Fast and low-dose medical imaging generation empowered by hybrid deep-learning and iterative reconstruction.
Ontology highlight
ABSTRACT:
SUBMITTER: Liao S
PROVIDER: S-EPMC10394257 | biostudies-literature | 2023 Jul
REPOSITORIES: biostudies-literature
Publications
Fast and low-dose medical imaging generation empowered by hybrid deep-learning and iterative reconstruction.
Liao Shu S Mo Zhanhao Z Zeng Mengsu M Wu Jiaojiao J Gu Yuning Y Li Guobin G Quan Guotao G Lv Yang Y Liu Lin L Yang Chun C Wang Xinglie X Huang Xiaoqian X Zhang Yang Y Cao Wenjing W Dong Yun Y Wei Ying Y Zhou Qing Q Xiao Yongqin Y Zhan Yiqiang Y Zhou Xiang Sean XS Shi Feng F Shen Dinggang D
Cell reports. Medicine 20230701 7
Fast and low-dose reconstructions of medical images are highly desired in clinical routines. We propose a hybrid deep-learning and iterative reconstruction (hybrid DL-IR) framework and apply it for fast magnetic resonance imaging (MRI), fast positron emission tomography (PET), and low-dose computed tomography (CT) image generation tasks. First, in a retrospective MRI study (6,066 cases), we demonstrate its capability of handling 3- to 10-fold under-sampled MR data, enabling organ-level coverage ...[more]
Similar Datasets
Project description:A Fourier-based iterative reconstruction technique, termed Equally Sloped Tomography (EST), is developed in conjunction with advanced mathematical regularization to investigate radiation dose reduction in x-ray CT. The method is experimentally implemented on fan-beam CT and evaluated as a function of imaging dose on a series of image quality phantoms and anonymous pediatric patient data sets. Numerical simulation experiments are also performed to explore the extension of EST to helical cone-beam geometry.EST is a Fourier based iterative algorithm, which iterates back and forth between real and Fourier space utilizing the algebraically exact pseudopolar fast Fourier transform (PPFFT). In each iteration, physical constraints and mathematical regularization are applied in real space, while the measured data are enforced in Fourier space. The algorithm is automatically terminated when a proposed termination criterion is met. Experimentally, fan-beam projections were acquired by the Siemens z-flying focal spot technology, and subsequently interleaved and rebinned to a pseudopolar grid. Image quality phantoms were scanned at systematically varied mAs settings, reconstructed by EST and conventional reconstruction methods such as filtered back projection (FBP), and quantified using metrics including resolution, signal-to-noise ratios (SNRs), and contrast-to-noise ratios (CNRs). Pediatric data sets were reconstructed at their original acquisition settings and additionally simulated to lower dose settings for comparison and evaluation of the potential for radiation dose reduction. Numerical experiments were conducted to quantify EST and other iterative methods in terms of image quality and computation time. The extension of EST to helical cone-beam CT was implemented by using the advanced single-slice rebinning (ASSR) method.Based on the phantom and pediatric patient fan-beam CT data, it is demonstrated that EST reconstructions with the lowest scanner flux setting of 39 mAs produce comparable image quality, resolution, and contrast relative to FBP with the 140 mAs flux setting. Compared to the algebraic reconstruction technique and the expectation maximization statistical reconstruction algorithm, a significant reduction in computation time is achieved with EST. Finally, numerical experiments on helical cone-beam CT data suggest that the combination of EST and ASSR produces reconstructions with higher image quality and lower noise than the Feldkamp Davis and Kress (FDK) method and the conventional ASSR approach.A Fourier-based iterative method has been applied to the reconstruction of fan-bean CT data with reduced x-ray fluence. This method incorporates advantageous features in both real and Fourier space iterative schemes: using a fast and algebraically exact method to calculate forward projection, enforcing the measured data in Fourier space, and applying physical constraints and flexible regularization in real space. Our results suggest that EST can be utilized for radiation dose reduction in x-ray CT via the readily implementable technique of lowering mAs settings. Numerical experiments further indicate that EST requires less computation time than several other iterative algorithms and can, in principle, be extended to helical cone-beam geometry in combination with the ASSR method.
