Convolutional neural networks for reconstruction of undersampled optical projection tomography data applied to in vivo imaging of zebrafish

Optical projection tomography (OPT) is a 3D mesoscopic imaging modality that can utilize absorption or fluorescence contrast. 3D images can be rapidly reconstructed from tomographic data sets sampled with sufficient numbers of projection angles using the Radon transform, as is typically implemented with optically cleared samples of the mm‐to‐cm scale. For in vivo imaging, considerations of phototoxicity and the need to maintain animals under anesthesia typically preclude the acquisition of OPT data at a sufficient number of angles to avoid artifacts in the reconstructed images. For sparse samples, this can be addressed with iterative algorithms to reconstruct 3D images from undersampled OPT data, but the data processing times present a significant challenge for studies imaging multiple animals. We show here that convolutional neural networks (CNN) can be used in place of iterative algorithms to remove artifacts—reducing processing time for an undersampled in vivo zebrafish dataset from 77 to 15 minutes. We also show that using CNN produces reconstructions of equivalent quality to compressed sensing with 40% fewer projections. We further show that diverse training data classes, for example, ex vivo mouse tissue data, can be used for CNN‐based reconstructions of OPT data of other species including live zebrafish.      

A Novel Iterative CT Reconstruction Approach Based on FBP Algorithm

The Filtered Back-Projection (FBP) algorithm and its modified versions are the most important techniques for CT (Computerized tomography) reconstruction, however, it may produce aliasing degradation in the reconstructed images due to projection discretization. The general iterative reconstruction (IR) algorithms suffer from their heavy calculation burden and other drawbacks. In this paper, an iterative FBP approach is proposed to reduce the aliasing degradation. In the approach, the image reconstructed by FBP algorithm is treated as the intermediate image and projected along the original projection directions to produce the reprojection data. The difference between the original and reprojection data is filtered by a special digital filter, and then is reconstructed by FBP to produce a correction term. The correction term is added to the intermediate image to update it. This procedure can be performed iteratively to improve the reconstruction performance gradually until certain stopping criterion is satisfied. Some simulations and tests on real data show the proposed approach is better than FBP algorithm or some IR algorithms in term of some general image criteria. The calculation burden is several times that of FBP, which is much less than that of general IR algorithms and acceptable in the most situations. Therefore, the proposed algorithm has the potential applications in practical CT systems.     

Sparse-view CT reconstruction based on multi-level wavelet convolution neural network

Sparse-view computed tomography (CT) is a recent approach to reducing the radiation dose in patients and speeding up the data acquisition. Consequently, sparse-view CT has been of particular interest among researchers within the CT community. Advanced reconstruction algorithms for sparse-view CT, such as iterative algorithms with total-variation (TV), have been studied along with the problem of increasing computational burden and the blurring of artifacts in the reconstructed images. Studies on deep-learning-based approaches applying U-NET have recently achieved remarkable outcomes in various domains including low-dose CT. In this study, we propose a new method for sparse-view CT reconstruction based on a multi-level wavelet convolutional neural network (MWCNN). First, a filtered backprojection (FBP) was used to reconstruct a sparsely sampled sinogram from 60, 120, and 180 projections. Subsequently, the sparse-view data obtained from FBP were fed to a deep-learning network, i.e., the MWCNN. Our network architecture combines a wavelet transform and modified U-NET without pooling. By replacing the pooling function with the wavelet transform, the receptive field is enlarged to improve the performance. We qualitatively and quantitatively evaluated the interpolation, iterative TV method, and standard U-NET in terms of a reduction in the streaking artifacts and a preservation of the anatomical structures. When compared with other methods, the proposed method showed the highest performance based on various evaluation parameters such as the structural similarity, root mean square error, and resolution. These results indicate that the MWCNN possesses a powerful potential for achieving a sparse-view CT reconstruction.     

