ABSTRACT
BACKGROUND: When applied to thermoacoustic imaging (TAI), the delay-and-sum (DAS) algorithm produces strong sidelobes due to its disadvantages of uniform aperture weighting. As a result, the quality of TAI images recovered by DAS is often severely degraded by strong non-coherent clutter, which restricts the development and application of TAI. PURPOSE: To address this issue, we propose an adaptive complementary neighboring sub-aperture (NSA) beamforming algorithm for TAI. METHODS: In NSA, we introduce a coordinate system transformation when calculating the normalized cross-correlation (NCC) matrix. This approach enables the computation of the NCC coefficient within the specified kernel without complex coordinate calculations. We first conducted the numerical simulation experiment to validate NSA using a tree branch phantom. In addition, we also conducted phantom (five sauce tubes), ex vivo (ablation needle in ex vivo porcine liver), and in vivo (human arm) TAI experiments using our TAI system with a center frequency of 3 GHz. RESULTS: In the numerical simulation experiment, the structural similarity index (SSIM) value for NSA is increased from 0.37828 for DAS to 0.75492. In the point target phantom TAI experiment, the generalized contrast-to-noise ratio (gCNR) value for NSA is increased from 0.936 for DAS to 0.962. The experimental results show that NSA can recover clearer thermoacoustic images compared to DAS. In the ex vivo TAI experiment, the full width at half maxima (FWHM) of an ablation needle (diameter = 1.5 mm) for coherence factor (CF) weighted DAS and NSA are 0.9 and 1.3 mm, respectively. Furthermore, in the in vivo TAI experiment, CF reduces the signals within the arm compared to NSA. Therefore, compared with CF, NSA can maintain the integrity of target information in TAI while effectively suppressing non-coherent background clutter. CONCLUSIONS: NSA can effectively reduce non-coherent background noise while ensuring the completeness of the target information. So, NSA offers the potential to provide high-quality thermoacoustic images and further advance their clinical application.
ABSTRACT
Compressed Sensing (CS) is a groundbreaking paradigm in image acquisition, challenging the constraints of the Nyquist-Shannon sampling theorem. This enables high-quality image reconstruction using a minimal number of measurements. Neural Networks' potent feature induction capabilities enable advanced data-driven CS methods to achieve high-fidelity image reconstruction. However, achieving satisfactory reconstruction performance, particularly in terms of perceptual quality, remains challenging at extremely low sampling rates. To tackle this challenge, we introduce a novel two-stage image CS framework based on latent diffusion, named LD-CSNet. In the first stage, we utilize an autoencoder pre-trained on a large dataset to represent natural images as low-dimensional latent vectors, establishing prior knowledge distinct from sparsity and effectively reducing the dimensionality of the solution space. In the second stage, we employ a conditional diffusion model for maximum likelihood estimates in the latent space. This is supported by a measurement embedding module designed to encode measurements, making them suitable for a denoising network. This guides the generation process in reconstructing low-dimensional latent vectors. Finally, the image is reconstructed using a pre-trained decoder. Experimental results across multiple public datasets demonstrate LD-CSNet's superior perceptual quality and robustness to noise. It maintains fidelity and visual quality at lower sampling rates. Research findings suggest the promising application of diffusion models in image CS. Future research can focus on developing more appropriate models for the first stage.
ABSTRACT
Game theory-inspired deep learning using a generative adversarial network provides an environment to competitively interact and accomplish a goal. In the context of medical imaging, most work has focused on achieving single tasks such as improving image resolution, segmenting images, and correcting motion artifacts. We developed a dual-objective adversarial learning framework that simultaneously 1) reconstructs higher quality brain magnetic resonance images (MRIs) that 2) retain disease-specific imaging features critical for predicting progression from mild cognitive impairment (MCI) to Alzheimer's disease (AD). We obtained 3-Tesla, T1-weighted brain MRIs of participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI, N=342) and the National Alzheimer's Coordinating Center (NACC, N = 190) datasets. We simulated MRIs with missing data by removing 50% of sagittal slices from the original scans (i.e., diced scans). The generator was trained to reconstruct brain MRIs using the diced scans as input. We introduced a classifier into the GAN architecture to discriminate between stable (i.e., sMCI) and progressive MCI (i.e., pMCI) based on the generated images to facilitate encoding of disease-related information during reconstruction. The framework was trained using ADNI data and externally validated on NACC data. In the NACC cohort, generated images had better image quality than the diced scans (Structural similarity (SSIM) index: 0.553 ± 0.116 versus 0.348 ± 0.108). Furthermore, a classifier utilizing the generated images distinguished pMCI from sMCI more accurately than with the diced scans (F1-score: 0.634 ± 0.019 versus 0.573 ± 0.028). Competitive deep learning has potential to facilitate disease-oriented image reconstruction in those at risk of developing Alzheimer's disease.
