Towards high resolution cardiac T1 mapping utilizing deep learning-based image reconstruction

To compare the image quality of brain computed tomography (CT) images reconstructed with deep learning-based image reconstruction (DLIR) and adaptive statistical iterative reconstruction-Veo (ASIR-V). Sixty-two patients underwent routine noncontrast brain CT scans and datasets were reconstructed with 30% ASIR-V and DLIR with three selectable reconstruction strength levels (low, medium, high). Objective parameters including CT attenuation, noise, noise reduction rate, artifact index of the posterior cranial fossa, and contrast-to-noise ratio (CNR) were measured at the levels of the centrum semiovale and basal ganglia. Subjective parameters including gray matter-white matter differentiation, sharpness, and overall diagnostic quality were also assessed and compared with the interobserver agreement. There was a gradual reduction in the image noise and artifact index of the posterior cranial fossa as the strength levels of DLIR increased (all P < 0.001) compared with that of ASIR-V. CNR in both the centrum semiovale and basal ganglia levels also improved from the low to high strength levels of DLIR compared with that of ASIR-V (all P < 0.001). DLIR images with medium and high strength levels demonstrated the best subjective image quality scores among the reconstruction datasets. There was moderate to good interobserver agreement for the subjective image quality assessments with ASIR-V and DLIR. On routine brain CT scans, optimized protocols with DLIR allowed significant reduction of noise and artifacts with improved subjective image quality compared with ASIR-V.

BackgroundCardiac complications are the leading cause of death in transfusion-dependent thalassemia major (TDT) patients due to increased cardiac iron overload (CIO), so this study’s purpose is to assess the diagnostic utility of cardiac magnetic resonance T1 mapping in correlation to T2* for the assessment of CIO in TDT patients. This prospective case–control study involved 55 TDT patients and 30 age- and gender-matched healthy controls. All the patients underwent assessment of myocardial T2*, and native T1 values by conventional cardiac magnetic resonance imaging, and native T1 mapping technique to assess the CIO. Comparison and association between T2*, and native T1 values among TDT patients were studied.ResultsA statistically significant difference is noted between control and patient groups in regard to mean T2* and native T1 values (P = 0.03 and 0.01, respectively). The mean native T1 value in TDT patients was lower than in the control group (992 ± 54 vs. 1234 ± 42), respectively, with a statistically significant P value of 0.04. The study evoked a statistically significant strong positive association, and correlation was noted between measured T2* and native T1 values (r = 0.84, P value = 0.001), while a moderate negative association was observed between native T1 values and S. ferritin levels (r = − 0.53, P value = 0.02).ConclusionsNative T1 mapping is a non-invasive promising complementary technique to T2* that helps to detect cardiac iron overload in TDT patients, as it is strongly associated with T2*. Also, native T1 value can differentiate mild, moderate, severe, and no cardiac iron overload.

Fat-suppressed T2-weighted imaging (T2-FS) requires a long scan time and can be wrought with motion artifacts, urging the development of a shorter and more motion robust sequence. We compare the image quality of a single-shot T2-weighted MRI prototype with deep-learning-based image reconstruction (DL HASTE-FS) with a standard T2-FS sequence for 3T liver MRI. 41 consecutive patients with 3T abdominal MRI examinations including standard T2-FS and DL HASTE-FS, between 5/6/2020 and 11/23/2020, comprised the study cohort. Three radiologists independently reviewed images using a 5-point Likert scale for artifact and image quality measures, while also assessing for liver lesions. DL HASTE-FS acquisition time was 54.93 ± 16.69, significantly (p < .001) shorter than standard T2-FS (114.00 ± 32.98s). DL HASTE-FS received significantly higher scores for sharpness of liver margin (4.3 vs 3.3; p < .001), hepatic vessel margin (4.2 vs 3.3; p < .001), pancreatic duct margin (4.0 vs 1.9; p < .001); in-plane (4.0 vs 3.2; p < .001) and through-plane (3.9 vs 3.4; p < .001) motion artifacts; other ghosting artifacts (4.3 vs 2.9; p < .001); and overall image quality (4.0 vs 2.9; p < .001), in addition to receiving a higher score for homogeneity of fat suppression (3.7 vs 3.4; p = .04) and liver-fat contrast (p = .03). For liver lesions, DL HASTE-FS received significantly higher scores for sharpness of lesion margin (4.4 vs 3.7; p = .03). Novel single-shot T2-weighted MRI with deep-learning-based image reconstruction demonstrated superior image quality compared with the standard T2-FS sequence for 3T liver MRI, while being acquired in less than half the time.