Low-Dose Micro-CT Imaging for Vascular Segmentation and Analysis Using Sparse-View Acquisitions

The aim of this study is to investigate whether reliable and accurate 3D geometrical models of the murine aortic arch can be constructed from sparse-view data in vivo micro-CT acquisitions. This would considerably reduce acquisition time and X-ray dose. In vivo contrast-enhanced micro-CT datasets were reconstructed using a conventional filtered back projection algorithm (FDK), the image space reconstruction algorithm (ISRA) and total variation regularized ISRA (ISRA-TV). The reconstructed images were then semi-automatically segmented. Segmentations of high- and low-dose protocols were compared and evaluated based on voxel classification, 3D model diameters and centerline differences. FDK reconstruction does not lead to accurate segmentation in the case of low-view acquisitions. ISRA manages accurate segmentation with 1024 or more projection views. ISRA-TV needs a minimum of 256 views. These results indicate that accurate vascular models can be obtained from micro-CT scans with 8 times less X-ray dose and acquisition time, as long as regularized iterative reconstruction is used.     

An iterative deconvolution algorithm for image recovery in clinical CT: A phantom study

PurposeTo study the feasibility of using an iterative reconstruction algorithm to improve previously reconstructed CT images which are judged to be non-diagnostic on clinical review. A novel rapidly converging, iterative algorithm (RSEMD) to reduce noise as compared with standard filtered back-projection algorithm has been developed.Materials and methodsThe RSEMD method was tested on in-silico, Catphan^®500, and anthropomorphic 4D XCAT phantoms. The method was applied to noisy CT images previously reconstructed with FBP to determine improvements in SNR and CNR. To test the potential improvement in clinically relevant CT images, 4D XCAT phantom images were used to simulate a small, low contrast lesion placed in the liver.ResultsIn all of the phantom studies the images proved to have higher resolution and lower noise as compared with images reconstructed by conventional FBP. In general, the values of SNR and CNR reached a plateau at around 20 iterations with an improvement factor of about 1.5 for in noisy CT images. Improvements in lesion conspicuity after the application of RSEMD have also been demonstrated. The results obtained with the RSEMD method are in agreement with other iterative algorithms employed either in image space or with hybrid reconstruction algorithms.ConclusionsIn this proof of concept work, a rapidly converging, iterative deconvolution algorithm with a novel resolution subsets-based approach that operates on DICOM CT images has been demonstrated. The RSEMD method can be applied to sub-optimal routine-dose clinical CT images to improve image quality to potentially diagnostically acceptable levels.     

A newly-developed metal artifact reduction algorithm improves the visibility of oral cavity lesions on 320-MDCT volume scans

PurposeTo investigate whether a newly-developed single-energy metal artifact reduction (SEMAR) algorithm applied to images acquired on a 320-MDCT volume scanner reduces image artifacts from dental metal.MethodsWe inserted the lower right teeth covered with a dental metal alloy and crown in a skull phantom and performed single-volume scanning on a second-generation 320-MDCT scanner. A 12-mm diameter spherical lesion was placed either close to or far from the dental metal. The tube voltage and current were 120 kVp and 80 or 155 mA, respectively. Images were reconstructed with filtered back projection (FBP) or iterative reconstruction (IR), with or without SEMAR. We calculated the signal-to-artifact ratios (SAR) to quantify the visibility of the lesion. Two radiologists inspected 96 images (48 with lesion and 48 without) for the presence or absence of the lesion using a 5-point ordinal scale (1 = definitely absent to 5 = definitely present).ResultsOn images reconstructed with FPB and IR with SEMAR, streak artifacts from the dental metal were reduced substantially compared to images without SEMAR. At 155 mA with the lesion near the dental metal, the SARs were better on FBP and IR images (FBP: 1.7 and 0.5 with and without SEMAR, respectively; IR: 1.6 and 0.9 with and without SEMAR, respectively). The observer visual scores improved with SEMAR (FBP: 4.2 and 3.2 with and without SEMAR, respectively; IR: 4.2 and 3.0).ConclusionThe SEMAR algorithm reduces dental metal artifacts and improves lesion detectability and image quality in patients with oral cavity lesions.     

Deep learning-based reconstruction in ultra-high-resolution computed tomography: Can image noise caused by high definition detector and the miniaturization of matrix element size be improved?