| S-EPMC3598761 | biostudies-other
Project description:BackgroundLow-kiloelectron volt (keV) virtual monochromatic images (VMIs) from low-dose (LD) dual-energy computed tomography (DECT) can enhance lesion contrast but suffer from high image noise. Recently, a deep learning image reconstruction (DLIR) algorithm has been developed and shown significant potential in suppressing image noise and improving image quality. To date, the capacity of LD low-keV thoracic-abdominal-pelvic DECT with DLIR to detect various types of tumor lesions have not been assessed. Hence, this study aimed to evaluate the image quality and lesion detection capabilities of LD VMIs using DLIR with thoracic-abdominal-pelvic DECT versus standard-dose (SD) iterative reconstruction (IR) in oncology patients.MethodsThis prospective intraindividual study included 56 oncology patients who received a SD (13.86 mGy) and a consecutive LD (7.15 mGy) thoracic-abdominal-pelvic DECT from April 2022 to July 2023 at The First Affiliated hospital of Zhengzhou University. SD VMIs were reconstructed using IR at 50 keV (SD-IR50 keV), while LD VMIs were processed using DLIR at 50 keV (LD-DL50 keV) and 40 keV (LD-DL40 keV), respectively. Quantitative image parameters [computed tomography (CT) values, image noise, and contrast-to-noise ratios (CNRs)], qualitative metrics (image noise, vessel conspicuity, image contrast, artificial sensation, and overall image quality), and lesion CNRs and conspicuity were compared. The lesion detection rates in the SD-IR50 keV, LD-DL50 keV, and LD-DL40 keV VMIs were assessed according to lesion location (lung, liver, and lymph), type, and size. Repeated measures analysis of variance and the Friedman test were applied for comparing quantitative and qualitative measures, respectively. The Cochran Q test was used for comparing lesion detection rates.ResultsCompared to SD-IR50 keV VMIs, LD-DL50 keV VMIs showed similar CT values and image noise (P>0.05), similar (P>0.05) or higher(P<0.05) CNRs, similar (P>0.05) or superior (P<0.05) perceptual image quality, and similar (P>0.05) or higher (P<0.001) lesion CNR and conspicuity. LD-DL40 keV VMIs exhibited higher CT values (by 40.4-47.1%) and CNRs (by 21.8-39.8%) (P<0.001), equivalent image noise, similar (P>0.05) or superior (P<0.05) perceptual image quality except for artificial sensation, and similar (P>0.05) or higher (P<0.001) lesion CNRs (by 16.5-46.3%) and conspicuity. The VMIs of LD-DL50 keV and LD-DL40 keV were consistent with those of SD-IR50 keV in terms of lesion detection capability in pulmonary nodules [SD-IR50 keV vs. LD-DL50 keV vs. LD-DL40 keV: 88/88 (100.0%) vs. 88/88 (100.0%) vs. 88/88 (100.0%); P>0.99], for lymph nodes [125/126 (99.2%) vs. 123/126 (97.6%) vs. 124/126 (98.4%); P>0.05], and high-contrast liver lesions [12/12 (100.0%) vs. 12/12 (100.0%) vs. 12/12 (100.0%); P>0.05], but not for small liver lesions (≤0.5 cm) [63/65 (96.9%) vs. 43/65 (66.2%) vs. 51/65 (78.5%); P<0.05] or low-contrast liver lesions [198/200 (99.0%) vs. 174/200 (87.0%) vs. 183/200 (91.5%); P<0.05].ConclusionsVMIs at 40 keV with DLIR enables a 50% decrease in the radiation dose while largely maintaining diagnostic capabilities for multidetection of pulmonary nodules, lymph nodes, and liver lesions in oncology patients.
