Search

Structured AbstractObjectiveTo evaluate accuracy and reproducibility of 2D echocardiography (2DE) left ventricular (LV) volumes and ejection fraction (LVEF) estimates by Deep Learning (DL) vs. manual contouring and against CMR.Background2DE LV manual segmentation for LV volumes and LVEF calculation is time consuming and operator dependent.MethodsA DL-based convolutional network (DL1) was trained on 2DE data from centre A, then evaluated on 171 subjects with a wide range of cardiac conditions (49 healthy) – 31 subjects from centre A (18%) and 140 subjects from centre B (82%) – who underwent 2DE and CMR on the same day. Two senior (A1 and B1) and one junior (A2) cardiologists manually contoured 2DE end-diastolic (ED) and end-systolic (ES) endocardial borders in the cycle and frames of their choice. Selected frames were automatically segmented by DL1 and two DL algorithms from the literature (DL2 and DL3), applied without adaptation to verify their generalizability to unseen data. Interobserver variability of DL was compared to manual contouring. All ESV, EDV and EF values were compared to CMR as reference.Results50% of 2DE images were of good quality. Interobserver agreement was better by DL1 and DL2 than by manual contouring for EF (Lin’s concordance = 0.9 and 0.91 vs. 0.84), EDV (0.98 and 0.99 vs. 0.82), and ESV (0.99 and 0.99 vs. 0.89). LVEF bias was similar or reduced using DL1 (-0.1) vs. manual contouring (3.0), and worse for DL2 and DL3. Agreement between 2DE and CMR LVEF was similar or higher for DL1 vs. manual contouring (Cohen’s kappa = 0.65 vs. 0.61) and degraded for DL2 and DL3 (0.48 and 0.29).ConclusionDL contouring yielded accurate EF measurements and generalized well to unseen data, while reducing interobserver variability. This suggests that DL contouring may improve accuracy and reproducibility of 2DE LVEF in routine practice.

Myocardial tissue tracking imaging techniques have been developed for a more accurate evaluation of myocardial deformation (i.e. strain), with the potential to overcome the limitations of ejection fraction (EF) and to contribute, incremental to EF, to the diagnosis and prognosis in cardiac diseases. While most of the deformation imaging techniques are based on the similar principles of detecting and tracking specific patterns within an image, there are intra- and inter-imaging modality inconsistencies limiting the wide clinical applicability of strain. In this review, we aimed to describe the particularities of the echocardiographic and cardiac magnetic resonance deformation techniques, in order to understand the discrepancies in strain measurement, focusing on the potential sources of variation: related to the software used to analyse the data, to the different physics of image acquisition and the different principles of 2D vs. 3D approaches. As strain measurements are not interchangeable, it is highly desirable to work with validated strain assessment tools, in order to derive information from evidence-based data. There is, however, a lack of solid validation of the current tissue tracking techniques, as only a few of the commercial deformation imaging softwares have been properly investigated. We have, therefore, addressed in this review the neglected issue of suboptimal validation of tissue tracking techniques, in order to advocate for this matter.

Background Cardiovascular magnetic resonance (CMR) 2D feature tracking (FT) left ventricular (LV) myocardial strain has seen widespread use to characterize myocardial deformation. Yet, validation of CMR FT measurements remains scarce, particularly for regional strain. Therefore, we aimed to perform intervendor comparison of 3 different FT software against tagging. Methods In 61 subjects (18 healthy subjects, 18 patients with chronic myocardial infarction, 15 with dilated cardiomyopathy, and 10 with LV hypertrophy due to hypertrophic cardiomyopathy or aortic stenosis) were prospectively compared global (G) and regional transmural peak-systolic Lagrangian longitudinal (LS), circumferential (CS) and radial strains (RS) by 3 FT software (cvi42, Segment, and Tomtec) among each other and with tagging at 3T. We also evaluated the ability of regional LS, CS, and RS by different FT software vs tagging to identify late gadolinium enhancement (LGE) in the 18 infarct patients. Results GLS and GCS by all 3 software had an excellent agreement among each other (ICC = 0.94–0.98 for GLS and ICC = 0.96–0.98 for GCS respectively) and against tagging (ICC = 0.92–0.94 for GLS and ICC = 0.88–0.91 for GCS respectively), while GRS showed inconsistent agreement between vendors (ICC 0.10–0.81). For regional LS, the agreement was good (ICC = 0.68) between 2 vendors but less vs the 3rd (ICC 0.50–0.59) and moderate to poor (ICC 0.44–0.47) between all three FT software and tagging. Also, for regional CS agreement between 2 software was higher (ICC = 0.80) than against the 3rd (ICC = 0.58–0.60), and both better agreed with tagging (ICC = 0.70–0.72) than the 3rd (ICC = 0.57). Regional RS had more variation in the agreement between methods ranging from good (ICC = 0.75) to poor (ICC = 0.05). Finally, the accuracy of scar detection by regional strains differed among the 3 FT software. While the accuracy of regional LS was similar, CS by one software was less accurate (AUC 0.68) than tagging (AUC 0.80, p Conclusions We confirm good agreement of CMR FT and little intervendor difference for GLS and GCS evaluation, with variable agreement for GRS. For regional strain evaluation, intervendor difference was larger, especially for RS, and the diagnostic performance varied more substantially among different vendors for regional strain analysis.