ABSTRACT
PURPOSE: To evaluate the feasibility and utility of a deep learning (DL)-based reconstruction for improving the SNR of hyperpolarized 129Xe lung ventilation MRI. METHODS: 129Xe lung ventilation MRI data acquired from patients with asthma and/or chronic obstructive pulmonary disease (COPD) were retrospectively reconstructed with a commercial DL reconstruction pipeline at five different denoising levels. Quantitative imaging metrics of lung ventilation including ventilation defect percentage (VDP) and ventilation heterogeneity index (VHI) were compared between each set of DL-reconstructed images and alternative denoising strategies including: filtering, total variation denoising and higher-order singular value decomposition. Structural similarity between the denoised and original images was assessed. In a prospective study, the feasibility of using SNR gains from DL reconstruction to allow natural-abundance xenon MRI was evaluated in healthy volunteers. RESULTS: 129Xe ventilation image SNR was improved with DL reconstruction when compared with conventionally reconstructed images. In patients with asthma and/or COPD, DL-reconstructed images exhibited a slight positive bias in ventilation defect percentage (1.3% at 75% denoising) and ventilation heterogeneity index (Ë1.4) when compared with conventionally reconstructed images. Additionally, DL-reconstructed images preserved structural similarity more effectively than data denoised using alternative approaches. DL reconstruction greatly improved image SNR (greater than threefold), to a level that 129Xe ventilation imaging using natural-abundance xenon appears feasible. CONCLUSION: DL-based image reconstruction significantly improves 129Xe ventilation image SNR, preserves structural similarity, and leads to a minor bias in ventilation metrics that can be attributed to differences in the image sharpness. This tool should help facilitate cost-effective 129Xe ventilation imaging with natural-abundance xenon in the future.
ABSTRACT
BACKGROUND: Photon-counting computed tomography (PCCT) enables new ways of image reconstruction, e.g. material decomposition and creation of virtual non-contrast (VNC) series with higher resolution and lower radiation dose than standard computed tomography (CT). Clinical experiences of this are limited. PURPOSE: To compare true non-contrast (TNC) series with VNC series derived from non-enhanced (VNCu), arterial phase (VNCa) and portal venous phase (VNCv) in clinically approved PCCT. MATERIAL AND METHODS: A total of 45 clinical, tri-phasic abdominal CT scans from the PCCT Naetom Alpha, between February 2022 and November 2022, were retrospectively assessed. Placing a region of interest in six different locations in each VNC series - right liver parenchyma, left liver parenchyma, spleen, aorta, erector spinae muscle, and in the subcutaneous fat - absolute Hounsfield values (HU) and standard deviations (SD) were collected. Differences in HU (ΔHU) were compared and statistically analyzed. RESULTS: Statistically significant differences between VNC and TNC were seen in all measurements, with the largest difference in the subcutaneous fat and the smallest difference in the erector spinae muscle. Only small differences were seen between VNCa and VNCv, where the largest differences were seen in the left and right liver lobes. CONCLUSION: VNC images from the first-generation clinically approved PCCT showed a significant difference between VNC and TNC images. The differences vary with the type of tissue. Only small differences were seen depending from which contrast phase the VNC was derived.