| S-EPMC11400683 | biostudies-literature
Project description:PurposeThis study aimed to investigate the impact of a deep learning (DL)-based denoising method on the image quality and lesion detectability of 18F-FDG positron emission tomography (PET) images.MethodsFifty-two oncological patients undergoing an 18F-FDG PET/CT imaging with an acquisition of 180 s per bed position were retrospectively included. The list-mode data were rebinned into four datasets: 100% (reference), 75%, 50%, and 33.3% of the total counts, and then reconstructed by OSEM algorithm and post-processed with the DL and Gaussian filter (GS). The image quality was assessed using a 5-point Likert scale, and FDG-avid lesions were counted to measure lesion detectability. Standardized uptake values (SUVs) in livers and lesions, liver signal-to-noise ratio (SNR) and target-to-background ratio (TBR) values were compared between the methods. Subgroup analyses compared TBRs after categorizing lesions based on parameters like lesion diameter, uptake or patient habitus.ResultsThe DL method showed superior performance regarding image noise and inferior performance regarding lesion contrast in the qualitative assessment. More than 96.8% of the lesions were successfully identified in DL images. Excellent agreements on SUV in livers and lesions were found. The DL method significantly improved the liver SNR for count reduction down to 33.3% (p < 0.001). Lesion TBR was not significantly different between DL and reference images of the 75% dataset; furthermore, there was no significant difference either for lesions of > 10 mm or lesions in BMIs of > 25. For the 50% dataset, there was no significant difference between DL and reference images for TBR of lesion with > 15 mm or higher uptake than liver.ConclusionsThe developed DL method improved both liver SNR and lesion TBR indicating better image quality and lesion conspicuousness compared to GS method. Compared with the reference, it showed non-inferior image quality with reduced counts by 25-50% under various conditions.
| S-EPMC8816983 | biostudies-literature
Project description:ObjectiveEvaluation of a new iterative reconstruction algorithm (IMR) for detection/rule-out of pulmonary embolism (PE) in ultra-low dose computed tomography pulmonary angiography (CTPA).MethodsLower dose CT data sets were simulated based on CTPA examinations of 16 patients with pulmonary embolism (PE) with dose levels (DL) of 50%, 25%, 12.5%, 6.3% or 3.1% of the original tube current setting. Original CT data sets and simulated low-dose data sets were reconstructed with three reconstruction algorithms: the standard reconstruction algorithm "filtered back projection" (FBP), the first generation iterative reconstruction algorithm iDose and the next generation iterative reconstruction algorithm "Iterative Model Reconstruction" (IMR). In total, 288 CTPA data sets (16 patients, 6 tube current levels, 3 different algorithms) were evaluated by two blinded radiologists regarding image quality, diagnostic confidence, detectability of PE and contrast-to-noise ratio (CNR).ResultsiDose and IMR showed better detectability of PE than FBP. With IMR, sensitivity for detection of PE was 100% down to a dose level of 12.5%. iDose and IMR showed superiority to FBP regarding all characteristics of subjective (diagnostic confidence in detection of PE, image quality, image noise, artefacts) and objective image quality. The minimum DL providing acceptable diagnostic performance was 12.5% (= 0.45 mSv) for IMR, 25% (= 0.89 mSv) for iDose and 100% (= 3.57 mSv) for FBP. CNR was significantly (p < 0.001) improved by IMR compared to FBP and iDose at all dose levels.ConclusionBy using IMR for detection of PE, dose reduction for CTPA of up to 75% is possible while maintaining full diagnostic confidence. This would result in a mean effective dose of approximately 0.9 mSv for CTPA.
| S-EPMC5017721 | biostudies-literature
Project description:PurposeTendency is to moderate the injected activity and/or reduce acquisition time in PET examinations to minimize potential radiation hazards and increase patient comfort. This work aims to assess the performance of regular full-dose (FD) synthesis from fast/low-dose (LD) whole-body (WB) PET images using deep learning techniques.MethodsInstead of using synthetic LD scans, two separate clinical WB 18F-Fluorodeoxyglucose (18F-FDG) PET/CT studies of 100 patients were acquired: one regular FD (~ 27 min) and one fast or LD (~ 3 min) consisting of 1/8th of the standard acquisition time. A modified cycle-consistent generative adversarial network (CycleGAN) and residual neural network (ResNET) models, denoted as CGAN and RNET, respectively, were implemented to predict FD PET images. The quality of the predicted PET images was assessed by two nuclear medicine physicians. Moreover, the diagnostic quality of the predicted PET images was evaluated using a pass/fail scheme for lesion detectability task. Quantitative analysis using established metrics including standardized uptake value (SUV) bias was performed for the liver, left/right lung, brain, and 400 malignant lesions from the test and evaluation datasets.ResultsCGAN scored 4.92 and 3.88 (out of 5) (adequate to good) for brain and neck + trunk, respectively. The average SUV bias calculated over normal tissues was 3.39 ± 0.71% and - 3.83 ± 1.25% for CGAN and RNET, respectively. Bland-Altman analysis reported the lowest SUV bias (0.01%) and 95% confidence interval of - 0.36, + 0.47 for CGAN compared with the reference FD images for malignant lesions.ConclusionCycleGAN is able to synthesize clinical FD WB PET images from LD images with 1/8th of standard injected activity or acquisition time. The predicted FD images present almost similar performance in terms of lesion detectability, qualitative scores, and quantification bias and variance.