OBJECTIVE: To evaluate image quality of deep learning-based image reconstruction (DLIR) in contrast-enhanced renal and adrenal computed tomography (CT) compared with adaptive statistical iterative reconstruction-Veo (ASiR-V).METHODS: We prospectively recruited 52 patients. All images were reconstructed with ASiR-V 30%, ASiR-V 70%, and DLIR at low, medium, and high reconstruction strengths. CT number, noise, noise reduction rate, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) were measured and calculated within the region of interest (ROI) on subcutaneous fat, bilateral renal cortices, renal medulla, renal arteries, and adrenal glands. For qualitative analyses, the differentiation of the renal cortex and medulla, conspicuity of the adrenal gland boundary, sharpness, artifacts, and subjective noise were assessed. The overall image quality was calculated on a scale from 0 (worst) to 15 (best) based on the five values above and the score≥9 was acceptable.RESULTS: CT number does not significantly differ between the reconstruction datasets. Noise does not significantly differ between ASiR-V 30% and DLIR-L, but it is significantly lower using ASiR-V 70%, DLIR-M, and DLIR-H. The noise reduction rate relative to ASiR-V 30% is significantly different between the DLIR groups and ASiR-V 70%, and DLIR-H yields the highest noise reduction rate (61.6%). SNR and CNR are higher for DLIR-M, DLIR-H, and ASiR-V 70% than for ASiR-V 30% and DLIR-L. DLIR-H shows the best SNR and CNR. The overall image quality yields the same pattern for DLIR-H, with the highest score. Percentages of cases with overall image quality score≥9 are 100% (DLIR-H), 94.23% (DLIR-M), 90.38% (ASiR-V70%), 67.31% (DLIR-L), and 63.46% (ASiR-V30%), respectively.CONCLUSIONS: DLIR significantly improved the objective and subjective image quality of renal and adrenal CTs, yielding superior noise reduction compared with ASiR-V.

Use of Cardiac Magnetic Resonance Imaging to Distinguish Between Acute Myocarditis and Takotsubo Cardiomyopathy

Motivation: Improved SNR characterization of images with Deep Learning-based image reconstruction (DLR). Goal(s): To evaluate whether SNR measurements using improved noise estimates demonstrate expected SNR changes with common sequence acquisition modifications in images reconstructed using DLR. Approach: Sequentially acquired images were used to calculate theoretically accurate noise estimates in the brain and abdomen. Each dataset consisted of two sequentially acquired T2-weighted series. Five distinct SNR estimates were calculated from the same data. Results: SNR estimates using region-based noise measurements were less accurate than sequentially acquired estimates. Furthermore, these SNR estimates cannot reliably measure relative SNR with common sequence parameters modifications. Impact: SNR measured from sequential images in the brain and abdomen demonstrated that DLR improved SNR. Results from single-image noise estimates were inaccurate. Further, results from known parameter modifications demonstrated that DLR underestimated the expected relative SNR differences in all methods.

The aim of this study was to evaluate the regional (i.e. myocardial segments) variability as well as the overall image quality of cardiac T1 and T2 maps obtained in diastole and in systole. In 22 healthy subjects (group-1), diastolic T1 and T2 maps were obtained at 1.5 T in short-axis view at basal, mid-ventricular and apical level, as well as in 4-chamber (4 ch) and in 2-chamber (2 ch) views. In another group of 25 patients (group-2), the maps were obtained in both diastole and systole. In the group-1, 15.4% of myocardial segments in T1 maps and 0.8% of myocardial segments in T2 maps, mainly located at apical level, showed relevant artifacts and/or partial-volume effect and had to be discarded. We found no significant difference in T1 values among basal, mid-ventricular and apical segments. T2 values at apical level were significantly higher than at basal and mid-ventricular level (short-axis, p < 0.0001; 4 ch, p < 0.009; 2 ch, p = 0.0002 at ANOVA tests). In the group-2, 21.1%/5.3% and 4.0%/0.8% of segments showed relevant artifacts in diastolic/systolic T1 and T2 maps, respectively. Apical T2 values were significantly lower in systole than in diastole. In systole, there were no significant differences in T1/T2 among basal, mid-ventricular and apical segments. The overall quality of T1 and T2 maps drops in apical segments. This could be problematic when evaluating focal myocardial changes. The acquisition in systole increases the number of evaluable segments.