Background: Pulmonary transit time (PTT) from first-pass perfusion imaging is a novel parameter to evaluate hemodynamic congestion by cardiac magnetic resonance (cMR). We sought to evaluate the additional prognostic value of PTT in heart failure with reduced ejection fraction over other well-validated predictors of risk including the Meta-Analysis Global Group in Chronic Heart Failure risk score and ischemic cause. Methods: We prospectively followed 410 patients with chronic heart failure with reduced ejection fraction (61±13 years, left ventricular (LV) ejection fraction 24±7%) who underwent a clinical cMR to assess the prognostic value of PTT for a primary endpoint of overall mortality and secondary composite endpoint of cardiovascular death and heart failure hospitalization. Normal reference values of PTT were evaluated in a population of 40 asymptomatic volunteers free of cardiovascular disease. Results PTT was significantly increased in patients with heart failure with reduced ejection fraction as compared to controls (9±6 beats and 7±2 beats, respectively, P 2 to improve, 12.3; hazard ratio, 1.35 [95% CI, 1.16–1.58]; P 2 to improve=20.1; hazard ratio, 1.23 [95% CI, 1.21–1.60]; P Conclusions: Despite limitations in temporal resolution, PTT derived from first-pass perfusion imaging provides higher and independent prognostic information in heart failure with reduced ejection fraction than clinical and other cMR parameters, including LV and RV ejection fraction or feature tracking global longitudinal strain. Registration: URL: https://www.clinicaltrials.gov ; Unique identifier: NCT03969394.

Funding Acknowledgements Fondation de Recherche Scientifique Belge FRSM PDR 19488731 BACKGROUND 2D-speckle-tracking (ST) echocardiography is currently widely used for estimation of global (G) and regional myocardial deformation. In previous works, we showed good correlation between global longitudinal (LS) and circumferential strain (CS) from one 2DST vendor with cMR-Tagging, however with significant bias between both methods. Also, we found poorer agreement between 2DST and cMR-Tagging on regional basis. However it is unknown how 2DST from other vendors would comparte to cMR tagging. PURPOSE To asssess vendor differences in global and regional strain assessment and compare 1) the agreement of 2 different 2DST softwares for global and regional LS and CS among each other and against cMR-Tagging as reference; and 2) the accuracy of both softwares to detect infarcted segments. METHODS 100 subjects with different cardiac disease (among which 31 with chronic infarct) underwent 2DST and tagging and LGE cMR on the same day. Global and regional CS (16 AHA segments) and LS (18 AHA segments) was computed using 2 different ST vendor softwares and compared to cMR-Tagging with HARP. Accuracy of regional 2D-ST by both vendors to detect infarcted segments (ie >75% transmurality of late gadolinium) was compared using ROC analysis. RESULTS Global LS (ICC = 0.87) and CS 2DST (ICC = 0.83, p Agreement for regional strains is shown in the figure below. Overall, regional LS and CS agreed adequately among both vendors. Agreement of regional LS and CS vs cMR-tagging was slightly better for vendor 1, with less bias than for vendor 2, and disagreement was similarly located (ie agreement with cMR-Tagging for LS in inferolateral inferior and inferoseptal basal segments). The predictive accuracy of regional CS and LS for detecting segments with infarct was higher for vendor 2 (AUC 0.76 and 0.68) than for vendor 1 (AUC 0.70 and 0.63) . CONCLUSION GLS agreed well among both vendors and with cMR-Tagging, confirming the universal validity of this measurement. However vendor 2 provided significantly greater GCS values and had higher bias against cMR-Tagging than vendor 1. On regional basis CS and LS agreed moderately well among both vendors, however vendor 2 agreed less with cMR-Tagging than vendor 1, but astoundingly had higher diagnostic accuracy for detecting infarct. Overall this findings call for further efforts in standardization of 2DST CS and regional strain. Abstract P1393 Figure.