PurposeThis study aimed to assess the noise characteristics of ultra-high-resolution computed tomography (UHRCT) with deep learning-based reconstruction (DLR).MethodsTwo different diameters of water phantom were scanned with three different resolution acquisition modes. Images were reconstructed by filtered back projection (FBP), hybrid iterative reconstruction (hybrid-IR), and DLR. Image noise analysis was performed with noise magnitude, peak frequency (f_p) of the noise power spectrum (NPS), and the square root of the area under the curve (√AUC_NPS) for the NPS curve.ResultsThe noise magnitude was up to 3.30 times higher for the FBP acquired in SHR mode than that for the NR mode.The f_p values of the FBP were 0.20–0.21, 0.34–0.36, and 0.34–0.37 cycles/mm for normal resolution (NR), high resolution (HR), and super high resolution (SHR) mode, respectively. The f_p of hybrid-IR was 0.16–0.19, 0.21–0.26, and 0.23–0.26 cycles/mm for NR, HR, and SHR mode, respectively. The f_p of DLR was 0.21–0.32 and 0.22–0.33 cycles/mm for HR and SHR mode, respectively.√AUC_NPS showed that the highest value in FBP images of the SHR mode was up to 1.89 times that of the NR mode. DLR in the HR and SHR modes showed high noise reduction while suppressing f_p shift with respect to FBP.ConclusionsThe new DLR algorithm could be a solution to the noise increase due to the high-definition detector elements and the small reconstruction matrix element size.     

The effect of overabundant projection directions on 3D reconstruction algorithms

The experimental process of collecting images from macromolecules in an electron microscope is such that it does not allow for prior specification of the angular distribution of the projection images. As a consequence, an uneven distribution of projection directions may occur. Concerns have been raised recently about the behavior of 3D reconstruction algorithms for the case of unevenly distributed projections. It has been illustrated on experimental data that in the case of a heavily uneven distribution of projection directions some algorithms tend to elongate the reconstructed volumes along the overloaded direction so much as to make a quantitative biological analysis impossible. In answer to these concerns we have developed a strategy for quantitative comparison and optimization of 3D reconstruction algorithms. We apply this strategy to quantitatively analyze algebraic reconstruction techniques (ART) with blobs, simultaneous iterative reconstruction techniques (SIRT) with voxels, and weighted backprojection (WBP). We show that the elongation artifacts that had been previously reported can be strongly reduced. With our specific choices for the free parameters of the three algorithms, WBP reconstructions tend to be inferior to those obtained with either SIRT or ART and the results obtained with ART are comparable to those with SIRT, but at a very small fraction of the computational cost of SIRT.     

Utility of fast non-local means (FNLM) filter for detection of pulmonary nodules in chest CT for pediatric patient

PurposeThis study was aimed to evaluate the utility based on imaging quality of the fast non-local means (FNLM) filter in diagnosing lung nodules in pediatric chest computed tomography (CT).MethodsWe retrospectively reviewed the chest CT reconstructed with both filtered back projection (FBP) and iterative reconstruction (IR) in pediatric patients with metastatic lung nodules. After applying FNLM filter with six h values (0.0001, 0.001, 0.01, 0.1, 1, and 10) to the FBP images, eight sets of images including FBP, IR, and FNLM were analyzed. The image quality of the lung nodules was evaluated objectively for coefficient of variation (COV), contrast to noise ratio (CNR), and point spread function (PSF), and subjectively for noise, sharpness, artifacts, and diagnostic acceptability.ResultsThe COV was lowest in IR images and decreased according to increasing h values and highest with FBP images (P < 0.001). The CNR was highest with IR images, increased according to increasing h values and lowest with FBP images (P < 0.001). The PSF was lower only in FNLM filter with h value of 0.0001 or 0.001 than in IR images (P < 0.001). In subjective analysis, only images of FNLM filter with h value of 0.0001 or 0.001 rarely showed unacceptable quality and had comparable results with IR images. There were less artifacts in FNLM images with h value of 0.0001 compared with IR images (p < 0.001).ConclusionFNLM filter with h values of 0.0001 allows comparable image quality with less artifacts compared with IR in diagnosing metastatic lung nodules in pediatric chest CT.     

Image Quality in Children with Low-Radiation Chest CT Using Adaptive Statistical Iterative Reconstruction and Model-Based Iterative Reconstruction

Objective

To evaluate noise reduction and image quality improvement in low-radiation dose chest CT images in children using adaptive statistical iterative reconstruction (ASIR) and a full model-based iterative reconstruction (MBIR) algorithm.

