Search

PURPOSE: To develop a three-dimensional (two dimensions + time) convolutional neural network trained with displacement encoding with stimulated echoes (DENSE) data for displacement and strain analysis of cine MRI. MATERIALS AND METHODS: In this retrospective multicenter study, a deep learning model (StrainNet) was developed to predict intramyocardial displacement from contour motion. Patients with various heart diseases and healthy controls underwent cardiac MRI examinations with DENSE between August 2008 and January 2022. Network training inputs were a time series of myocardial contours from DENSE magnitude images, and ground truth data were DENSE displacement measurements. Model performance was evaluated using pixelwise end-point error (EPE). For testing, StrainNet was applied to contour motion from cine MRI. Global and segmental circumferential strain (E(cc)) derived from commercial feature tracking (FT), StrainNet, and DENSE (reference) were compared using intraclass correlation coefficients (ICCs), Pearson correlations, Bland-Altman analyses, paired t tests, and linear mixed-effects models. RESULTS: The study included 161 patients (110 men; mean age, 61 years ± 14 [SD]), 99 healthy adults (44 men; mean age, 35 years ± 15), and 45 healthy children and adolescents (21 males; mean age, 12 years ± 3). StrainNet showed good agreement with DENSE for intramyocardial displacement, with an average EPE of 0.75 mm ± 0.35. The ICCs between StrainNet and DENSE and FT and DENSE were 0.87 and 0.72, respectively, for global E(cc) and 0.75 and 0.48, respectively, for segmental E(cc). Bland-Altman analysis showed that StrainNet had better agreement than FT with DENSE for global and segmental E(cc). CONCLUSION: StrainNet outperformed FT for global and segmental E(cc) analysis of cine MRI. Keywords: Image Postprocessing, MR Imaging, Cardiac, Heart, Pediatrics, Technical Aspects, Technology Assessment, Strain, Deep Learning, DENSE Supplemental material is available for this article. © RSNA, 2023

Objectives To investigate the efficacy of an in-line non-rigid motion-compensated reconstruction (NRC) in an image-navigated high-resolution three-dimensional late gadolinium enhancement (LGE) sequence with Dixon water–fat separation, in a clinical setting. Methods Forty-seven consecutive patients were enrolled prospectively and examined with 1.5 T MRI. NRC reconstructions were compared to translational motion-compensated reconstructions (TC) of the same datasets in overall and different sub-category image quality scores, diagnostic confidence, contrast ratios, LGE pattern, and semiautomatic LGE quantification. Results NRC outperformed TC in all image quality scores (p < 0.001 to 0.016; e.g., overall image quality 5/5 points vs. 4/5). Overall image quality was downgraded in only 23% of NRC datasets vs. 53% of TC datasets due to residual respiratory motion. In both reconstructions, LGE was rated as ischemic in 11 patients and non-ischemic in 10 patients, while it was absent in 26 patients. NRC delivered significantly higher LGE-to-myocardium and blood-to-myocardium contrast ratios (median 6.33 vs. 5.96, p < 0.001 and 4.88 vs. 4.66, p < 0.001, respectively). Automatically detected LGE mass was significantly lower in the NRC reconstruction (p < 0.001). Diagnostic confidence was identical in all cases, with high confidence in 89% and probable in 11% datasets for both reconstructions. No case was rated as inconclusive. Conclusions The in-line implementation of a non-rigid motion-compensated reconstruction framework improved image quality in image-navigated free-breathing, isotropic high-resolution 3D LGE imaging with undersampled spiral-like Cartesian sampling and Dixon water–fat separation compared to translational motion correction of the same datasets. The sharper depictions of LGE may lead to more accurate measures of LGE mass. Key Points • 3D LGE imaging provides high-resolution detection of myocardial scarring. • Non-rigid motion correction provides better image quality in cardiac MRI. • Non-rigid motion correction may lead to more accurate measures of LGE mass.