Funding Acknowledgements Philips BACKGROUND Developing new training tools for TransThoracic Echocardiography (TTE) is more important than ever, as the use of ultrasound expands with the advent of mobile devices, both reducing costs and increasing the number of sites and operators. Two major challenges are the lack of experts to meet the growing demand for training and the risk of unnecessary examinations and misdiagnosis by users who lack proper training. PURPOSE To evaluate Artificial Intelligence (AI)-assisted TTE for assessing and improving novices" echocardiography skills. METHODS AI-assisted TTE relies on real-time analysis of the ultrasound stream by AI algorithms (e.g. for automated view recognition) to provide adaptive feedback to the user. It was compared to standard TTE in a prospective study including 40 medical students with no prior ultrasound experience ("novices") and 40 healthy volunteers of varying echogenicity. Novices received a standardized 10-minute presentation of the basic machine controls and requirements of an apical 4-chamber (AP4) acquisition. They were then asked to perform three consecutive AP4 acquisitions on a randomly assigned healthy volunteer: 1) "initial" standard TTE, 2) AI-assisted TTE and 3) "repetition" standard TTE, with a maximum of 3 minutes per acquisition. Additionally, for reference purposes, an AP4 acquisition was performed by an expert. Both the performance over time and the final acquisition performance were assessed by an AI-based AP4 quality score and expressed relatively to the reference score. The suitability for clinical use was assessed by an expert on a 0-3 semi-quantitative scale. RESULTS Kaplan-Meier analysis of the probability to match reference performance over time (see left part in figure below) validated the positive and significant contribution of AI-assisted TTE wrt initial TTE (median time = 16.1s vs. 45.7s, p = 0.026). There was also a positive, though non-significant, contribution wrt repetition TTE (median time = 25.7s, p = 0.20), while the difference between standard TTEs was non-significant (p = 0.24). A majority of novices (70%, resp. 72.5%) improved their final acquisition with AI-assisted TTE wrt initial TTE, resp. repetition TTE. The final performance gain (81.7 ± 29.7% vs. 60.8 ± 40.4%, resp. 70.4 ± 33.9%, see right part in figure below) was shown to be significant by Wilcoxon matched-pair signed-rank test (p CONCLUSION AI-assisted TTE can assess and improve novices" echocardiography skills. Developing and embedding AI-assistance in ultrasound devices could be a cost-effective way to support the training of novices and key to acquisition standardization. Abstract 541 Figure.

