Tomographic Reconstruction Research Articles

Advanced image reconstruction for electrostatic tomography in gas-solid two-phase flow based on convolutional autoencoder neural network

The image reconstruction of flowing charged particles in gas-solid two-phase (GSTP) flow can be achieved through electrostatic tomography (EST). Accurate image reconstruction is crucial for detecting the movement patterns of the particles. In order to improve the quality of reconstructed images, a unique convolutional autoencoder neural network (CANN) is proposed. This study uses an image set generated by the linear backprojection (LBP) algorithm to train the CANN, which consists of an encoder and a decoder. The encoder utilizes convolutional and max-pooling layers to reduce the dimensionality of the images and extract key features, while the decoder restores and reconstructs the images through up-sampling and convolutional operations to closely approximate the reference image. To prevent overfitting, dropout layers are introduced after each max-pooling layer in the encoder. To verify the anti-noise capability of the network, Gaussian white noise ranging from 10 dB to 20 dB is added to the test set. The proposed CANN has been validated through simulations and experiments, demonstrating its effectiveness in overcoming noticeable artifacts and noise in reconstructed images when identifying GSTP flow patterns. Furthermore, it shows significant enhancements in imaging outcomes compared to conventional image reconstruction techniques and some current deep learning algorithms.

Precision compensation method for large-scale measurements based on full-space refractive index field reconstruction with vibratory mirror background oriented schlieren (VMBOS)

With the development of the large-scale equipment manufacturing industry, such as aircraft component assembly, wind turbine rotor blade measurement and large ship assembly the demand for optical measurement accuracy in large-dimensional spaces within industrial environments has become increasingly stringent. However, due to the presence of internal interference sources, light will no longer propagate in a straight line, thereby introducing measurement errors that cannot be ignored. To compensate for the measurement error, the refractive index field distribution within the large-dimensional measurement space must be reconstructed. Therefore, this paper proposed a vibratory mirror background oriented schlieren (VMBOS) based on vibratory scanning. This method scans the measurement space through the rotation of vibratory mirror to obtain light deflection angles at different directions. Subsequently, the refractive index field of the large-dimensional space is reconstructed by using the tomographic reconstruction algorithm. Then the optical path of the light in the optical measurement can be corrected through the Hamiltonian ray tracing algorithm, thereby achieving compensation for the measurement error. Finally, the VMBOS method has been verified through both simulation and experimental methods. The experimental results demonstrated that the method proposed in this paper can be applied to optical path correction and error compensation in large scale spaces.

DynaCT biliary reconstruction via a 3D C-arm cholangiography system: clinical application in hepatolithiasis.

Dyna computed tomography (DynaCT) is an innovative clinical imaging tool used to obtain three-dimensional (3D) images of biliary structures via the Artis Zee DSA system (SIEMENS Company, Germany). DynaCT is a type of 3D cone beam computed tomography (CBCT) reconstruction produced from a two-dimensional (2D) cholangiography system by rotating the C-arm without moving the patient. The aim of this study was to evaluate the technical approach and application value of DynaCT to diagnosis hepatolithiasis and biliary stenosis. This was a retrospective single-centre series of 37 hepatolithiasis patients with tubes receiving one-step percutaneous transhepatic cholangioscopic lithotripsy (one-step PTCSL) between October 2021 and October 2022: twenty-one patients were guided by CT (CT group) and sixteen by DynaCT biliary reconstruction (DynaCT group). We compared DynaCT biliary reconstruction technology with computed tomography (CT) in the application of bile ducts. DynaCT biliary reconstruction was successfully performed in 37 patients. Biliary stenosis, including anatomy, morphology, and size, was visualized via DynaCT. Compared with the CT group, the DynaCT group was characterized by significantly more target biliary branches with stones (92 vs. 48, P < 0.05), a higher percentage of secondary stenosis (75.76% vs. 24.24%, p < 0.05), a greater percentage of biliary infection (37.5% vs. 9.5%, P = 0.041), a shorter overall stone clearance time (26.38 ± 13.49 vs. 52.67 ± 30.10, P = 0.001), and a lower rate of reoperation for residual stones (25.00% vs. 66.67%, P = 0.012). DynaCT had a lower contrast agent (25.61 ± 5.13 vs. 42.69 ± 11.15, p < 0.05). However, DynaCT increased radiation exposure (38.12 ± 10.59 vs. 25.79 ± 4.76, p < 0.05). There were no significant differences between the two groups regarding the clearance ratio of the calculus or several postoperative complications. DynaCT for biliary reconstruction has the potential to be a powerful evaluation tool for one-step PTCSL surgery and could lead to new possibilities for hepatobiliary surgery. DynaCT was used for the first time in patients with hepatolithiasis and biliary stenosis. Compared with CT, DynaCT for biliary reconstruction results in higher-quality 3D biliary, blood vessel and liver images. On the basis of the DynaCT biliary model, one-step PTCSL has the potential to improve the stone clearance ratio and shorten the stone clearance time and reoperation ratio.