ABSTRACT
Compressed ultrafast photography (CUP) can capture irreversible or difficult-to-repeat dynamic scenes at the imaging speed of more than one billion frames per second, which is obtained by compressive sensing-based image reconstruction from a compressed 2D image through the discretization of detector pixels. However, an excessively high data compression ratio in CUP severely degrades the image reconstruction quality, thereby restricting its ability to observe ultrafast dynamic scenes with complex spatial structures. To address this issue, a discrete illumination-based CUP (DI-CUP) with high fidelity is reported. In DI-CUP, the dynamic scenes are loaded into an ultrashort laser pulse train with controllable sub-pulse number and time interval, thus the data compression ratio, as well as the overlap between adjacent frames, is greatly decreased and flexibly controlled through the discretization of dynamic scenes based on laser pulse train illumination, and high-fidelity image reconstruction can be realized within the same observation time window. Furthermore, the superior performance of DI-CUP is verified by observing femtosecond laser-induced ablation dynamics and plasma channel evolution, which are hardly resolved in the spatial structures using conventional CUP. It is anticipated that DI-CUP will be widely and dependably used in the real-time observations of various ultrafast dynamics.
ABSTRACT
Micro-ultrasound has recently been introduced as a low-cost alternative to multi-parametric MRI for imaging prostate cancer. Early clinical studies have demonstrated promising results; however, robust validation via comparison with whole-mount pathology has yet to be achieved. Due to micro-ultrasound probe design and tissue deformation during scanning, it is difficult to accurately correlate micro-ultrasound imaging planes with ground truth whole-mount pathology slides. In this study, we developed a multi-step methodology to co-register micro-ultrasound and MRI to whole-mount pathology. The three-step process had a registration error of 3.90 ± 0.11 mm and consists of: (1) micro-ultrasound image reconstruction, (2) 3D landmark registration of micro-ultrasound to MRI, and (3) 2D capsule registration of MRI to whole-mount pathology. This process was then used in a preliminary reader study to compare the diagnostic accuracy of micro-ultrasound and MRI in 15 patients who underwent radical prostatectomy for prostate cancer. Micro-ultrasound was found to have equivalent performance to retrospective MRI review for index lesion detection (91.7% vs. 80%), while demonstrating an increased detection of tumor extent (52.5% vs. 36.7%) with similar false positive regions-of-interest (38.3% vs. 40.8%). Prospective MRI review had reduced detection of index lesions (73.3%) and tumor extent (18.9%) but improved false positive regions-of-interest (22.7%) relative to micro-ultrasound and retrospective MRI. Further evaluation is needed with a larger sample size.
Subject(s)
Magnetic Resonance Imaging , Prostatic Neoplasms , Ultrasonography , Humans , Prostatic Neoplasms/diagnostic imaging , Prostatic Neoplasms/pathology , Male , Magnetic Resonance Imaging/methods , Ultrasonography/methods , Aged , Retrospective Studies , Middle Aged , Prostate/diagnostic imaging , Prostate/pathology , Prostatectomy , Image Processing, Computer-Assisted/methodsABSTRACT
BACKGROUND: Subtraction of single-photon emission computed tomography (SPECT) images has a number of clinical applications in e.g. foci localization in ictal/inter-ictal SPECT and defect detection in rest/stress cardiac SPECT. In this work, we investigated the technical performance of SPECT subtraction for the purpose of quantifying the effect of a vasoconstricting drug (angiotensin-II, or AT2) on the Tc-99m-MAA liver distribution in hepatic radioembolization using an innovative interventional hybrid C-arm scanner. Given that subtraction of SPECT images is challenging due to high noise levels and poor resolution, we compared four methods to obtain a difference image in terms of image quality and quantitative accuracy. These methods included (i) image subtraction: subtraction of independently reconstructed SPECT images, (ii) projection subtraction: reconstruction of a SPECT image from subtracted projections, (iii) projection addition: reconstruction by addition of projections as a background term during the iterative reconstruction, and (iv) image addition: simultaneous reconstruction of the difference image and the subtracted image. RESULTS: Digital simulations (XCAT) and phantom studies (NEMA-IQ and anthropomorphic torso) showed that all four methods were able to generate difference images but their performance on specific metrics varied substantially. Image subtraction had the best quantitative performance (activity recovery coefficient) but had the worst visual quality (contrast-to-noise ratio) due to high noise levels. Projection subtraction showed a slightly better visual quality than image subtraction, but also a slightly worse quantitative accuracy. Projection addition had a substantial bias in its quantitative accuracy which increased with less counts in the projections. Image addition resulted in the best visual image quality but had a quantitative bias when the two images to subtract contained opposing features. CONCLUSION: All four investigated methods of SPECT subtraction demonstrated the capacity to generate a feasible difference image from two SPECT images. Image subtraction is recommended when the user is only interested in quantitative values, whereas image addition is recommended when the user requires the best visual image quality. Since quantitative accuracy is most important for the dosimetric investigation of AT2 in radioembolization, we recommend using the image subtraction method for this purpose.
