel shutter, thus yielding an effective time-multiplexed coded aperture (TMCA). We show that the proposed TMCA entails a spatially-variant point spread function (PSF) for a constant depth in a scene, which, in turn, facilitates the distinguishability, and therefore, better recovery of the depth information. Further, the selective filtering of specific spectral bands by the CCA encodes relevant spectral information that is disentangled using a reconstruction algorithm. We leverage the advances of deep learning techniques to jointly learn the optical modulation and the computational decoding algorithm in an end-to-end (E2E) framework. We demonstrate via simulations and with a real testbed prototype that the proposed TMCA strategy outperforms state-of-the-art snapshot SD imaging alternatives in both spectral and depth reconstruction quality.

Deep learning has been successfully explored in addressing different multimedia topics recent years, ranging from object detection, semantic classification, entity annotation, to multimedia captioning, multimedia question answering and storytelling. Open source libraries and platforms such as Tensorflow, Caffe, MXnet significantly help promote the wide deployment of deep learning in solving real-world applications. On one hand, deep learning practitioners, while not necessary to understand the involved math behind, are able to set up and make use of a complex deep network. One recent deep learning tool based on Keras even provides the graphical interface to enable straightforward 'drag and drop' operation for deep learning programming. On the other hand, however, some general theoretical problems of learning such as the interpretation and generalization, have only achieved limited progress. Most deep learning papers published these days follow the pipeline of designing/modifying network structures - tuning parameters - reporting performance improvement in specific applications. We have even seen many deep learning application papers without one single equation. Theoretical interpretation and the science behind the study are largely ignored. While excited about the successful application of deep learning in classical and novel problems, we multimedia researchers are responsible to think and solve the fundamental topics in deep learning science. Prof. Guanrong Chen recently wrote an editorial note titled 'Science and Technology, not SciTech' [1]. This panel falls into similar discussion and aims to invite prestigious multimedia researchers and active deep learning practitioners to discuss the positioning of deep learning research now and in the future. Specifically, each panelist is asked to present their opinions on the following five questions: 1)How do you think the current phenomenon that deep learning applications are explosively growing, while the general theoretical problems remain slow progress? 2)Do you agree that deployment of deep learning techniques is getting easy (with a low barrier), while deep learning research is difficult (with a high barrier) 3)What do you think are the core problems for deep learning techniques? 4)What do you think are the core problems for deep learning science? 5)What's your suggestion on the multimedia research in the post-deep learning era?

In this paper, we propose a simulated annealing method for timeliness and energy aware deep learning (DL) job assignment. In the proposed method, we make a decision of server position for DL training jobs in order to consider both the timeliness of deep neural network (DNN) model updating and the associated energy consumption. For timeliness management, we design three penalty functions; step, linear, and exponential functions, so as to practically penalize the service quality degradation of DL inference due to delay of DL training. For energy management, we formulate the operation cost optimization problem considering energy consumption for DL training. By using these management schemes, our method is able to find the trade-off of the timeliness and energy consumption of DL training. Especially, in order to overcome the non-smoothness of the defined problem, we design a simulated annealing (SA) based metaheuristic which finds approximated optimal solution for DL training job assignment. For the performance evaluation, we show the preliminary experimental results of DL training jobs with AlexNet, Inception-V3, and ResNet on NVIDIA GPU devices.

Objectives:Radiation exposure has been a cause of concern in computed tomography imaging. Reducing radiation dose increases the image noise which can be compensated by using reconstruction techniques. Recently artificial intelligence-based reconstruction technique has been introduced. Therefore, the purpose of the study was to prospectively compare the image quality between Idose4 and Precise Image in normal BMI individuals.Methods:Sixty-six consecutive patients with a normal body habitus undergoing contrast-enhanced abdomen and pelvis scan were included in the study. All scans were performed using 100 kVp and tube current modulation. The acquired images were reconstructed to iDose4 and precise imaging. Quantitatively images were analyzed by placing regions of interest in different organs to estimate the image noise, signal-to-noise ratio, and contrast-to-noise ratio. Qualitative analysis was done by two radiologists on a five-point Likert scale.Results:Image noise was significantly reduced using Precise Image across the plain (9.11 ± 1.43 vs 8.18 ± 1.2), arterial (14.34 ± 2.1 vs 10.21 ± 1.5), and portovenous phase (14.78 ± 2.30 vs 11.97 ± 2.07) with maximum noise reduction in the arterial and portovenous phases. Signal-to-noise ratio and contrast-to-noise ratio was significantly improved in all the organs across the plain, arterial, and portovenous phases. Qualitative analysis showed no significant difference between Idose4 and Precise Image with regards to visualization of large vessels in the arterial and portovenous phases. However, precise image was graded better than Idose4 with respect to visualization/conspicuity, image noise, and artifacts.Conclusion:Precise Image can be useful in reducing the image noise and improving the signal-to-noise ratio and contrast-to-noise ratio in low-dose computed tomography protocol among normal BMI individuals.

Over the past decades, MRI has become increasingly important for diagnosing and longitudinally monitoring musculoskeletal disorders, with ongoing hardware and software improvements aiming to optimize image quality and speed. However, surging demand for musculoskeletal MRI and increased interest to provide more personalized care will necessitate a stronger emphasis on efficiency and specificity. Ongoing hardware developments include more powerful gradients, improvements in wide-bore magnet designs to maintain field homogeneity, and high-channel phased-array coils. There is also interest in low-field-strength magnets with inherently lower magnetic footprints and operational costs to accommodate global demand in middle- and low-income countries. Previous approaches to decrease acquisition times by means of conventional acceleration techniques (eg, parallel imaging or compressed sensing) are now largely overshadowed by deep learning reconstruction algorithms. It is expected that greater emphasis will be placed on improving synthetic MRI and MR fingerprinting approaches to shorten overall acquisition times while also addressing the demand of personalized care by simultaneously capturing microstructural information to provide greater detail of disease severity. Authors also anticipate increased research emphasis on metal artifact reduction techniques, bone imaging, and MR neurography to meet clinical needs.

Super-resolution reconstruction of pneumocystis carinii pneumonia images based on generative confrontation network