ElectronCT - an imaging technique using very-high energy electrons

The electronCT technique is an imaging method based on the multiple Coulomb scattering of relativistic electrons and has potential applications in medical and industrial imaging. It utilizes a pencil beam of electrons in the very high energy electron (VHEE, 50–250 MeV) range and a single detection layer for the determination of the beam profile. The technique constitutes a projectional, two-dimensional imaging method and thus also qualifies for the tomographic reconstruction of samples. Given the simplicity of the technical setup and its location behind the sample, the electronCT technique has potential synergies with VHEE radiotherapy, making use of the same electron source for both treatment and diagnostics and thus being a candidate for in situ imaging and patient localization. At the same time, several technical challenges arise from the measurement technique when applied for the imaging of living beings. Measurements performed at the ARES linear particle accelerator at an electron energy of 155 MeV using a mouse phantom and a Timepix3 silicon pixel detector assembly demonstrate the feasibility of this technique. Both projectional and tomographic reconstructions are presented and the potential and limits of the technology are discussed.

Optical Time-Domain Quantum State Tomography on a Subcycle Scale

Following recent progress in the experimental application of electro-optic sampling to the detection of the quantum fluctuations of the electromagnetic-field ground state and ultrabroadband squeezed states on a subcycle scale, we propose an approach to elevate broadband electro-optic sampling from a spectroscopic method to a full quantum tomography scheme, able to reconstruct a free-space quantum state directly in the time domain. By combining two recently developed methods to theoretically describe quantum electro-optic sampling, we analytically relate the photon-count probability distribution of the electro-optic signal to a transformed phase-space quasiprobability distribution of the sampled quantum state as a function of the time delay between the sampled midinfrared pulsed state and an ultrabroadband near-infrared probe pulse. We catalog and analyze sources of noise and show that in quantum electro-optic sampling with an ultrabroadband probe pulse one can expect to observe thermalization due to entanglement breaking. Mitigation of the thermalization noise enables a tomographic reconstruction of broadband quantum states while granting access to its dynamics on a subcycle scale. Published by the American Physical Society 2024

Bias and precision of SPECT-based 177Lu activity-concentration estimation using a ring-configured solid-state versus a dual-headed anger system

BackgroundThe aim was to compare bias and precision for 177Lu-SPECT activity-concentration estimation using a dual-headed Anger SPECT system and a ring-configured CZT SPECT system. This was investigated for imaging at 208 keV and 113 keV, respectively.MethodsPhantom experiments were performed on a GE Discovery 670 system with 5/8’’ NaI(Tl) crystal (dual-headed Anger system) and a GE StarGuide (ring-configured CZT system). Six spheres (1.2 mL to 113 mL) in a NEMA PET body phantom were filled with 99mTc and 177Lu, separately. Mean relative errors and coefficients of variation (CV) in estimated sphere activity concentration were studied over six timeframes of 10 min each for the two systems. For 177Lu, similar acquisitions were also performed for an anthropomorphic phantom with two spheres (10 mL and 25 mL) in a liver with non-radioactive background and a sphere-to-background ratio of 15:1. Tomographic reconstruction was performed using OS-EM with 10 subsets with compensation for attenuation, scatter, and distance-dependent spatial resolution. For the Anger system, up to 40 iterations were used and for the ring-configured CZT system up to 30 iterations were used.ResultsThe two systems showed similar mean relative errors and CVs for 177Lu when using an energy window around 208 keV, while the ring-configured system demonstrated a lower bias for a similar CV compared to the Anger system for 99mTc and for 177Lu when using an energy window around 113 keV. However, total activity in the phantom tended to be overestimated in both systems for these cases.ConclusionsThe ring-configured CZT system is a viable alternative to the dual-headed Anger system equipped with medium-energy collimators for 177Lu-SPECT and shows a potential advantage for activity-concentration estimation when operated at 113 keV. However, further consideration of the preservation of total activity is warranted.