ABSTRACT
OBJECTIVE: Multi-energy CT conducted by photon-counting detector has a wide range of applications, especially in multiple contrast agents imaging. However, static multi-energy (SME) CT imaging suffers from higher statistical noise because of increased energy bins with static energy thresholds. Our team has proposed a dynamic dual-energy (DDE) CT detector model and the corresponding iterative reconstruction algorithm to solve this problem. However, rigorous and detailed analysis of the statistical noise characterization in this DDE CT was lacked. APPROACH: Starting from the properties of the Poisson random variable, this paper analyzes the noise characterization of the DDE CT and compares it with the SME CT. It is proved that the multi-energy CT projections and reconstruction images calculated from the proposed DDE CT algorithm have less statistical noise than that of the SME CT. MAIN RESULTS: Simulations and experiments verify that the expectations of the multi-energy CT projections calculated from DDE CT are the same as those of the SME projections. Still, the variance of the former is smaller. We further analyze the convergence of the iterative DDE CT algorithm through simulations and prove that the derived noise characterization can be realized under different CT imaging configurations. SIGNIFICANCE: The low statistical noise characteristics demonstrate the value of DDE CT imaging technology.
ABSTRACT
Deep learning techniques offer improvements in computer-aided diagnosis systems. However, acquiring image domain annotations is challenging due to the knowledge and commitment required of expert pathologists. Pathologists often identify regions in whole slide images with diagnostic relevance rather than examining the entire slide, with a positive correlation between the time spent on these critical image regions and diagnostic accuracy. In this paper, a heatmap is generated to represent pathologists' viewing patterns during diagnosis and used to guide a deep learning architecture during training. The proposed system outperforms traditional approaches based on color and texture image characteristics, integrating pathologists' domain expertise to enhance region of interest detection without needing individual case annotations. Evaluating our best model, a U-Net model with a pre-trained ResNet-18 encoder, on a skin biopsy whole slide image dataset for melanoma diagnosis, shows its potential in detecting regions of interest, surpassing conventional methods with an increase of 20%, 11%, 22%, and 12% in precision, recall, F1-score, and Intersection over Union, respectively. In a clinical evaluation, three dermatopathologists agreed on the model's effectiveness in replicating pathologists' diagnostic viewing behavior and accurately identifying critical regions. Finally, our study demonstrates that incorporating heatmaps as supplementary signals can enhance the performance of computer-aided diagnosis systems. Without the availability of eye tracking data, identifying precise focus areas is challenging, but our approach shows promise in assisting pathologists in improving diagnostic accuracy and efficiency, streamlining annotation processes, and aiding the training of new pathologists.
ABSTRACT
Objective: This study aims to investigate the image quality of a high-resolution, low-dose coronary CT angiography (CCTA) with deep learning image reconstruction (DLIR) and second-generation motion correction algorithms, namely, SnapShot Freeze 2 (SSF2) algorithm, and its diagnostic accuracy for in-stent restenosis (ISR) in patients after percutaneous coronary intervention (PCI), in comparison with standard-dose CCTA with high-definition mode reconstructed by adaptive statistical iterative reconstruction Veo algorithm (ASIR-V) and the first-generation motion correction algorithm, namely, SnapShot Freeze 1 (SSF1). Methods: Patients after PCI and suspected of having ISR scheduled for high-resolution CCTA (randomly for 100â kVp low-dose CCTA or 120â kVp standard-dose) and invasive coronary angiography (ICA) were prospectively enrolled in this study. After the basic information pairing, a total of 105 patients were divided into the LD group (60 patients underwent 100â kVp low-dose CCTA reconstructed with DLIR and SSF2) and the SD group (45 patients underwent 120â kVp standard-dose CCTA reconstructed with ASIR-V and SSF1). Radiation and contrast medium doses, objective image quality including CT value, image noise (standard deviation), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) for the aorta, left main artery (LMA), left ascending artery (LAD), left circumflex artery (LCX), and right coronary artery (RCA) of the two groups were compared. A five-point scoring system was used for the overall image quality and stent appearance evaluation. Binary ISR was defined as an in-stent neointimal proliferation with diameter stenosis ≥50% to assess the diagnostic performance between the LD group and SD group with ICA as the standard reference. Results: The LD group achieved better objective and subjective image quality than that of the SD group even with 39.1% radiation dose reduction and 28.0% contrast media reduction. The LD group improved the diagnostic accuracy for coronary ISR to 94.2% from the 83.8% of the SD group on the stent level and decreased the ratio of false-positive cases by 19.2%. Conclusion: Compared with standard-dose CCTA with ASIR-V and SSF1, the high-resolution, low-dose CCTA with DLIR and SSF2 reconstruction algorithms further improves the image quality and diagnostic performance for coronary ISR at 39.1% radiation dose reduction and 28.0% contrast dose reduction.