| S-EPMC8241799 | biostudies-literature
Project description:Iterative reconstruction has demonstrated superior performance in medical imaging under compressed, sparse, and limited-view sensing scenarios. However, iterative reconstruction algorithms are slow to converge and rely heavily on hand-crafted parameters to achieve good performance. Many iterations are usually required to reconstruct a high-quality image, which is computationally expensive due to repeated evaluations of the physical model. While learned iterative reconstruction approaches such as model-based learning (MBLr) can reduce the number of iterations through convolutional neural networks, it still requires repeated evaluations of the physical models at each iteration. Therefore, the goal of this study is to develop a Fast Iterative Reconstruction (FIRe) algorithm that incorporates a learned physical model into the learned iterative reconstruction scheme to further reduce the reconstruction time while maintaining robust reconstruction performance. We also propose an efficient training scheme for FIRe, which releases the enormous memory footprint required by learned iterative reconstruction methods through the concept of recursive training. The results of our proposed method demonstrate comparable reconstruction performance to learned iterative reconstruction methods with a 9x reduction in computation time and a 620x reduction in computation time compared to variational reconstruction.
| S-EPMC9867977 | biostudies-literature
Project description:Purpose. The purpose of this study was to evaluate the technical and diagnostic performance of sub-milliSievert ultralow-dose (ULD) CT colonograpy (CTC) in the detection of colonic and extracolonic lesions. Materials and Methods. CTC with standard dose (SD) and ULD acquisitions of 64 matched patients, half of them with colonic findings, were reconstructed with filtered back projection (FBP), hybrid (HIR) and iterative model reconstruction techniques (IMR). Image noise in six colonic segments, in the left psoas muscle and aorta were measured. Image quality of the left adrenal gland and of the colon in the endoscopic and 2D view was rated on a five point Likert scale by two observers, who also completed the reading of CTC for colonic and extracolonic findings. Results. The mean radiation dose estimate was 4.1 ± 1.4 mSv for SD and 0.86 ± 0.17 mSv for ULD for both positions (p < 0.0001). In ULD-IMR, SD-IMR and SD-HIR, the endoluminal noise was decreased in all colonic segments compared to SD-FBP (p < 0.001). There were 27 small (6-9 mm) and 17 large (≥10 mm) colonic lesions that were classified as sessile polyps (n = 38), flat lesions (n = 3), or as a mass (n = 3). Per patient sensitivity and specificity were 0.82 and 0.93 for ULD-FBP, 0.97 and 0.97 for ULD-HIR, 0.97 and 1.0 for ULD-IMR. Per polyp sensitivity was 0.84 for ULD-FBP, 0.98 for ULD-HIR, 0.98 for ULD-IMR. Significantly less extracolonic findings were detected in ULD-FBP and ULD-HIR, but in the E4 category by C-RADS (potentially important findings), the detection was similar. Conclusion. Both HIR and IMR are suitable for sub-milliSievert ULD CTC without sacrificing diagnostic performance of the study.
| S-EPMC4824919 | biostudies-literature
Project description:A Compton camera is a device for imaging a radio-source distribution without using a mechanical collimator. Ordered-subset expectation-maximization (OS-EM) is widely used to reconstruct Compton images. However, the OS-EM algorithm tends to over-concentrate and amplify noise in the reconstructed image. It is, thus, necessary to optimize the number of iterations to develop high-quality images, but this has not yet been achieved. In this paper, we apply a median filter to an OS-EM algorithm and introduce a median root prior expectation-maximization (MRP-EM) algorithm to overcome this problem. In MRP-EM, the median filter is used to update the image in each iteration. We evaluated the quality of images reconstructed by our proposed method and compared them with those reconstructed by conventional algorithms using mathematical phantoms. The spatial resolution was estimated using the images of two point sources. Reproducibility was evaluated on an ellipsoidal phantom by calculating the residual sum of squares, zero-mean normalized cross-correlation, and mutual information. In addition, we evaluated the semi-quantitative performance and uniformity on the ellipsoidal phantom. MRP-EM reduces the generated noise and is robust with respect to the number of iterations. An evaluation of the reconstructed image quality using some statistical indices shows that our proposed method delivers better results than conventional techniques.