Analytic sparse-view CBCT reconstruction using a novel sinogram restoration method based on higher-order Fourier harmonic peaks

This study presents an analytic sparse-view cone-beam computed tomography (CBCT) reconstruction method using a novel sinogram restoration technique based on higher-order Fourier harmonic peaks to obtain CBCT images of reasonable quality with a reduced radiation dose. The proposed method involves four main steps: (1) acquiring sparse-view (P144) sinogram from a CBCT system, (2) upsampling (P720) the original sparse-view sinogram with equally spaced zero padding, (3) separating higher-order Fourier harmonic peaks of the upsampled sinogram using a band-pass filter, and (4) restoring the sinogram of the higher-order harmonic peaks by taking its magnitude, followed by computationally cost-efficient filtered-backprojection (FBP)-based CBCT reconstruction and denoising processes. The notation P144 indicates that the number of projections is 144. To verify the efficacy of the proposed method, an experiment was performed on a chest phantom using a prototype CBCT system comprising an X-ray tube (90 kV p and 40 mA), a large-area flat-panel detector (388 μm pixel size), and a rotational gantry. The quality of the restored sparse-view CBCT images was quantitatively evaluated in terms of the image intensity profile, universal quality index (UQI), and peak signal-to-noise ratio (PSNR). The results indicate that the proposed sinogram restoration method effectively recovered zero-padded areas in the original upsampled sinogram with image intensities nearly identical to those of the reference dense-view (P720) sinogram, which demonstrates the efficacy of the proposed sinogram restoration method. The quality of the sparse-view CBCT image restored with higher-order harmonic peaks was close to that of the reference dense-view CBCT image, maintaining the image quality. The UQI and PSNR values of the restored sparse-view CBCT image were 0.93 and 34.84, respectively, which are approximately 1.3 and 1.2 times greater than those of the original sparse-view CBCT image, respectively, improving the image quality. Consequently, the proposed method has the potential to resolve the issues caused by sparse-view sampling in FBP-based CBCT.

Age estimation on post-mortem CT based on pelvic bone mineral density measurement and the state of putrefaction: a multivariate method.

Age-at-death estimation is an important issue in forensic medicine and anthropology. Initially, methods relied on morphological criteria, but with the advancement of radiology, new techniques such as morphological studies on multi-slice computed tomography (CT) reconstructions have emerged. Recent studies have shown promising results by investigating the correlation between age and bone mineral density (BMD). However, there is currently a lack of data on post-mortem CTs (PMCT) involving decomposed bodies, and limited information exists regarding changes in Hounsfield Units measurement in a post-mortem context. In light of these gaps, our study aimed to examine the relationship between age at death and pubic and ilium BMD using a sample of forensic bodies. We also aimed to determine whether post-mortem processes, such as putrefaction, could interfere with this correlation. Our retrospective analysis encompassed 637 PMCTs conducted before medicolegal autopsies at Tours University Hospital. Utilizing simple and multiple linear regressions, we investigated the correlation between age and pubic and ilium BMD, as well as the relationship between BMD and the radiologic alteration index (RAI), a scale employed to quantify the degree of putrefaction. Our findings indicate promising outcomes in age-at-death estimation using pubic and/or ilium BMD for bodies exhibiting no or moderate decomposition (RAI < 80), particularly among individuals under 40years old. However, for highly decomposed corpses (RAI ≥ 80), the presence of gas infiltration significantly influences the BMD of both the ilium and pubis. Consequently, we advocate for the incorporation of the RAI score into the age estimation equation to enhance the accuracy of our results in such cases. Further investigation involving a larger cohort of decomposed bodies could facilitate refinement and validation of our method within this specific population.