ABSTRACT
To evaluate the usefulness of low-keV multiphasic computed tomography (CT) with deep learning image reconstruction (DLIR) in improving the delineation of pancreatic ductal adenocarcinoma (PDAC) compared to conventional hybrid iterative reconstruction (HIR). Thirty-five patients with PDAC who underwent multiphasic CT were retrospectively evaluated. Raw data were reconstructed with two energy levels (40 keV and 70 keV) of virtual monochromatic imaging (VMI) using HIR (ASiR-V50%) and DLIR (TrueFidelity-H). Contrast-to-noise ratio (CNRtumor) was calculated from the CT values within regions of interest in tumor and normal pancreas in the pancreatic parenchymal phase images. Lesion conspicuity of PDAC in pancreatic parenchymal phase on 40-keV HIR, 40-keV DLIR, and 70-keV DLIR images was qualitatively rated on a 5-point scale, using 70-keV HIR images as reference (score 1 = poor; score 3 = equivalent to reference; score 5 = excellent) by two radiologists. CNRtumor of 40-keV DLIR images (median 10.4, interquartile range (IQR) 7.8-14.9) was significantly higher than that of the other VMIs (40 keV HIR, median 6.2, IQR 4.4-8.5, P < 0.0001; 70-keV DLIR, median 6.3, IQR 5.1-9.9, P = 0.0002; 70-keV HIR, median 4.2, IQR 3.1-6.1, P < 0.0001). CNRtumor of 40-keV DLIR images were significantly better than those of the 40-keV HIR and 70-keV HIR images by 72 ± 22% and 211 ± 340%, respectively. Lesion conspicuity scores on 40-keV DLIR images (observer 1, 4.5 ± 0.7; observer 2, 3.4 ± 0.5) were significantly higher than on 40-keV HIR (observer 1, 3.3 ± 0.9, P < 0.0001; observer 2, 3.1 ± 0.4, P = 0.013). DLIR is a promising reconstruction method to improve PDAC delineation in 40-keV VMI at the pancreatic parenchymal phase compared to conventional HIR.
ABSTRACT
Background: Deep learning image reconstruction (DLIR) is a novel computed tomography (CT) reconstruction technique that minimizes image noise, enhances image quality, and enables radiation dose reduction. This study aims to compare the diagnostic performance of DLIR and iterative reconstruction (IR) in the evaluation of focal hepatic lesions. Methods: We conducted a retrospective study of 216 focal hepatic lesions in 109 adult participants who underwent abdominal CT scanning at our institution. We used DLIR (low, medium, and high strength) and IR (0 %, 10 %, 20 %, and 30 %) techniques for image reconstruction. Four experienced abdominal radiologists independently evaluated focal hepatic lesions based on five qualitative aspects (lesion detectability, lesion border, diagnostic confidence level, image artifact, and overall image quality). Quantitatively, we measured and compared the level of image noise for each technique at the liver and aorta. Results: There were significant differences (p < 0.001) among the seven reconstruction techniques in terms of lesion borders, image artifacts, and overall image quality. Low-strength DLIR (DLIR-L) exhibited the best overall image quality. Although high-strength DLIR (DLIR-H) had the least image noise and fewest artifacts, it also had the lowest scores for lesion borders and overall image quality. Image noise showed a weak to moderate positive correlation with participants' body mass index and waist circumference. Conclusions: The optimal-strength DLIR significantly improved overall image quality for evaluating focal hepatic lesions compared to the IR technique. DLIR-L achieved the best overall image quality while maintaining acceptable levels of image noise and quality of lesion borders.