Funding Acknowledgements Philips BACKGROUND Accurate and reproducible echocardiographic measurements are paramount for objective assessment and follow-up of the cardiac function. However, manual contouring – e.g., for determining left ventricular (LV) volumes and ejection fraction (EF) – is limited by image quality and operator experience. Meanwhile, despite the wider availability of (semi-)automated tools, strong multimodal validation is still lacking for their widespread and safe use in the clinical routine. PURPOSE To evaluate the accuracy and reproducibility of an Artificial Intelligence (AI)-based semi-automated tool to compute LV volumes and EF, in comparison with manual contouring, using cardiac magnetic resonance (cMR) as reference. METHODS Manual and AI measurements from echocardiography were compared to measurements from cMR in a retrospective two-centre study. One hundred fourteen patients in sinus rhythm were included; among those, 85 had abnormal LV function (56 dilated and 29 hypertrophic). Three successive cardiac cycles were available for apical 4- and 2-chamber views. Two senior (A1 and B1) and one junior (A2) cardiologists contoured the ED and ES endocardial borders in the cardiac cycle of their choice, while blinded to quantitative outcomes. For AI analysis, a deep convolutional neural networks model was used to segment the LV cavity on the frames selected by the three observers. This model was trained using ED and ES manual contouring from senior cardiologist A1 on an independent single-centre dataset that consisted of 700 apical 4- and 2-chamber views. The same biplane Simpson’s method was used to compute all LV volumes and EF. RESULTS Despite challenging image quality (poor: 6%; fair: 33%; high: 61%, as rated by observers), the majority of the AI segmentations were deemed acceptable (75% in total; 80% for images of high quality). Overall, inter-observer agreement was better by AI than by manual contouring (ICC = 0.99 vs. 0.89, 1.00 vs. 0.95 and 0.95 vs. 0.89 for LVED, LVES and LVEF respectively, all p CONCLUSION The AI model generalized well to different sites, observers and image quality. Compared to manual contouring, LV volumes and EF by AI showed comparable or improved accuracy and higher reproducibility. These findings demonstrate the value of AI-based tools, with potential for full automation, for objective assessment and follow-up of the cardiac function. Abstract 154 Figure.

Background Assessment of congestion and cardiac function has been shown to have both therapeutic and prognostic implication for the management of patient with CHF. Pulmonary transit time (PTT) assessed by cMR is a novel parameter, which reflects not only hemodynamic congestion but also LV and RV function. Purpose We sought to explore the prognostic value of the pulmonary transit time assessed in seconds (PTT) and in beats (PTB) and the pulmonary blood volume indexed (PBVi) above conventional well-known risk factors including cMR-RVEF and estimated pulmonary artery pressure (eSPAP) in predicting outcomes. PBVi is defined by the product of PTB and the stroke volume indexed to body surface area. Methods 401 patients in sinus rhythm with a LVEF Results Average cMR-LVEF was 23±7%, cMR-RVEF was 43±15%, average estimated systolic pulmonary pressure (eSPAP) was 33±12mmH, average PTT was 11±6s, PTB 8.9±5.6 bpm and average PBVi 305.5±254.9ml/m2. After a median follow-up of 6 years, 182 reached the primary endpoint. In univariate cox regression, age, ischemic cardiomyopathy, hypertension, diabetes, NYHA class III-IV, eSPAP >40mmHg, E/A ratio, e/e'ratio, cMR-RVEF, LV scar, PTT, PTB, PBVi, GFR, beta blockers and diuretics were associated with overall mortality. For the multivariate analysis, a baseline model was created where age, ischemic etiology, NYHA functional class III-IV, eSPAP >40 mmHg, beta-blockers and cMR-RVEF were found to be significantly and independently associated with the primary endpoint. PTT (X2 to improve = 5.3, HR: 1.03; 95% CI: [1.01; 1.06]; P=0.015), PTB (X2 to improve = 11.8, HR: 1.06; 95% CI: [1.03; 1.09]; P Conclusion Pulmonary transit time and pulmonary blood volume provide higher prognostic information over well-known risk factors including cMR-RVEF and eSPAP with high power to stratify prognosis in HF-rEF and might be promising tools to identify patients at higher risk among HF patients. Acknowledgement/Funding Fond National de recherche scientifique (FNRS)

Clinical decision requires reasoning in the presence of imperfect data. DTs are a well-known decision support tool, owing to their interpretability, fundamental in safety-critical contexts such as medical diagnosis. However, learning DTs from uncertain data leads to poor generalization, and generating predictions for uncertain data hinders prediction accuracy. Several methods have suggested the potential of probabilistic decisions at the internal nodes in making DTs robust to uncertainty. Some approaches only employ probabilistic thresholds during evaluation. Others also consider the uncertainty in the learning phase, at the expense of increased computational complexity or reduced interpretability. The existing methods have not clarified the merit of a probabilistic approach in the distinct phases of DT learning, nor when the uncertainty is present in the training or the test data. We present a probabilistic DT approach that models measurement uncertainty as a noise distribution, independently realized: (1) when searching for the split thresholds, (2) when splitting the training instances, and (3) when generating predictions for unseen data. The soft training approaches (1, 2) achieved a regularizing effect, leading to significant reductions in DT size, while maintaining accuracy, for increased noise. Soft evaluation (3) showed no benefit in handling noise.