Allometry of human calvaria bones during development from birth to 8 years of age shows a nonlinear growth pattern

Pediatric skulls change rapidly in size and shape during development, especially for children up to 8 years of age. This project was developed to address the gap in understanding of the three-dimensional growth parameters of the human skull during this period and the impact these growth patterns have on fontanelle closure and suture formation. This study offers novel data on the dynamic changes in the anatomy of the skull with the intention of providing better guidance for pediatric surgical care. Craniometric landmarks defined on three-dimensional computed tomography reconstructions were used to map skull development in children aged 0 to 8 years old. A total of 364 datasets were analyzed and statistically representative 3D skulls with anatomical craniometric features such as head shape, bone size, suture and fontanelle closure time were generated for 17 age groups spanning birth to 8 years of age to provide a comprehensive neuroanatomical understanding of how the pediatric skull changes over time. This study indicates that the cranial bones follow a non-linear growth pattern, with the occipital and frontal bones driving the directionality of fontanelle closure and delivers a 3D visualization of the developmental characteristics of the skull providing a landmark resource for understanding the growth dynamics of the human skull. While clinical measurements remain valid approaches for the planning of surgical interventions, these 3D models may provide a more accurate planning paradigm.

Neural-field-based image reconstruction for bioluminescence tomography

Neural-field-based image reconstruction for bioluminescence tomography

Electron tomography visualization of HIV-1 virions trapped by fusion inhibitors to host cells in infected tissues.

HIV-1 delivers its genetic material to infect a cell after fusion of the viral and host cell membranes, which takes place after the viral envelope (Env) binds host receptor and co-receptor proteins. Binding of host receptor CD4 to Env results in conformational changes that allow interaction with a host co-receptor (CCR5 or CXCR4). Further conformational rearrangements result in an elongated pre-hairpin intermediate structure in which Env is anchored to the viral membrane by its transmembrane region and to the host cell membrane by its fusion peptide. Although budding virions can be readily imaged by electron tomography (ET) of HIV-1-infected tissues and cultured cells, virions that are fusing (attached to host cells via pre-hairpin intermediates) are not normally visualized, perhaps because the process of membrane fusion is too fast to capture by ET. To image virions during fusion, we used fusion inhibitors to prevent downstream conformational changes in Env that lead to membrane fusion, thereby trapping HIV-1 virions linked to target cells by pre-hairpin intermediates. ET of HIV-1 pseudovirions bound to CD4+/CCR5+ TZM-bl cells revealed presumptive pre-hairpin intermediates as 2-4 narrow spokes linking a virion to the cell surface. To extend these results to a more physiological setting, we used ET to image tissues and organs derived from humanized bone marrow/liver/thymus mice infected with HIV-1 and then treated with CPT31, a high-affinity D-peptide fusion inhibitor linked to cholesterol. Trapped HIV-1 virions were found in all tissues studied (small intestine, mesenteric lymph nodes, spleen, and bone marrow), and spokes representing pre-hairpin intermediates linking trapped virions to cell surfaces were similar in structure and number to those seen in the previous pseudovirus and cultured cell ET study.IMPORTANCETrapped and untrapped HIV-1 virions, both mature and immature, were distinguished by localizing spokes via 3D tomographic reconstructions of HIV-1 infected and fusion-inhibitor-treated tissues of humanized mice. The findings of trapped HIV-1 virions in all tissues examined demonstrate a wide distribution of the CPT31 inhibitor, a desirable property for a potential therapeutic. In addition, the presence of virions trapped by spokes, particularly in vascular endothelial cells, demonstrates that the fusion inhibitors can be used as markers for potential HIV-1-target cells within tissues, facilitating the mapping of HIV-1 target cells within the complex cellular milieu of infected tissues.

Evaluation of a Vendor-Agnostic Deep Learning Model for Noise Reduction and Image Quality Improvement in Dental CBCT.