ABSTRACT
Background: High-quality images with minimum radiation dose are considered a challenge in Computed Tomography (CT) scans. Objective: The current study aimed to assess the efficacy of the Iterative Reconstruction in Image Space (IRIS) algorithm combined with Automatic Tube Current Modulation (ATCM) compared to Filtered Back Projection (FBP) in brain CT scans. Material and Methods: In this cross-sectional study, 200 patients underwent to brain CT scan, and images were then reconstructed using both FBP and IRIS. The CT Number (CTN), noise, and Signal-to-Noise Ratio (SNR) were computed for different tissues from CT images. The performance of two algorithms under different exposure conditions was evaluated using a water phantom. Two experienced radiologists assessed the image quality. Volume CT Dose Index (CTDIvol) and Dose Length Product (DLP) were recorded for each scan. Results: FBP reconstruction exhibited higher noise and lower SNR compared to IRIS, both with and without ATCM. Noise levels significantly increased for FBP combined with ATCM. Subjective analysis showed higher performance for IRIS without ATCM compared to other approaches. The mean CTDIvol with and without ATCM was 20.04±3.33 and 36.37±4.65 mGy, respectively. In the phantom study, the noise with IRIS remained lower than that with FBP even with a 42% dose reduction. Conclusion: IRIS algorithm can preserve the image quality when radiation dose is significantly reduced by ATCM in brain CT scan. Implementation of IRIS combined with ATCM is recommended for brain CT examinations.
ABSTRACT
Deep learning, particularly convolutional neural networks (CNNs), has advanced positron emission tomography (PET) image reconstruction. However, it requires extensive, high-quality training datasets. Unsupervised learning methods, such as deep image prior (DIP), have shown promise for PET image reconstruction. Although DIP-based PET image reconstruction methods demonstrate superior performance, they involve highly time-consuming calculations. This study proposed a two-step optimization method to accelerate end-to-end DIP-based PET image reconstruction and improve PET image quality. The proposed two-step method comprised a pre-training step using conditional DIP denoising, followed by an end-to-end reconstruction step with fine-tuning. Evaluations using Monte Carlo simulation data demonstrated that the proposed two-step method significantly reduced the computation time and improved the image quality, thereby rendering it a practical and efficient approach for end-to-end DIP-based PET image reconstruction.
Subject(s)
Deep Learning , Image Processing, Computer-Assisted , Monte Carlo Method , Positron-Emission Tomography , Image Processing, Computer-Assisted/methods , Positron-Emission Tomography/methods , Humans , Phantoms, ImagingABSTRACT
OBJECTIVE: Penalty parameters in penalized likelihood positron emission tomography (PET) reconstruction are typically determined empirically. The cross-validation log-likelihood (CVLL) method has been introduced to optimize these parameters by maximizing a CVLL function, which assesses the likelihood of reconstructed images using one subset of a list-mode dataset based on another subset. This study aims to validate the efficacy of the CVLL method in whole-body imaging for cancer patients using a conventional clinical PET scanner. APPROACH: Fifteen lung cancer patients were injected with 243.7±23.8 MBq of [18F]FDG and underwent a 22-minute PET scan on a Biograph mCT PET/CT scanner, starting at 60±5 minutes post-injection. The PET list-mode data were partitioned by subsampling without replacement, with 20 minutes used for image reconstruction using an in-house ordered subset expectation maximization algorithm and the remaining 2 minutes for cross-validation. Two penalty parameters, penalty strength ß and Fair penalty function parameter δ, were subjected to optimization. Whole-body images were reconstructed, and CVLL values were computed across various penalty parameter combinations. The optimal image corresponding to the maximum CVLL value was selected by a grid search for each patient. MAIN RESULTS: The δ value required to maximize the CVLL value was notably small (≤ 10-6 in this study). The influences of voxel size and scan duration on image optimization were investigated. A correlation analysis revealed a significant inverse relationship between optimal ß and scan count level, with a correlation coefficient of -0.68 (p-value = 3.5×10-5). The optimal images selected by the CVLL method were compared with those chosen by two radiologists based on their diagnostic preferences. Differences were observed in the selection of optimal images. SIGNIFICANCE: This study demonstrates the feasibility of incorporating the CVLL method into routine imaging protocols, potentially allowing for a wide range of combinations of injected radioactivity amounts and scan durations in modern PET imaging.