BACKGROUND: The aim of this study was to evaluate the accuracy of mitral regurgitation (MR) volume quantified on three-dimensional (3D) color Doppler transesophageal echocardiography (TEE) using new semiautomated software compared with conventional two-dimensional (2D) proximal isovelocity surface area (PISA) transthoracic echocardiography (TTE) and TEE and cardiac magnetic resonance imaging (CMR). METHODS: Fifty-one patients (mean age, 63 ± 16 years; 35 men) prospectively underwent TTE, TEE, and CMR for MR evaluation. Regurgitant volume (RVol) by 3D MR flow quantification was compared with 2D TTE, TEE, and CMR, and the accuracy of evaluation of severe MR by 3D MR flow quantification was compared against guideline criteria by TEE. RESULTS: Twenty-nine patients had severe MR, 16 had moderate MR, and six had mild MR. Three-dimensional MR flow quantification was feasible in all patients, including prolapse (n = 37), restriction (n = 9), functional MR (n = 5), and eccentric or multiple jects (n = 41). RVol on 3D MR flow quantification correlated well with RVol on 2D PISA TTE (interclass correlation coefficient [ICC] = 0.75, P < .001), quantitatively estimated RVol (ICC = 0.74, P < .001), and 2D PISA TEE (ICC = 0.79, P < .001). Three-dimensional MR flow quantification agreed better with CMR (ICC = 0.86, P < .001) than did RVol on 2D PISA TTE (ICC = 0.66, P < .001) and 2D PISA TEE (ICC = 0.69, P < .001), with narrower limits of agreement on Bland-Altman analysis. Three-dimensional MR flow quantification had high accuracy for diagnosing severe MR using TEE (area under the curve = 0.85, 95% CI 0.74-0.96, P < .001) or CMR (area under the curve = 0.95; 95% CI, 0.89-1.00; P < .001) as the criterion. CONCLUSIONS: The new software enabled semiautomated 3D MR flow quantification in complex MR with multiple and eccentric jets and showed better agreement with CMR than 2D PISA TTE or TEE, suggesting that this method is more accurate than conventional 2D PISA TTE and TEE.

Objective: The aim of this paper is to describe an automated diagnostic pipeline which uses as input only ultrasound (US) data, but is at the same time informed by a training database of multimodal magnetic resonance (MR) and US image data.Methods: We create a multimodal cardiac motion atlas from 3D MR and 3D US data followed by multi-view machine learning algorithms to combine and extract the most meaningful cardiac descriptors for classification of dilated cardiomyopathy (DCM) patients using US data only. More specifically, we propose two algorithms based on: multi-view linear discriminant analysis (MLDA) and multi-view Laplacian support vector machines (MvLapSVM). Furthermore, a novel regional multi-view approach is proposed to exploit the regional relationships between the two modalities.Results: We evaluate our pipeline on the classification task of discriminating between normals and DCM patients. Results show that the use of multi-view classifiers together with a cardiac motion atlas results in a statistically significant improvement in accuracy compared to classification without the multimodal atlas. MvLapSVM was able to achieve the highest accuracy for both the globalapproach (92.71%) and the regional approach (94.32%).Conclusion: Our work represents an important contribution to the understandingof cardiac motion, which is an important aid in the quantification of the contractility and function of the left ventricular myocardium.Significance: The intended workflow of the developed pipeline is to make use of the prior knowledge from the multimodal atlas to enable robust extraction of indicators from 3D US images for detecting DCM patients.

Cardiovascular magnetic resonance myocardial feature tracking (CMR-FT) is a promising method for quantification of cardiac function from standard steady-state free precession (SSFP) images. However, currently available techniques require operator dependent and time-consuming manual intervention, limiting reproducibility and clinical use. In this paper, we propose a fully automated pipeline to compute left ventricular (LV) longitudinal and radial strain from 2- and 4-chamber cine acquisitions, and LV circumferential and radial strain from the short-axis imaging. The method employs a convolutional neural network to automatically segment the myocardium, followed by feature tracking and strain estimation. Experiments are performed using 40 healthy volunteers and 40 ischemic patients from the UK Biobank dataset. Results show that our method obtained strain values that were in excellent agreement with the commercially available clinical CMR-FT software CVI42 (Circle Cardiovascular Imaging, Calgary, Canada).