Methods

Forty-five children (age ranging from 28 days to 6 years, median of 1.8 years) who received low-dose chest CT scans were included. Age-dependent noise index (NI) was used for acquisition. Images were retrospectively reconstructed using three methods: MBIR, 60% of ASIR and 40% of conventional filtered back-projection (FBP), and FBP. The subjective quality of the images was independently evaluated by two radiologists. Objective noises in the left ventricle (LV), muscle, fat, descending aorta and lung field at the layer with the largest cross-section area of LV were measured, with the region of interest about one fourth to half of the area of descending aorta. Optimized signal-to-noise ratio (SNR) was calculated.

Result

In terms of subjective quality, MBIR images were significantly better than ASIR and FBP in image noise and visibility of tiny structures, but blurred edges were observed. In terms of objective noise, MBIR and ASIR reconstruction decreased the image noise by 55.2% and 31.8%, respectively, for LV compared with FBP. Similarly, MBIR and ASIR reconstruction increased the SNR by 124.0% and 46.2%, respectively, compared with FBP.

Conclusion

Compared with FBP and ASIR, overall image quality and noise reduction were significantly improved by MBIR. MBIR image could reconstruct eligible chest CT images in children with lower radiation dose.     

Single-shot optical projection tomography for high-speed volumetric imaging of dynamic biological samples

A single-shot adaptation of Optical Projection Tomography (OPT) for high-speed volumetric snapshot imaging of dynamic mesoscopic biological samples is presented. Conventional OPT has been applied to in vivo imaging of animal models such as D. rerio, but the sequential acquisition of projection images typically requires samples to be immobilized during the acquisition. A proof-of-principle system capable of single-shot tomography of a ~1 mm³ volume is presented, demonstrating camera-limited rates of up to 62.5 volumes/s, which has been applied to 3D imaging of a freely swimming zebrafish embryo. This is achieved by recording eight projection views simultaneously on four low-cost CMOS cameras. With no stage required to rotate the sample, this single-shot OPT system can be implemented with a component cost of under £5000. The system design can be adapted to different sized fields of view and may be applied to a broad range of dynamic samples, including high throughput flow cytometry applied to model organisms and fluid dynamics studies.     

A Novel Weighted Total Difference Based Image Reconstruction Algorithm for Few-View Computed Tomography

In practical applications of computed tomography (CT) imaging, due to the risk of high radiation dose imposed on the patients, it is desired that high quality CT images can be accurately reconstructed from limited projection data. While with limited projections, the images reconstructed often suffer severe artifacts and the edges of the objects are blurred. In recent years, the compressed sensing based reconstruction algorithm has attracted major attention for CT reconstruction from a limited number of projections. In this paper, to eliminate the streak artifacts and preserve the edge structure information of the object, we present a novel iterative reconstruction algorithm based on weighted total difference (WTD) minimization, and demonstrate the superior performance of this algorithm. The WTD measure enforces both the sparsity and the directional continuity in the gradient domain, while the conventional total difference (TD) measure simply enforces the gradient sparsity horizontally and vertically. To solve our WTD-based few-view CT reconstruction model, we use the soft-threshold filtering approach. Numerical experiments are performed to validate the efficiency and the feasibility of our algorithm. For a typical slice of FORBILD head phantom, using 40 projections in the experiments, our algorithm outperforms the TD-based algorithm with more than 60% gains in terms of the root-mean-square error (RMSE), normalized root mean square distance (NRMSD) and normalized mean absolute distance (NMAD) measures and with more than 10% gains in terms of the peak signal-to-noise ratio (PSNR) measure. While for the experiments of noisy projections, our algorithm outperforms the TD-based algorithm with more than 15% gains in terms of the RMSE, NRMSD and NMAD measures and with more than 4% gains in terms of the PSNR measure. The experimental results indicate that our algorithm achieves better performance in terms of suppressing streak artifacts and preserving the edge structure information of the object.     