Sparsity-based and deep learning-based image reconstruction algorithms are two promising approaches to accelerate the image acquisition process for localization-based super-resolution microscopy, by allowing a higher density of fluorescing emitters to be imaged in a single frame. Despite the surging popularity, a comprehensive parametric study guiding the practical applications of sparsity-based and deep learning-based image reconstruction algorithms is yet to be conducted. In this study, we examined the performance of sparsity- and deep learning-based algorithms in reconstructing super-resolution images using simulated fluorescent microscopy images. The simulated images were synthesized with varying levels of sparsity and connectivity. We found the deep learning-based VDSR recovers image faster, with a higher recall rate and localization accuracy. The sparsity-based SPIDER recovers more zero pixels truthfully. We also compared the two algorithms using images acquired from a real super-resolution experiment, yielding results agreeing with the results from the evaluation using simulated images. We concluded that VDSR is preferable when accurate emitter localization is needed while SPIDER is more suitable when evaluation of the number of emitters is critical.

Background Accumulation of diffuse myocardial fibrosis / extracellular volume (ECV) in the myocardium is common to various cardiac diseases and is associated with an unfavorable prognosis. Various cardiac magnetic resonance (CMR) T1 mapping techniques have recently been proposed for the quantification of ECV, including: 1. Modified Look-Locker Inversion recovery (MOLLI) T1 mapping including calculation of ECV, 2. Post-contrast multiple-breath-hold T1 mapping. 3. Native (precontrast) T1 mapping. T1 mapping is a promising technique, however, validation data based on myocardial biopsy are sparse. Methods 36 heart failure patients underwent CMR T1 mapping and left-ventricular biopsy within 4 weeks. CMR protocols (1.5-T scanner, (Magnetom Avanto, Siemens Healthcare, Erlangen, Germany) included non-product MOLLI and multiple-breath-hold T1 mapping. The population consisted of 22 patients with HFpEF (heart failure with preserved ejection fraction), 7 with cardiac amyloidosis, 3 with HFrEF (heart failure with reduced ejection fraction) and 4 with organic mitral regurgitation (MR). Specimens were stained using modified Trichrome. TissueFAXS and dedicated software were used to quantify ECV. Results ECV by TissueFAXS was 32±17% and 35±13% by CMR MOLLI. Post-contrast T1 times by the multiple-breathhold sequence were 411±79ms, native T1 times were 1009±73ms. Patients with amyloidosis had significantly more ECV by TissueFAXS (59±13%, p<0.001) and with CMR MOLLI (56±12%, p<0.001) compared with the other pathologies. The amount of TissueFAXS ECV correlated significantly with native T1 times (r=0.657, p<0.001), MOLLI ECV (r=0.907, p<0.001) and with multiple-breath-hold post-contrast T1 times (r=-0.683, p<0.001). When excluding patients with amyloidosis, only MOLLI ECV significantly correlated with histology (r=0.491, p=0.015), whereas multiple-breath-hold post-contrast T1 times just missed the level of significance (r=-0.413, p=0.056).

To improve the accuracy and robustness of T1 estimation by MyoMapNet, a deep learning-based approach using 4 inversion-recovery T1 -weighted images for cardiac T1 mapping. MyoMapNet is a fully connected neural network for T1 estimation of an accelerated cardiac T1 mapping sequence, which collects 4 T1 -weighted images by a single Look-Locker inversion-recovery experiment (LL4). MyoMapNet was originally trained using in vivo data from the modified Look-Locker inversion recovery sequence, which resulted in significant bias and sensitivity to various confounders. This study sought to train MyoMapNet using signals generated from numerical simulations and phantom MR data under multiple simulated confounders. The trained model was then evaluated by phantom data scanned using new phantom vials that differed from those used for training. The performance of the new model was compared with modified Look-Locker inversion recovery sequence and saturation-recovery single-shot acquisition for measuring native and postcontrast T1 in 25 subjects. In the phantom study, T1 values measured by LL4 with MyoMapNet were highly correlated with reference values from the spin-echo sequence. Furthermore, the estimated T1 had excellent robustness to changes in flip angle and off-resonance. Native and postcontrast myocardium T1 at 3Tesla measured by saturation-recovery single-shot acquisition, modified Look-Locker inversion recovery sequence, and MyoMapNet were 1483 ± 46.6ms and 791 ± 45.8ms, 1169 ± 49.0ms and 612 ± 36.0ms, and 1443 ± 57.5ms and 700 ± 57.5ms, respectively. The corresponding extracellular volumes were 22.90% ± 3.20%, 28.88% ± 3.48%, and 30.65% ± 3.60%, respectively. Training MyoMapNet with numerical simulations and phantom data will improve the estimation of myocardial T1 values and increase its robustness to confounders while also reducing the overall T1 mapping estimation time to only 4 heartbeats.