ABSTRACT
PURPOSE: To develop a neural network architecture for improved calibrationless reconstruction of radial data when no ground truth is available for training. METHODS: NLINV-Net is a model-based neural network architecture that directly estimates images and coil sensitivities from (radial) k-space data via nonlinear inversion (NLINV). Combined with a training strategy using self-supervision via data undersampling (SSDU), it can be used for imaging problems where no ground truth reconstructions are available. We validated the method for (1) real-time cardiac imaging and (2) single-shot subspace-based quantitative T1 mapping. Furthermore, region-optimized virtual (ROVir) coils were used to suppress artifacts stemming from outside the field of view and to focus the k-space-based SSDU loss on the region of interest. NLINV-Net-based reconstructions were compared with conventional NLINV and PI-CS (parallel imaging + compressed sensing) reconstruction and the effect of the region-optimized virtual coils and the type of training loss was evaluated qualitatively. RESULTS: NLINV-Net-based reconstructions contain significantly less noise than the NLINV-based counterpart. ROVir coils effectively suppress streakings which are not suppressed by the neural networks while the ROVir-based focused loss leads to visually sharper time series for the movement of the myocardial wall in cardiac real-time imaging. For quantitative imaging, T1-maps reconstructed using NLINV-Net show similar quality as PI-CS reconstructions, but NLINV-Net does not require slice-specific tuning of the regularization parameter. CONCLUSION: NLINV-Net is a versatile tool for calibrationless imaging which can be used in challenging imaging scenarios where a ground truth is not available.
ABSTRACT
OBJECTIVES: Detection of liver metastases is crucial for guiding oncological management. Computed tomography through iterative reconstructions is widely used in this indication but has certain limitations. Deep learning image reconstructions (DLIR) use deep neural networks to achieve a significant noise reduction compared to iterative reconstructions. While reports have demonstrated improvements in image quality, their impact on liver metastases detection remains unclear. Our main objective was to determine whether DLIR affects the number of detected liver metastasis. Our secondary objective was to compare metastases conspicuity between the two reconstruction methods. METHODS: CT images of 121 patients with liver metastases were reconstructed using a 50% adaptive statistical iterative reconstruction (50%-ASiR-V), and three levels of DLIR (DLIR-low, DLIR-medium, and DLIR-high). For each reconstruction, two double-blinded radiologists counted up to a maximum of ten metastases. Visibility and contour definitions were also assessed. Comparisons between methods for continuous parameters were performed using mixed models. RESULTS: A higher number of metastases was detected by one reader with DLIR-high: 7 (2-10) (median (Q1-Q3); total 733) versus 5 (2-10), respectively for DLIR-medium, DLIR-low, and ASiR-V (p < 0.001). Ten patents were detected with more metastases with DLIR-high simultaneously by both readers and a third reader for confirmation. Metastases visibility and contour definition were better with DLIR than ASiR-V. CONCLUSION: DLIR-high enhanced the detection and visibility of liver metastases compared to ASiR-V, and also increased the number of liver metastases detected. CRITICAL RELEVANCE STATEMENT: Deep learning-based reconstruction at high strength allowed an increase in liver metastases detection compared to hybrid iterative reconstruction and can be used in clinical oncology imaging to help overcome the limitations of CT. KEY POINTS: Detection of liver metastases is crucial but limited with standard CT reconstructions. More liver metastases were detected with deep-learning CT reconstruction compared to iterative reconstruction. Deep learning reconstructions are suitable for hepatic metastases staging and follow-up.