To assess the impact of a vendor-agnostic deep learning model (DLM) on image quality parameters and noise reduction in dental cone-beam computed tomography (CBCT) reconstructions. This retrospective study was conducted on CBCT scans of 93 patients (41 males and 52 females, mean age 41.2 years, SD 15.8 years) from a single center using the inclusion criteria of standard radiation dose protocol images. Objective and subjective image quality was assessed in three predefined landmarks through contrast-to-noise ratio (CNR) measurements and visual assessment using a 5-point scale by three experienced readers. The inter-reader reliability and repeatability were calculated. Eighty patients (30 males and 50 females; mean age 41.5 years, SD 15.94 years) were included in this study. The CNR in DLM reconstructions was significantly greater than in native reconstructions, and the mean CNR in regions of interest 1-3 (ROI1-3) in DLM images was 11.12 ± 9.29, while in the case of native reconstructions, it was 7.64 ± 4.33 (p < 0.001). The noise level in native reconstructions was significantly higher than in the DLM reconstructions, and the mean noise level in ROI1-3 in native images was 45.83 ± 25.89, while in the case of DLM reconstructions, it was 35.61 ± 24.28 (p < 0.05). Subjective image quality assessment revealed no statistically significant differences between native and DLM reconstructions. The use of deep learning-based image reconstruction algorithms for CBCT imaging of the oral cavity can improve image quality by enhancing the CNR and lowering the noise.

PET image reconstruction using weighted nuclear norm maximization and deep learning prior

The ill-posed Positron emission tomography (PET) reconstruction problem usually results in limited resolution and significant noise. Recently, deep neural networks have been incorporated into PET iterative reconstruction framework to improve the image quality. In this paper, we propose a new neural network-based iterative reconstruction method by using weighted nuclear norm (WNN) maximization, which aims to recover the image details in the reconstruction process. The novelty of our method is the application of WNN maximization rather than WNN minimization in PET image reconstruction. Meanwhile, a neural network is used to control the noise originated from WNN maximization. Our method is evaluated on simulated and clinical datasets. The simulation results show that the proposed approach outperforms state-of-the-art neural network-based iterative methods by achieving the best contrast/noise tradeoff with a remarkable contrast improvement on the lesion contrast recovery. The study on clinical datasets also demonstrates that our method can recover lesions of different sizes while suppressing noise in various low-dose PET image reconstruction tasks. Our code is available athttps://github.com/Kuangxd/PETReconstruction.

Endosomal membrane budding patterns in plants

Multivesicular endosomes (MVEs) sequester membrane proteins destined for degradation within intralumenal vesicles (ILVs), a process mediated by the membrane-remodeling action of Endosomal Sorting Complex Required for Transport (ESCRT) proteins. In Arabidopsis, endosomal membrane constriction and scission are uncoupled, resulting in the formation of extensive concatenated ILV networks and enhancing cargo sequestration efficiency. Here, we used a combination of electron tomography, computer simulations, and mathematical modeling to address the questions of when concatenated ILV networks evolved in plants and what drives their formation. Through morphometric analyses of tomographic reconstructions of endosomes across yeast, algae, and various land plants, we have found that ILV concatenation is widespread within plant species, but only prevalent in seed plants, especially in flowering plants. Multiple budding sites that require the formation of pores in the limiting membrane were only identified in hornworts and seed plants, suggesting that this mechanism has evolved independently in both plant lineages. To identify the conditions under which these multiple budding sites can arise, we used particle-based molecular dynamics simulations and found that changes in ESCRT filament properties, such as filament curvature and membrane binding energy, can generate the membrane shapes observed in multiple budding sites. To understand the relationship between membrane budding activity and ILV network topology, we performed computational simulations and identified a set of membrane remodeling parameters that can recapitulate our tomographic datasets.

Integrating 3D technology with the Sampaio classification for enhanced percutaneous nephrolithotomy in complex renal calculi treatment.