A vectorized Levenberg-Marquardt model fitting algorithm for efficient post-processing of cardiac T1 mapping MRI

Background The development of myocardial fibrosis is a key mechanism in the transition from compensated hypertrophy to heart failure in aortic stenosis (AS). Focal and diffuse fibrosis can be quantified using cardiac magnetic resonance (CMR) imaging late gadolinium-enhanced (LGE) and T1 mapping techniques. Purpose To assess T1 mapping measures of fibrosis in patients with severe AS referred for aortic valve intervention, and determine their associations with clinical characteristics, disease severity and long-term clinical outcome. Methods In this international prospective cohort study, patients with severe AS underwent contrast enhanced CMR with T1 mapping and LGE prior to aortic valve intervention. Image analysis was performed by a single core laboratory and the extracellular volume fraction [ECV%] calculated from T1 mapping images. The presence of LGE was determined visually and quantified using the full-width-at-half-maximum technique. Results Four-hundred and forty patients (70±10 years, 59% male) from ten international centres were enrolled. Aortic valve intervention was performed 15 [4 to 58] days following CMR. Within a follow-up of 3.8 [2.8 to 4.6] years, 52 patients died. ECV% (mean 27.7±3.6%) correlated with increasing age, Society of Thoracic Surgeons Predicted Risk of Mortality score, known coronary artery disease, lower peak aortic-jet velocity, larger left ventricular (LV) mass, lower LV ejection fraction, and presence of LGE (P&lt;0.05 for all). Following adjustment for all demographic and clinical variables, ECV% remained associated with age (P=0.028), LV ejection fraction (P&lt;0.001) and presence of LGE (P=0.035). Univariable predictors of all-cause mortality included age, male sex, impaired LV ejection fraction and presence of LGE (all P&lt;0.05). A progressive increase in all-cause mortality was seen across tertiles of ECV% (17.3, 31.6 and 52.7 deaths per 1000 patient-years; log-rank test, P=0.009). ECV% was independently associated with all-cause mortality following adjustment for age, sex, impaired LV ejection fraction and presence of LGE (HR per unit increase in ECV: 1.10, 95%, (1.02–1.19), P=0.013). ECV440 abstract iamge Conclusion In patients with severe aortic stenosis scheduled for aortic valve intervention, extracellular volume-based T1 mapping correlates with LV decompensation. ECV% is a strong independent predictor of late all-cause mortality and is a potential therapeutic target.

Magnetic resonance imaging (MRI) is the keystone of modern musculoskeletal imaging; however, long pulse sequence acquisition times may restrict patient tolerability and access. Advances in MRI scanners, coil technology, and innovative pulse sequence acceleration methods enable 4-fold turbo spin echo pulse sequence acceleration in clinical practice; however, at this speed, conventional image reconstruction approaches the signal-to-noise limits of temporal, spatial, and contrast resolution. Novel deep learning image reconstruction methods can minimize signal-to-noise interdependencies to better advantage than conventional image reconstruction, leading to unparalleled gains in image speed and quality when combined with parallel imaging and simultaneous multislice acquisition. The enormous potential of deep learning-based image reconstruction promises to facilitate the 10-fold acceleration of the turbo spin echo pulse sequence, equating to a total acquisition time of 2-3 minutes for entire MRI examinations of joints without sacrificing spatial resolution or image quality. Current investigations aim for a better understanding of stability and failure modes of image reconstruction networks, validation of network reconstruction performance with external data sets, determination of diagnostic performances with independent reference standards, establishing generalizability to other centers, scanners, field strengths, coils, and anatomy, and building publicly available benchmark data sets to compare methods and foster innovation and collaboration between the clinical and image processing community. In this article, we review basic concepts of deep learning-based acquisition and image reconstruction techniques for accelerating and improving the quality of musculoskeletal MRI, commercially available and developing deep learning-based MRI solutions, superresolution, denoising, generative adversarial networks, and combined strategies for deep learning-driven ultra-fast superresolution musculoskeletal MRI. This article aims to equip radiologists and imaging scientists with the necessary practical knowledge and enthusiasm to meet this exciting new era of musculoskeletal MRI.