| S-EPMC7059936 | biostudies-literature
Project description:Very high temporal and spatial resolution is mandatory for the diagnosis of arteriovenous malformations (AVM) of the hand. Until now, magnetic resonance imaging (MRI) has not fulfilled both requirements simultaneously. This study presents how the combination of a very fast TWIST MRI (time-resolved angiography with interleaved stochastic trajectories) sequence and iterative reconstructions optimizes temporal as well as spatial resolution. 11 patients were examined at a 3-T MRI scanner with two different TWIST protocols: the standard and the study protocol, acquiring a data set every 5.57 s and 1.44 s respectively. The study data was retrospectively iteratively reconstructed with different regularization factors (0.001, 0.002, 0.004, 0.008). Results were compared using the sign-test. P-values?<?0.05 were regarded statistically significant. With a low amount of contrast medium, the temporal resolution of the study protocol enabled the differentiation of arteries from veins in all patients whereas the signal-to-noise ratio (SNR) deteriorated. Depending on the regularization factors, SNR, delineation of arterial feeders and non-involved hand and interdigital arteries, as well as artefact levels varied. Overall, iterative reconstruction with regularization factor 0.004 achieved the best results, consequently showing the ability of MRI as a reliable diagnostic method in AVMs of the hand.
| S-EPMC7529883 | biostudies-literature
Project description:OBJECTIVE:To investigate the accuracy of model-based iterative reconstruction (MIR) for volume measurement of part-solid nodules (PSNs) and solid nodules (SNs) in comparison with filtered back projection (FBP) or hybrid iterative reconstruction (HIR) at various radiation dose settings. MATERIALS AND METHODS:CT scanning was performed for eight different diameters of PSNs and SNs placed in the phantom at five radiation dose levels (120 kVp/100 mAs, 120 kVp/50 mAs, 120 kVp/20 mAs, 120 kVp/10 mAs, and 80 kVp/10 mAs). Each CT scan was reconstructed using FBP, HIR, or MIR with three different image definitions (body routine level 1 [IMR-R1], body soft tissue level 1 [IMR-ST1], and sharp plus level 1 [IMR-SP1]; Philips Healthcare). The SN and PSN volumes including each solid/ground-glass opacity portion were measured semi-automatically, after which absolute percentage measurement errors (APEs) of the measured volumes were calculated. Image noise was calculated to assess the image quality. RESULTS:Across all nodules and dose settings, the APEs were significantly lower in MIR than in FBP and HIR (all p < 0.01). The APEs of the smallest inner solid portion of the PSNs (3 mm) and SNs (3 mm) were the lowest when MIR (IMR-R1 and IMR-ST1) was used for reconstruction for all radiation dose settings. (IMR-R1 and IMR-ST1 at 120 kVp/100 mAs, 1.06 ± 1.36 and 8.75 ± 3.96, p < 0.001; at 120 kVp/50 mAs, 1.95 ± 1.56 and 5.61 ± 0.85, p = 0.002; at 120 kVp/20 mAs, 2.88 ± 3.68 and 5.75 ± 1.95, p = 0.001; at 120 kVp/10 mAs, 5.57 ± 6.26 and 6.32 ± 2.91, p = 0.091; at 80 kVp/10 mAs, 5.84 ± 1.96 and 6.90 ± 3.31, p = 0.632). Image noise was significantly lower in MIR than in FBP and HIR for all radiation dose settings (120 kVp/100 mAs, 3.22 ± 0.66; 120 kVp/50 mAs, 4.19 ± 1.37; 120 kVp/20 mAs, 5.49 ± 1.16; 120 kVp/10 mAs, 6.88 ± 1.91; 80 kVp/10 mAs, 12.49 ± 6.14; all p < 0.001). CONCLUSION:MIR was the most accurate algorithm for volume measurements of both PSNs and SNs in comparison with FBP and HIR at low-dose as well as standard-dose settings. Specifically, MIR was effective in the volume measurement of the smallest PSNs and SNs.
| S-EPMC6609437 | biostudies-literature