Context: Hamstring muscle injury location using magnetic resonance imaging (MRI) is not so well described in the literature. Objective: To describe the location of hamstring injuries using MRI. Data Sources: PubMed, Web of Science, Scopus, SPORTDiscus, Cochrane Library. Study Selection: The full text of studies, in English, had to be available. Case reports and reviews were excluded. Included studies must report the location of hamstring injuries using MRI within 8 days of the acute injury. Study Design: Systematic review. Level of Evidence: Level 4. Data Extraction: A first screening was conducted based on title and abstract of the articles. In the second screening, the full text of the remaining articles was evaluated for the fulfillment of the inclusion criteria. Results: From the 2788 references initially found in 5 databases, we included 34 studies, reporting a total of 2761 acute hamstring injuries. The most frequent muscle head involved was the long head of the biceps femoris (BFLH) (70%), followed by the semitendinosus (ST) (15%), generally associated with BFLH. The most frequent tissue affected was the myotendinous junction (MTJ) accounting for half the cases (52%). Among all lesions, the distribution between proximal, central, and distal lesions looked homogenous, with 34.0%, 33.4% and 32.6%, respectively. The stretching mechanism, while only reported in 2 articles, represented 3% of all reported mechanisms, appears responsible for a specific lesion involving the proximal tendon of the semimembranosus (SM), and leading to a longer time out from sport. Conclusion: BFLH was the most often affected hamstring injuries and MTJ was the most affected tissue. In addition, the distal, central, and proximal locations were homogeneously distributed. We also noted that MRI descriptions of hamstring injuries are often poor and did not take full advantage of the MRI strengths. Systematic Review Registration: Before study initiation, the study was registered in the PROSPERO International prospective register of systematic reviews (registration number CRD42018107580).

International audience; Cardiac diffusion tensor imaging (DTI) is an emerging technique for the in vivo characterisation of myocardial microstructure, and there is a growing need for its validation and standardisation. We sought to establish the accuracy, precision, repeatability and reproducibility of state-of-the-art pulse sequences for cardiac DTI among 10 centres internationally. Phantoms comprising 0%-20% polyvinylpyrrolidone (PVP) were scanned with DTI using a product pulsed gradient spin echo (PGSE; N = 10 sites) sequence, and a custom motion-compensated spin echo (SE; N = 5) or stimulated echo acquisition mode (STEAM; N = 5) sequence suitable for cardiac DTI in vivo. A second identical scan was performed 1-9 days later, and the data were analysed centrally. The average mean diffusivities (MDs) in 0% PVP were (1.124, 1.130, 1.113) x 10-3 mm2 /s for PGSE, SE and STEAM, respectively, and accurate to within 1.5% of reference data from the literature. The coefficients of variation in MDs across sites were 2.6%, 3.1% and 2.1% for PGSE, SE and STEAM, respectively, and were similar to previous studies using only PGSE. Reproducibility in MD was excellent, with mean differences in PGSE, SE and STEAM of (0.3 ± 2.3, 0.24 ± 0.95, 0.52 ± 0.58) x 10-5 mm2 /s (mean ± 1.96 SD). We show that custom sequences for cardiac DTI provide accurate, precise, repeatable and reproducible measurements. Further work in anisotropic and/or deforming phantoms is warranted.

International audience; Cardiac perfusion MRI is proving to be a valuable noninvasive assessment tool in the armamentarium for patients with suspected coronary artery disease or microvascular disease, both for initial diagnosis and for follow-up to therapeutic decision. In particular, it has recently been shown to be comparable and more cost-effective than invasive fractional flow reserve for therapeutic guidance of patients with suspected CAD 1. This is especially true since the long-awaited results of the ISCHEMIA study challenge invasive management of coronary artery disease and demonstrate the value of optimal medical treatment 2. The Orbita study also masterfully demonstrated that an invasive approach is not the ultimate solution to propose to symptomatic coronary patients 3. In this context of decreasing radicality and slowing down of the therapeutic approach with an increasing role of an optimized medical approach, it is clear that dynamic cardiac perfusion MRI with contrast imaging therefore promises to become the cornerstone of regular follow-up that will be proposed in future guidelines.

In our book “Navier–Stokes Equations in Planar Domains”, Imperial College Press, 2013, we have suggested a fourth-order compact scheme for the Navier–Stokes equations in streamfunction formulation $$\partial _t(\Delta \psi )+(\nabla ^{\perp }\psi ) \cdot \nabla (\Delta \psi ) =\nu \Delta ^2 \psi $$ . Here we present a new approach for the analysis of a high-order compact scheme for the Navier–Stokes equations in cases where the convective term vanishes, or in cases where the viscous term dominates the convective term. In these cases the Navier–Stokes equations is replaced by the time-dependent Stokes equation $$\partial _t(\Delta \psi )=\nu \Delta ^2 \psi $$ . The same type of fourth-order compact schemes that were proposed for the Navier–Stokes equations, may be adopted to the time-dependent Stokes problem. For these methods the truncation error is only of first-order at near-boundary points, but is of fourth order at interior points. We prove that the rate of convergence is actually four, thus the error tends to zero as $$O(h^4)$$ , where h is the size of the mesh.