Background Antibody-mediated rejection (AMR) is characterized by histopathological and immunophenotypic findings such as activated endothelial cells, intravascular macrophages and evidence of capillary C4d deposition. This inflammatory reaction could be followed by diffuse fibrosis. Cardiac magnetic resonance (CMR) with recently T1 mapping is a promising technique to identify diffuse myocardial fibrosis. The purpose of this study was to assess T1 mapping in patients with AMR. Methods Two patients with clinical AMR (histopathological and immunophenotypic findings, presence of donor-specific allo antibodies and allograft dysfunction) performed a CMR study one week (for the first patient) and 3 weeks (for the second patient) after the treatment of AMR (plasmapheresis, IV Immunoglobulins and Rituximab). Images were acquired on a 1.5 Tesla scanner (Siemens) including T1 mapping using a shortened modified look-locker inversion-recovery sequence and T2 mapping in a matched mid-ventricular short axis slice using a black- blood single shot fast spin echo pulse sequence. Segmental and global T1 values were measured before and 15 minutes after administration of 0.2 mmol/kg of Gadoteric acid and compared to our cohort of 17 controls. Results Mean non contrast T1 values were significantly higher in heart transplants patients compared to controls (1100 ± 5 ms vs 947 ± 29 ms, P < 0.001). Segmental T1 values were significantly higher in the 6 regions of interest compared to controls (P < 0.001 in all segments). Mean post contrast T1 values were not significantly different in patients and controls. Mean T2 value was higher in patients compared to controls (73 ± 13 vs 50 ± 4 ms), suggesting the presence of global edema. Conclusions Heart transplant patients with clinical antibodymediated rejection show a significant increased global and segmental non contrast T1 values suggesting the presence of diffuse myocardial fibrosis. Further studies are required to confirm these data. Funding

Introduction:Primary aldosteronism (PA) is associated with cardiovascular disease and has increased cardiovascular morbidity. Several studies had demonstrated that PA shows more serious myocardial fibrosis compared with essential hypertension (EH) using late Gadolinium-enhanced (LGE) imaging [1, 2]. However, the early myocardial change may not be detected by LGE. Recently, cardiac MRI T1 mapping emerged to be a sensitive technique in detecting early changes of myocardium quantitatively. Our study aimed to demonstrate the early myocardial change in PA patients by applying the T1-mapping technique.Method:81 (male: 25; age: 48±12 years) PA[3] and 44 (male: 21; age: 47±14 years) EH patients diagnosed by the Department of Endocrinology and metabolism according to the 2016 TES guidelines for PA were recruited from September 2018 to May 2019. All the subjects underwent cardiac MRI examinations on a 3T MRI scanner (MAGNETOM Trio a Tim System, Siemens Healthcare, Erlangen, Germany), including steady-state free-precession (SSFP) cine imaging and modified Look-Locker inversion recovery (MOLLI) imaging for T1 mapping. Functional parameters were obtained from cine images by the Argus software (Siemens Healthcare, Erlangen, Germany).The mid-ventricular native T1 value was acquired using the QMass software (Medis, Leiden, The Netherlands). Native T1 value and cardiac function parameters were analyzed between the two groups. Differences of continuous variables were analyzed by Student’s t-test or Mann-Whitney U-test, and the relationships between native T1 value and physiological variables were analyzed by multiple linear regression method.Results:The age (PA: 48±12 years vs EH: 47±14 years; p=0.720) and BMI (PA: 25.06±3.73 vs EH: 25.17±2.83; P=0.874) between two groups had no significant differences. The PA group had a significant higher aldosterone/renin ratio (ARR) than those in the EH group (114.6 (48.7-474.8)ng/dl:ng/ml.h vs 7.6 (4.2-13.8) ng/dl:ng/ml.h; p<0.001). All functional parameters including LVEDVI (PA: 78.66 (72.4-90.6) ml/m2 vs EH: 75.8 (64.3-85.8) ml/m2; p=0.054); LVESVI (PA: 31.7 (26.2-38.4) ml/m2 vs EH: 29.3 (24.9-36.2) ml/m2; p=0.152); LVEF (PA: 59.9 (53.3-65.0)% VS EH: 59.2(55.3-65.6)%; p=0.679); LVmassi (PA: 58.5(47.7-67.7) g/m2 vs EH: 54.9 (47.6-60.1) g/m2; p=0.463) had no differences between the two groups. However, native T1 values were higher in PA group than those in EH group (1227±40ms vs 1203±45ms). The multiple linear regression analysis showed that gender (Beta=-27.678, p<0.001) and PA (Beta=-17.287, p=0.031) were independently related to the native T1 values.Conclusion:PA patients represent more severe and earlier myocardium damage and this might be related to cardiovascular morbidity. The T1-mapping technique in cardiac MRI is more sensitive to evaluate the left ventricular function and to detect the myocardial fibrosis in PA patients.