To investigate the safety and efficacy of percutaneous nephrolithotomy (PCNL) in the treatment of complicated renal calculi by integrating three-dimensional (3D) computed tomography (CT) reconstruction with the Sampaio classification of the renal collecting system. Sixty-four consecutive patients with complex kidney calculi who underwent PCNL between January 2019 and October 2023 were retrospectively analyzed and divided into experimental group (3D printing) and control group (CT imaging) according to their willingness to pay for 3D imaging. Both groups underwent preoperative CT urography. The Digital Imaging and Communications (DICOM) in Medicine data of the experimental group from CT imaging were used for 3D reconstruction and model printing. Then, the Sampaio classification system was used to design the puncture channel and develop a surgical strategy. The 3D-printed models of the experimental group successfully displayed the Sampaio classification system. There was no significant difference in the baseline parameters between the groups. Compared with the control group, the experimental group exhibited significant improvements in the puncture time, number of puncture needles, number of puncture channels, target calyx consistency, number of first puncture channels, and stone clearance. There were no significant differences in the total operative time, decrease in the hemoglobin level, length of hospital stay, and postoperative complications between the groups. Integration of 3D technology with the Sampaio classification of the renal collecting system can enhance the preoperative evaluation and planning of percutaneous renal access. This approach allows a more precise method of PCNL for treating complex renal calculi.

Approach to frontal sinus via five frontal sinus drainage pathways.

This study aimed to validate a method for successful frontal sinus surgery. The method classifies the frontal sinus drainage pathway (FSDP) into five categories based on three bony walls of the anterior ethmoid sinus, including the uncinate process (UP), accessory uncinate process (UPa), and basal lamella of the ethmoid bulla (BLEB), which was tested in actual surgical procedures. This study analyzed 53 sides of 48 patients who underwent frontal sinus surgery between October 2022 and March 2023. We classified the FSDPs using preoperative computed tomography (CT) and multiplanar reconstruction (MPR). During surgeries for FSDPs located anterior to the BLEB, we used a two-step method involving resection of the turbinal UP, followed by upward resection from the lower edge of the ethmoidal UP. For FSDPs located posterior to the BLEB, we resected the BLEB at the superior semilunar recess. We confirmed the origin of each of the five types of FSDP during surgery. These origins, which were located at the lowest part of the anterior ethmoid, could be identified in the early stages of ethmoid sinus surgery before proceeding to the frontal recess area. The origins of the five types of FSDP, classified based on the bony walls, can be clearly and accurately identified during surgery. This provides a reliable method for preoperatively predicting and locating the inferior end of the FSDP (origin), without extensive manipulation of the cells formed in the frontal fossa.

A dual-domain cone beam computed tomography sparse-view reconstruction method based on generative projection interpolation

To propose a dual-domain CBCT reconstruction framework (DualSFR-Net) based on generative projection interpolation to reduce artifacts in sparse-view cone beam computed tomography (CBCT) reconstruction. The proposed method DualSFR-Net consists of a generative projection interpolation module, a domain transformation module, and an image restoration module. The generative projection interpolation module includes a sparse projection interpolation network (SPINet) based on a generative adversarial network and a full-view projection restoration network (FPRNet). SPINet performs projection interpolation to synthesize full-view projection data from the sparse-view projection data, while FPRNet further restores the synthesized full-view projection data. The domain transformation module introduces the FDK reconstruction and forward projection operators to complete the forward and gradient backpropagation processes. The image restoration module includes an image restoration network FIRNet that fine-tunes the domain-transformed images to eliminate residual artifacts and noise. Validation experiments conducted on a dental CT dataset demonstrated that DualSFR-Net was capable to reconstruct high-quality CBCT images under sparse-view sampling protocols. Quantitatively, compared to the current best methods, the DualSFR-Net method improved the PSNR by 0.6615 and 0.7658 and increased the SSIM by 0.0053 and 0.0134 under 2-fold and 4-fold sparse protocols, respectively. The proposed generative projection interpolation-based dual-domain CBCT sparse-view reconstruction method can effectively reduce stripe artifacts to improve image quality and enables efficient joint training for dual-domain imaging networks in sparse-view CBCT reconstruction.

Structural properties and mechanical behavior of three-dimensional cylindrical particle-like systems under in situ loading

To investigate the mechanical behavior of granular systems and their impact on density uniformity during compression, a three-dimensional micro-CT in situ loading experiment was conducted on a cylindrical granular system. The granular system was subjected to tomographic scanning and reconstruction, and the variation patterns of structural parameters such as volume fraction, porosity, and intrinsic density during the in situ loading process were analyzed. Internal pores within the granular system were extracted to obtain changes in pore area at each layer, allowing for an analysis of the evolving patterns of layered pore areas across different regions of the granular system during loading. This characterization elucidated the dynamic evolution process of compaction within the granular system. A contact network model for the granular system was established using particle contact area as a representation of inter-particle contact forces, revealing inherent connections between particle contact networks and macroscopic mechanical behaviors within the granular system.