Digital holographic microtomography for high‐resolution refractive index mapping of live cells

Quantification of three‐dimensional (3D) refractive index (RI) with sub‐cellular resolution is achieved by digital holographic microtomography (DHμT) using quantitative phase images measured at multiple illumination angles. The DHμT system achieves sensitive and fast phase measurements based on iterative phase extraction algorithm and asynchronous phase shifting interferometry without any phase monitoring or active control mechanism. A reconstruction algorithm, optical diffraction tomography with projection on convex sets and total variation minimization, is implemented to substantially reduce the number of angular scattered fields needed for reconstruction without sacrificing the accuracy and quality of the reconstructed 3D RI distribution. Tomogram of a living CA9‐22 cell is presented to demonstrate the performance of the method. Further, a statistical analysis of the average RI of the nucleoli, the nucleus excluding the nucleoli and the cytoplasm of twenty CA9‐22 cells is performed. (© 2013 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)     

Set-up error validation with EPID images: Measurements vs Egs_cbct simulation

AimIn this study, the egs_cbct code’s ability to replicate an electronic portal imaging device (EPID) is explored.BackgroundWe have investigated head and neck (H&N) setup verification on an Elekta Precise linear accelerator. It is equipped with an electronic portal imaging device (EPID) that can capture a set of projection images over different gantry angles.Methods and materialsCone-beam computed tomography (CBCT) images were reconstructed from projection images of two different setup scenarios. Projections of an Anthropomorphic Rando head phantom were also simulated by using the egs_cbct Monte Carlo code for comparison with the measured projections.Afterwards, CBCT images were reconstructed from this data. Image quality was evaluated against a metric defined as the image acquisition interval (IAI). It determines the number of projection images to be used for CBCT image reconstruction.ResultsFrom this results it was established that phantom shifts could be determined within 2 mm and rotations within one degree accuracy using only 20 projection images (IAI = 10 degrees). Similar results were obtained with the simulated data.ConclusionIn this study it is demonstrated that a head and neck setup can be verified using substantially fewer projection images. Bony landmarks and air cavities could still be observed in the reconstructed Rando head phantom. The egs_cbct code can be used as a tool to investigate setup errors without tedious measurements with an EPID system.     

Image quality assessment of an iterative reconstruction algorithm applied to abdominal CT imaging

PurposeTo compare the noise and accuracy on images of the whole porcine liver acquired with iterative reconstruction (IMR, Philips Healthcare, Cleveland, OH, USA) and filtered back projection (FBP) methods.Materials and methodsWe used non-enhanced porcine liver to simulate the human liver and acquired it 100 times to obtain the average FBP value as the ground-truth. The mean and the standard deviation (“inter-scan SD”) of the pixel values on the 100 image acquisitions were calculated for FBP and for three levels of IMR (L1, L2, and L3). We also calculated the noise power spectrum (NPS) and the normalized NPS for the 100 image acquisitions.ResultsThe spatial SD for the porcine liver parenchyma on these slices was 9.92, 4.37, 3.63, and 2.30 Hounsfield units with FBP, IMR-L1, IMR-L2, and IMR-L3, respectively. The detectability of small faint features was better on single IMR than single FBP images. The inter-scan SD value for IMR-L3 images was 53% larger at the liver edges than at the liver parenchyma; it was only 10% larger on FBP images. Assessment of the normalized NPS showed that the noise on IMR images was comprised primarily of low-frequency components.ConclusionIMR images yield the same structure informations as FBP images and image accuracy is maintained. On level 3 IMR images the image noise is more strongly suppressed than on IMR images of the other levels and on FBP images.     

System Matrix Analysis for Computed Tomography Imaging

In practical applications of computed tomography imaging (CT), it is often the case that the set of projection data is incomplete owing to the physical conditions of the data acquisition process. On the other hand, the high radiation dose imposed on patients is also undesired. These issues demand that high quality CT images can be reconstructed from limited projection data. For this reason, iterative methods of image reconstruction have become a topic of increased research interest. Several algorithms have been proposed for few-view CT. We consider that the accurate solution of the reconstruction problem also depends on the system matrix that simulates the scanning process. In this work, we analyze the application of the Siddon method to generate elements of the matrix and we present results based on real projection data.     

Analysis of symmetry and three-dimensional reconstruction of thin gp32*I crystals.

Thin, multilayered crystals of gp32*I were analyzed by negative stain electron microscopy and image processing. Images of untilted crystals exhibited different projection symmetries and structural motifs. Systematic analysis of these images categorized the projections into four types. Areas producing the type 1 projection were reconstructed in three-dimensions from four tilt series containing 111 images. The three-dimensional data has excellent p121 plane group symmetry and reveals that the gp32*I molecule contains two large domains linked together by a small domain. Computer simulations utilizing projection data suggested that the type 2 and 3 projections arise from superposition of type 1 projections related by a 21 screw axis along the projection axis. The three-dimensional reconstruction was utilized in a final simulation that explained the occurrence of the fourth type of projection. This work provides a firm foundation for future high-resolution analysis of the crystal by electron cryomicroscopy.     