BACKGROUND: In prior work, the authors demonstrated that two-dimensional speckle-tracking (2DST) correlated well but systematically overestimated global longitudinal strain (LS) and circumferential strain (CS) compared with two-dimensional cardiac magnetic resonance tagging (2DTagg) and had poor agreement on a segmental basis. Because three-dimensional speckle-tracking (3DST) has recently emerged as a new, more comprehensive evaluation of myocardial deformation, this study was undertaken to evaluate whether it would compare more favorably with 2DTagg than 2DST. METHODS: In a prospective two-center trial, 119 subjects (29 healthy volunteers, 63 patients with left ventricular dysfunction, and 27 patients with left ventricular hypertrophy) underwent 2DST, 3DST, and 2DTagg. Global, regional (basal, mid, and apical), and segmental (18 and 16 segments per patient) LS and CS by 2DST and 3DST were compared with 2DTagg using intraclass correlation coefficients (ICCs) and Bland-Altman analysis. Test-retest reproducibility of 3DST and 2DST was compared in 48 other patients. RESULTS: Both global LS and CS by 3DST agreed better with 2DTagg (ICC = 0.89 and ICC = 0.83, P < .001 for both; bias = 0.5 ± 2.3% and 0.2 ± 3%) than 2DST (ICC = 0.65 and ICC = 0.55, P < .001 for both; bias = -5.5 ± 2.5% and -7 ± 5.3%). Unlike 2DST, 3DST did not overestimate deformation at the regional and particularly the apical levels and at the segmental level had lower bias (LS, 0.8 ± 2.8% vs -5.3 ± 2.4%; CS, -0.01 ± 2.8% vs -7 ± 2.8%, respectively) but similar agreement with 2DST (LS: ICC = 0.58 ± 0.16 vs 0.56 ± 0.12; CS: ICC = 0.58 ± 0.12 vs 0.51 ± 0.1) with 2DTagg. Finally, 3DST had similar global LS, but better global CS test-retest variability than 2DST. CONCLUSIONS: Using 2DTagg as reference, 3DST had better agreement and less bias for global and regional LS and CS. At the segmental level, 3DST demonstrated comparable agreement but lower bias versus 2DTagg compared with 2DST. Also, test-retest variability for global CS by 3DST was better than by 2DST. This suggests that 3DST is superior to 2DST for analysis of global and regional myocardial deformation, but further refinement is needed for both 3DST and 2DST at the segmental level.

Comunicació presentada a: 17th IEEE International Conference on Machine Learning and Applications (ICMLA) celebrada del 17 al 20 de 2018 a Orlando, Estats Units. Decision trees (DTs) are a widely used prediction tool, owing to their interpretability. Standard learning methods follow a locally-optimal approach that trades off prediction performance for computational efficiency. Such methods can however be far from optimal, and it may pay off to spend more computational resources to increase performance. Monte Carlo tree search (MCTS) is an approach to approximate optimal choices in exponentially large search spaces. Since exploring the space of all possible DTs is computationally intractable, we propose a DT learning approach based on MCTS. To bound the branching factor of MCTS, we limit the number of decisions at each level of the search tree, and introduce mechanisms to balance exploration, DT size and the statistical significance of the predictions. To mitigate the computational cost of our method, we employ a move pruning strategy that discards some branches of the search tree, leading to improved performance. The experiments show that our approach outperformed locally optimal search in 20 out of 31 datasets, with a reduction in DT size in most of the cases. This work is supported by the European Union Horizon 2020 Programme for Research and Innovation, under grant agreement No. 642676 (CardioFunXion).

A cardiac motion atlas provides a space of reference in which the cardiac motion fields of a cohort of subjects can be directly compared. From such atlases, descriptors can be learned for subsequent diagnosis and characterization of disease. Traditionally, such atlases have been formed from imaging data acquired using a single modality. In this work we propose a framework for building a multimodal cardiac motion atlas from MR and ultrasound data and incorporate a multiview classifier to exploit the complementary information provided by the two modalities. We demonstrate that our novel framework is able to detect non ischemic dilated cardiomyopathy patients from ultrasound data alone, whilst still exploiting the MR based information from the multimodal atlas. We evaluate two different approaches based on multiview learning to implement the classifier and achieve an improvement in classification performance from 77.5% to 83.50% compared to the use of US data without the multimodal atlas.