Fast reconstruction 3D computed tomography image of stacked cell under faster scanning by dual-branch cross-fusion flat bottom network

Abstract Cone beam computed tomography (CBCT) fast scanning and reconstruction is a key step to achieve rapid detection of internal defects in batteries. In this work, we have achieved a faster CT scanning just in 5 seconds by reducing the X-ray exposure time in sparse view CT. However, the CT data is extremely incomplete by faster scanning; the existing reconstruction methods are difficult to reconstruct a high quality three-dimensional (3D) CT image of stacked cells. To address this issue, we propose a 3D CT image reconstruction network, which can reconstruct higher quality CT images from low quality 3D volume data. The input data of the reconstruction network is not 2D projection data, but 3D volume data. In this network, a high and low resolution dual-branch cross-fusion flat bottom structure is designed. The high resolution flat bottom branch aims to preserve detailed information, while the low resolution flat bottom branch focuses on capturing more semantic information. Cross-fusion between these branches mitigates the loss of semantic details. Additionally, the auxiliary loss function, the main loss function, and the 3D attention module are designed to enhance semantic accuracy and the learning performance of the network. The 3D training data is collected under a fast scanning strategy spanning 5-60 seconds. During the training phase, we use clipping block technology to cut the 3D volume data, enabling direct training on the 3D volume data. Our experimental results demonstrate that our 3D reconstruction network outperforms mainstream algorithms under this faster scanning strategy, which is able to reconstruct higher quality 3D CT images just in 15 seconds. Ablation experiments confirm the positive impact of the dual-branch cross-fusion flat bottom structure, attention module, and loss functions on improving the quality of 3D CT images.

Evaluation of colorectal liver metastases using virtual monoenergetic images obtained from dual-layer spectral computed tomography.

To assess the potential of virtual monoenergetic images in assessing colorectal liver metastasis (CRLM) compared with conventional CT images. This single-center, retrospective study included 173 consecutive patients (mean age, 65.5 ± 10.6years; 106 men) who underwent dual-layer spectral CT (DLSCT) between November 2016 and April 2021. Portal venous phase images were reconstructed using hybrid iterative reconstruction (iDose) and virtual monoenergetic imaging at 50keV. Four radiologists independently and randomly reviewed the de-identified iDose and 50keV images. Lesion detection, CRLM conspicuity, and CRLM diagnosis were compared between these images using a generalized estimating equation analysis. The reference standards used were histopathology and follow-up imaging findings. The study included 797 focal liver lesions, including 463 CRLMs (median size, 18.1mm [interquartile range, 10.9-37.7mm]). Lesion detection was better with 50keV images than with iDose images (45.0% [95% confidence interval [CI]: 39-50] vs 40.0% [95% CI: 34-46], P = 0.003). CRLM conspicuity was higher in the 50keV images than in the iDose images (3.27 [95% CI: 3.09-3.46] vs 3.09 [95% CI: 2.90-3.28], P < 0.001). However, the specificity for diagnosing CRLM was lower with 50keV images than with iDose images (94.5% [95% CI: 91.6-96.4] vs 96.0% [95% CI: 93.2-98.1], P = 0.022), whereas sensitivity did not differ significantly (77.6% [95% CI: 70.3-83.5] vs 76.9% [95% CI: 70.0-82.7], P = 0.736). Indeterminate lesions were more frequently noted in 50keV images than in iDose images (13% [445/3188] vs 9% [313/3188], P = 0.005), and 56% (247/445) of the indeterminate lesions at 50keV were not CRLMs. The 50keV images obtained from DLSCT were better than the iDose images in terms of CRLM conspicuity and lesion detection. However, 50keV images did not improve CRLM diagnosis but slightly increased the reporting of indeterminate focal liver lesions associated with CRLMs.