X-ray scatter data for flat-panel detector CT

In modern X-ray computed tomography (CT) a trend to increased volume coverage by using multi-row detectors is apparent. Flat-panel detector CT (FPD-CT) systems provide an even larger field of measurement which, however, results in an increased scatter fraction. We investigated the scatter intensities registered in the case of FPD-CT. A hybrid model for the simulation of scatter combining deterministic and Monte Carlo methods was used for the scatter calculations. The influence of imaging parameters on the registered scatter intensity was examined both in single projections and reconstructed images. Scatter-to-primary ratios (SPRs) are given for various values of object thickness, field size, object-to-detector distance, incident energy and projection angle. For the simulations, homogeneous water phantoms and realistic patient data sets were used to produce scatter data representative for clinical situations. The SPR increases with object size, collimation and z-extent resulting in SPR ≫ 1 and respective scatter artifacts in the reconstructed images. In contrary, the scatter intensity decreases non-linearly with the object-to-detector distance. The angular and spatial distributions of scatter form a flat function as compared to the distribution of the primary signal. Single scatter appears to determine the distribution and magnitude of the total-scatter intensity at the detector.     

Automatic exposure control at single- and dual-heartbeat CTCA on a 320-MDCT volume scanner: Effect of heart rate,exposure phase window setting,and reconstruction algorithm

PurposeTo investigate whether electrocardiogram (ECG)-gated single- and dual-heartbeat computed tomography coronary angiography (CTCA) with automatic exposure control (AEC) yields images with uniform image noise at reduced radiation doses.Materials and methodsUsing an anthropomorphic chest CT phantom we performed prospectively ECG-gated single- and dual-heartbeat CTCA on a second-generation 320-multidetector CT volume scanner. The exposure phase window was set at 75%, 70–80%, 40–80%, and 0–100% and the heart rate at 60 or 80 or corr80 bpm; images were reconstructed with filtered back projection (FBP) or iterative reconstruction (IR, adaptive iterative dose reduction 3D). We applied AEC and set the image noise level to 20 or 25 HU. For each technique we determined the image noise and the radiation dose to the phantom center.ResultsWith half-scan reconstruction at 60 bpm, a 70–80% phase window- and a 20-HU standard deviation (SD) setting, the imagenoise level and -variation along the z axis manifested similar curves with FBP and IR. With half-scan reconstruction, the radiation dose to the phantom center with 70–80% phase window was 18.89 and 12.34 mGy for FBP and 4.61 and 3.10 mGy for IR at an SD setting SD of 20 and 25 HU, respectively. At 80 bpm with two-segment reconstruction the dose was approximately twice that of 60 bpm at both SD settings. However, increasing radiation dose at corr80 bpm was suppressed to 1.39 times compared to 60 bpm.ConclusionAEC at ECG-gated single- and dual-heartbeat CTCA controls the image noise at different radiation dose.     

Fast and accurate three-dimensional reconstruction from projections with random orientations via radon transforms

A new algorithm for three-dimensional reconstruction from randomly oriented projections has been developed. The algorithm recovers the 3D Radon transform from the 2D Radon transforms (sinograms) of the projections. The structure in direct space is obtained by an inversion of the 3D Radon transform. The mathematical properties of the Radon transform are exploited to design a special filter that can be used to correct inconsistencies in a data set and to fill the gaps in the Radon transform that originate from missing projections. Several versions of the algorithm have been implemented, with and without a filter and with different interpolation methods for merging the sinograms into the 3D Radon transform. The algorithms have been tested on analytical phantoms and experimental data and have been compared with a weighted back projection algorithm (WBP). A quantitative analysis of phantoms reconstructed from noise-free and noise-corrupted projections shows that the new algorithms are more accurate than WBP when the number of projections is small. Experimental structures obtained by the new methods are strictly comparable to those obtained by WBP. Moreover, the algorithm is more than 10 times faster than WPB when applied to a data set of 1000-5000 projections. Copyright 1999 Academic Press.