ABSTRACT
Background: Simultaneous multi-slice (SMS) bSSFP imaging enables stress myocardial perfusion imaging with high spatial resolution and increased spatial coverage. Standard parallel imaging techniques (e.g., TGRAPPA) can be used for image reconstruction but result in high noise level. Alternatively, iterative reconstruction techniques based on temporal regularization (ITER) improve image quality but are associated with reduced temporal signal fidelity and long computation time limiting their online use. The aim is to develop an image reconstruction technique for SMS-bSSFP myocardial perfusion imaging combining parallel imaging and image-based denoising using a novel noise map estimation network (NoiseMapNet), which preserves both sharpness and temporal signal profiles and that has low computational cost. Methods: The proposed reconstruction of SMS images consists of a standard temporal parallel imaging reconstruction (TGRAPPA) with motion correction (MOCO) followed by image denoising using NoiseMapNet. NoiseMapNet is a deep learning network based on a 2D Unet architecture and aims to predict a noise map from an input noisy image, which is then subtracted from the noisy image to generate the denoised image. This approach was evaluated in 17 patients who underwent stress perfusion imaging using a SMS-bSSFP sequence. Images were reconstructed with (a) TGRAPPA with MOCO (thereafter referred to as TGRAPPA), (b) iterative reconstruction with integrated motion compensation (ITER), and (c) proposed NoiseMapNet-based reconstruction. Normalized mean squared error (NMSE) with respect to TGRAPPA, myocardial sharpness, image quality, perceived SNR (pSNR), and number of diagnostic segments were evaluated. Results: NMSE of NoiseMapNet was lower than using ITER for both myocardium (0.045 ± 0.021 vs. 0.172 ± 0.041, p < 0.001) and left ventricular blood pool (0.025 ± 0.014 vs. 0.069 ± 0.020, p < 0.001). There were no significant differences between all methods for myocardial sharpness (p = 0.77) and number of diagnostic segments (p = 0.36). ITER led to higher image quality than NoiseMapNet/TGRAPPA (2.7 ± 0.4 vs. 1.8 ± 0.4/1.3 ± 0.6, p < 0.001) and higher pSNR than NoiseMapNet/TGRAPPA (3.0 ± 0.0 vs. 2.0 ± 0.0/1.3 ± 0.6, p < 0.001). Importantly, NoiseMapNet yielded higher pSNR (p < 0.001) and image quality (p < 0.008) than TGRAPPA. Computation time of NoiseMapNet was only 20s for one entire dataset. Conclusion: NoiseMapNet-based reconstruction enables fast SMS image reconstruction for stress myocardial perfusion imaging while preserving sharpness and temporal signal profiles.
ABSTRACT
BACKGROUND: The best settings of deep learning image reconstruction (DLIR) algorithm for abdominal low-kiloelectron volt (keV) virtual monoenergetic imaging (VMI) have not been determined. PURPOSE: To determine the optimal settings of the DLIR algorithm for abdominal low-keV VMI. MATERIAL AND METHODS: The portal-venous phase computed tomography (CT) scans of 109 participants with 152 lesions were reconstructed into four image series: VMI at 50 keV using adaptive statistical iterative reconstruction (Asir-V) at 50% blending (AV-50); and VMI at 40 keV using AV-50 and DLIR at medium (DLIR-M) and high strength (DLIR-H). The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of nine anatomical sites were calculated. Noise power spectrum (NPS) using homogenous region of liver, and edge rise slope (ERS) at five edges were measured. Five radiologists rated image quality and diagnostic acceptability, and evaluated the lesion conspicuity. RESULTS: The SNR and CNR values, and noise and noise peak in NPS measurements, were significantly lower in DLIR images than AV-50 images in all anatomical sites (all P < 0.001). The ERS values were significantly higher in 40-keV images than 50-keV images at all edges (all P < 0.001). The differences of the peak and average spatial frequency among the four reconstruction algorithms were significant but relatively small. The 40-keV images were rated higher with DLIR-M than DLIR-H for diagnostic acceptance (P < 0.001) and lesion conspicuity (P = 0.010). CONCLUSION: DLIR provides lower noise, higher sharpness, and more natural texture to allow 40 keV to be a new standard for routine VMI reconstruction for the abdomen and DLIR-M gains higher diagnostic acceptance and lesion conspicuity rating than DLIR-H.