Purpose: This paper addresses the reconstruction of x-ray cone-beam computed tomography (CBCT) for interventional C-arm systems. Subsampling of CBCT is a significant issue with C-arms due to their slow rotation and to the low frame rate of their flat panel x-ray detectors. The aim of this work is to propose a novel method able to handle the subsampling artifacts generally observed with analytical reconstruction, through a content-driven hierarchical reconstruction based on compressed sensing. Methods: The central idea is to proceed with a hierarchical method where the most salient features (high intensities or gradients) are reconstructed first to reduce the artifacts these features induce. These artifacts are addressed first because their presence contaminates less salient features. Several hierarchical schemes aiming at streak artifacts reduction are introduced for C-arm CBCT: the empirical orthogonal matching pursuit approach with the l0 pseudonorm for reconstructing sparse vessels; a convex variant using homotopy with the l1-norm constraint of compressed sensing, for reconstructing sparse vessels over a nonsparse background; homotopy with total variation (TV); and a novel empirical extension to nonlinear diffusion (NLD). Such principles are implemented with penalized iterative filtered backprojection algorithms. For soft-tissue imaging, the authors compare the use of TV and NLD filters as sparsity constraints, both optimized with the alternating direction method of multipliers, using a threshold for TV and a nonlinear weighting for NLD. Results: The authors show on simulated data that their approach provides fast convergence to good approximations of the solution of the TV-constrained minimization problem introduced by the compressed sensing theory. Using C-arm CBCT clinical data, the authors show that both TV and NLD can deliver improved image quality by reducing streaks. Conclusions: A flexible compressed-sensing-based algorithmic approach is proposed that is able to accommodate for a wide range of constraints. It is successfully applied to C-arm CBCT images that may not be so well approximated by piecewise constant functions.

Background— Despite widespread use to characterize and refine prognosis, validation data of two-dimensional (2D) speckle tracking (2DST) echocardiography myocardial strain measurement remain scarce. Methods and Results— Global and regional subendocardial peak-systolic Lagrangian longitudinal (LS) and circumferential strain (CS) by 2DST and 2D-tagged (2DTagg) cardiac magnetic resonance imaging were compared against sonomicrometry in a dynamic heart phantom and among each other in 136 patients included prospectively at 2 centers. The ability of regional LS and CS 2DST and 2DTagg to identify late gadolinium enhancement was compared using receiver operating characteristics curves. In vitro, both LS-2DST and 2DTagg highly agreed with sonomicrometry (intraclass correlation coefficient [ICC], 0.89 and ICC, 0.90, both P P P =0.08), while it was lower for CS 2DST than 2DTagg (area under the curve, 0.61 versus 0.75; P Conclusions— The high accuracy against sonomicrometry and good agreement of global LS and global CS by 2DST and 2DTagg confirm the overall validity of 2DST strain measurement. Yet, higher intertechnique segmental variability and lower ability for detecting infarct suggest that 2DST strain estimates may be less performant on regional than on global basis.

In this paper, we address the problem of automated pose classification and segmentation of the left ventricle (LV) in 2D echocardiographic images. For this purpose, we compare two complementary approaches. The first one is based on engineering ad-hoc features according to the traditional machine learning paradigm. Namely, we extract phase features to build an unsupervised LV pose estimator, as well as a global image descriptor for view type classification. We also apply the Supervised Descent Method (SDM) to iteratively refine the LV contour. The second approach follows the deep learning framework, where a Convolutional Network (ConvNet) learns the visual features automatically. Our experiments on a large database of apical sequences show that the two approaches yield comparable results on view classification, but SDM outperforms ConvNet on LV segmentation at a significantly lower training computational cost.