Background While multiple cardiovascular magnetic resonance (CMR) methods provide excellent reproducibility of global circumferential and global longitudinal strain, achieving highly reproducible segmental strain is more challenging. Previous single-center studies have demonstrated excellent reproducibility of displacement encoding with stimulated echoes (DENSE) segmental circumferential strain. The present study evaluated the reproducibility of DENSE for measurement of whole-slice or global circumferential (Ecc), longitudinal (Ell) and radial (Err) strain, torsion, and segmental Ecc at multiple centers. Methods Six centers participated and a total of 81 subjects were studied, including 60 healthy subjects and 21 patients with various types of heart disease. CMR utilized 3 T scanners, and cine DENSE images were acquired in three short-axis planes and in the four-chamber long-axis view. During one imaging session, each subject underwent two separate DENSE scans to assess inter-scan reproducibility. Each subject was taken out of the scanner and repositioned between the scans. Intra-user, inter-user-same-site, inter-user-different-site, and inter-user-Human-Deep-Learning (DL) comparisons assessed the reproducibility of different users analyzing the same data. Inter-scan comparisons assessed the reproducibility of DENSE from scan to scan. The reproducibility of whole-slice or global Ecc, Ell and Err, torsion, and segmental Ecc were quantified using Bland–Altman analysis, the coefficient of variation (CV), and the intraclass correlation coefficient (ICC). CV was considered excellent for CV ≤ 10%, good for 10% 40. ICC values were considered excellent for ICC > 0.74, good for ICC 0.6 Results Based on CV and ICC, segmental Ecc provided excellent intra-user, inter-user-same-site, inter-user-different-site, inter-user-Human-DL reproducibility and good–excellent inter-scan reproducibility. Whole-slice Ecc and global Ell provided excellent intra-user, inter-user-same-site, inter-user-different-site, inter-user-Human-DL and inter-scan reproducibility. The reproducibility of torsion was good–excellent for all comparisons. For whole-slice Err, CV was in the fair-good range, and ICC was in the good–excellent range. Conclusions Multicenter data show that 3 T CMR DENSE provides highly reproducible whole-slice and segmental Ecc, global Ell, and torsion measurements in healthy subjects and heart disease patients.

Supplemental Digital Content is available in the text., Background: Inflammation is a key factor of myocardial damage in reperfused ST-segment–elevation myocardial infarction. We hypothesized that colchicine, a potent anti-inflammatory agent, may reduce infarct size (IS) and left ventricular (LV) remodeling at the acute phase of ST-segment–elevation myocardial infarction. Methods: In this double-blind multicenter trial, we randomly assigned patients admitted for a first episode of ST-segment–elevation myocardial infarction referred for primary percutaneous coronary intervention to receive oral colchicine (2-mg loading dose followed by 0.5 mg twice a day) or matching placebo from admission to day 5. The primary efficacy outcome was IS determined by cardiac magnetic resonance imaging at 5 days. The relative LV end-diastolic volume change at 3 months and IS at 3 months assessed by cardiac magnetic resonance imaging were among the secondary outcomes. Results: We enrolled 192 patients, 101 in the colchicine group and 91 in the control group. At 5 days, the gadolinium enhancement–defined IS did not differ between the colchicine and placebo groups with a mean of 26 interquartile range (IQR) [16–44] versus 28.4 IQR [14–40] g of LV mass, respectively (P=0.87). At 3 months follow-up, there was no significant difference in LV remodeling between the colchicine and placebo groups with a +2.4% (IQR, –8.3% to 11.1%) versus –1.1% (IQR, –8.0% to 9.9%) change in LV end-diastolic volume (P=0.49). Infarct size at 3 months was also not significantly different between the colchicine and placebo groups (17 IQR [10–28] versus 18 IQR [10–27] g of LV mass, respectively; P=0.92). The incidence of gastrointestinal adverse events during the treatment period was greater with colchicine than with placebo (34% versus 11%, respectively; P=0.0002). Conclusions: In this randomized, placebo-controlled trial, oral administration of high-dose colchicine at the time of reperfusion and for 5 days did not reduce IS assessed by cardiac magnetic resonance imaging. Registration: URL: https://www.clinicaltrials.gov; Unique identifier: NCT03156816.