International audience; This work tackles three-dimensional reconstruction of tomographic acquisitions in C-arm-based rotational angiography. The relatively slow rotation speed of C-arm systems involves motion artifacts that limit the use of three-dimensional imaging in interventional procedures. The main contribution of this paper is a reconstruction algorithm that deals with the temporal variations due to intra-arterial injections. Based on a compressed-sensing approach, we propose a multiple phase reconstruction with spatio-temporal constraints. The algorithm was evaluated by qualitative and quantitative assessment of image quality on both numerical phantom experiments and clinical data from vascular C-arm systems. In this latter case, motion artifacts reduction was obtained in spite of the cone-beam geometry, the short-scan acquisition, and the truncated and subsampled data.

International audience; Digital Subtraction Rotational Angiography (DSRA) is a clinical protocol that allows three-dimensional (3D) visualization of vasculature during minimally invasive procedures. C-arm systems that are used to generate 3D reconstructions in interventional radiology have limited sampling rate and thus, contrast resolution. To address this particular subsampling problem, we propose a novel iterative reconstruction algorithm based on compressed sensing. To this purpose, we exploit both spatial and temporal sparsity of DSRA. For computational efficiency, we use a proximal implementation that accommodates multiple '1-penalties. Experiments on both simulated and clinical data confirm the relevance of our strategy for reducing subsampling streak artifacts.

International audience; In this paper, we address three-dimensional tomographic re-construction of rotational angiography acquisitions. In clinical routine, angular subsampling commonly occurs, due to the technical limitations of C-arm systems or possible improper injection. Standard methods such as ltered backprojection yield a reconstruction that is deteriorated by subsampling artifacts, which potentially hampers medical interpretation. Recent developments of compressed sensing have demonstrated that it is possible to signi cantly improve reconstruction of subsampled datasets by generating sparse approximations through '1-penalized minimization. Based on these results, we present an extension of the iterative ltered backprojection that includes a sparsity constraint called soft background subtraction. This approach is shown to provide subsampling artifact reduction when reconstructing sparse objects, and more interestingly, when reconstructing sparse objects over a non-sparse background. The relevance of our approach is evaluated in cone beam geometry on real clinical data.

International audience; Interpretation of medical infrared images is complex due to thermal noise, absence of texture, and small temperature differences in pathological zones. Acute inflammatory response is a characteristic symptom of some knee injuries like anterior cruciate ligament sprains, muscle or tendons strains, and meniscus tear. Whereas artificial coloring of the original grey level images may allow to visually assess the extent inflammation in the area, their automated segmentation remains a challenging problem. This paper presents a hybrid segmentation algorithm to evaluate the extent of inflammation after knee injury, in terms of temperature variations and surface shape. It is based on the intersection of rapid color segmentation and homogeneous region segmentation, to which a Laplacian of a Gaussian filter is applied. While rapid color segmentation enables to properly detect the observed core of swollen area, homogeneous region segmentation identifies possible inflammation zones, combining homogeneous grey level and hue area segmentation. The hybrid segmentation algorithm compares the potential inflammation regions partially detected by each method to identify overlapping areas. Noise filtering and edge segmentation are then applied to common zones in order to segment the swelling surfaces of the injury. Experimental results on images of a patient with anterior cruciate ligament sprain show the improved performance of the hybrid algorithm with respect to its separated components. The main contribution of this work is a meaningful automatic segmentation of abnormal skin temperature variations on infrared thermography images of knee injury swelling.

Comunicació presentada a: 9th international conference on Functional Imaging and Modeling of the Heart (FIMH 2017), celebrada de l'11 al 13 de juny de 2017 a Toronto, Canada. In this paper, we capture patterns of response to cardiac stress-testing using a multiview dimensionality reduction technique that allows the compact representation of patient response to stress, regarding multiple features over consecutive cycles, as a low-dimensional trajectory. In this low-dimensional space, patients can be compared and clustered in distinct healthy and pathological responses, and the patterns that characterize each of them can be reconstructed. Experiments were performed on (a) synthetic data simulating different types of response and (b) a real acquisition during a cold pressor test. Results show that the proposed approach allows the clustering of healthy and pathological responses, as well as the reconstruction of characteristic patterns of such responses, in terms of multiple features of interest. This work is supported by the European Union Horizon 2020 Programme for Research and Innovation, under grant agreement No. 642676 (CardioFunXion).