Among imaging techniques, magnetic resonance imaging (MRI) is one of the most versatile, capable of producing not only morphological but also quantitative and functional data. Due to its non-irradiative nature, it allows in particular temporal and longitudinal studies. The optimal exploitation of all these potentialities nevertheless requires constructing and optimizing an automatic framework for image processing and analysis. The image segmentation is a fundamental step providing the separation of anatomical structures into individual regions. The massive production of the available image data and the multiplication and complexity of the anatomical structures to be by-passed makes manual segmentation obsolete due to the time required and the lack of motivation concerning the task. It is, therefore, necessary to implement suitable automatic segmentation methods.This thesis work was based on the questions of the MUST research project, whose objective was to help understand the effect of ultra-endurance on the body and its impact on the muscular level in particular. The researchers at the CREATIS laboratory chose to conduct this study on the mountain ultra-marathon, long considered the most extreme event in the world: the Tor des Géants. This study notably used a mobile MRI equipped with advanced imaging techniques to study the longitudinal variations during a supra-physiological effort. In order to study the evolution of inflammation and muscle functional parameters at the quadriceps level, our thesis studies and improves automatic segmentation methods based on supervised approaches. Our objective is to provide a clinically applicable method to segment the quadriceps muscle heads as precisely as possible in longitudinal and requiring as little manual segmentation as possible for the learning phase.With this objective, we first explore and apply multi-atlas approaches for quadriceps segmentation. Our contributions allow us to obtain quality segmentations on a large part of our dataset. However, in order to have a faster and more robust approach, we then oriented our work towards deep learning approaches. In this methodological context, our two main contributions are i) the proposal of a segmentation correction step based on machine learning and ii) the proposal of data augmentation strategies to optimize the learning of U-Net type networks, in particular, a strategy based on morphological resemblance evaluated thanks to an original measurement. This morphology measurement also proved to be very efficient for selecting atlases for the multi-atlas segmentation approach. Finally, we show that the proposed approaches can be generalized to other muscle segmentation problems in MRI and allow longitudinal and localized statistical studies.Our work shows that, even if the quantity of annotated data is crucial in supervised learning, we must also pay attention to our database's morphological diversity, which eventually reduces computation time and increases the precision and robustness of the segmentation methods., Parmi les techniques d’imagerie, l'imagerie par résonance magnétique (IRM) est l’une des techniques d'imagerie les plus versatiles, capable de produire des données morphologiques mais également quantitatives et fonctionnelles. De par son caractère non-irradiant, elle permet en particulier les études temporelles et longitudinales. L’exploitation optimale de toutes ces potentialités requiert néanmoins la construction et l’optimisation de séquences de processus automatiques de traitements et d'analyses des images. L’étage de segmentation des images est une étape fondamentale garantissant la précision de l’exploitation des données, en permettant la séparation des structures anatomiques en différentes régions. La production massive des données image disponibles et la multiplication et complexité des structures anatomiques à contourer, rend caduque une segmentation manuelle en raison de la durée requise et la démotivation au regard de la tâche. Dès lors, il est nécessaire de mettre en place des méthodes de segmentation automatique adaptées.Ce travail de thèse s’est nourri des problématiques du projet de recherche MUST dont l’objectif a été d’aider à comprendre l’effet de l’ultra-endurance sur l’organisme et son impact au niveau notamment musculaire. L’équipe de chercheurs du laboratoire CREATIS a choisi de mener cette étude sur l’Ultramarathon de Montagne longtemps considérée comme l’épreuve la plus extrême au monde : le Tor des Géants. Cette étude a notamment utilisé une IRM mobile dotées de techniques d’imagerie avancées pour étudier les variations longitudinales survenant lors d’un effort supra-physiologique. Afin d'étudier l'évolution de l'inflammation et de paramètres fonctionnels musculaires au niveau des quadriceps, notre thèse étudie et améliore les méthodes de segmentation automatique basées sur des approches supervisées. Notre objectif est de fournir une méthode cliniquement applicable permettant de segmenter les chefs musculaires des quadriceps aussi précisément que possible en longitudinal, et nécessitant le moins possible de segmentations manuelles pour la phase d'apprentissage. A cet effet, nous explorons et appliquons dans un premier temps les approches multi-atlas pour la segmentation des quadriceps. Nos contributions permettent d'obtenir des segmentations de qualité sur une grande partie de la base de données. Cependant, afin de disposer d'une approche plus rapide et plus robuste, nous avons secondairement orienté nos travaux vers les approches de deep learning. Dans ce contexte méthodologique, nos deux contributions principales sont i) la proposition d'une étape de correction des segmentations basée sur un apprentissage machine, et ii) la proposition de stratégies d'augmentation de données pour optimiser l'apprentissage de réseaux de type U-Net, notamment une stratégie basée sur la ressemblance morphologique qui est évaluée grâce à une mesure originale. Cette mesure de la morphologie s'est aussi révélée très efficace pour sélectionner les atlas pour l'approche de segmentation multi-atlas. Enfin, nous montrons que les approches proposées se généralisent à d'autres problématiques de segmentation musculaires en IRM et permettent aussi des études statistiques longitudinales et localisées.Notre travail met en évidence que, même si la quantité de données annotées est essentielle dans l'apprentissage supervisé, nous devons également prêter attention à la diversité morphologique de notre base de données, ce qui permet de réduire le temps de calcul et d'augmenter la précision ainsi que la robustesse des méthodes de segmentation.}