The emergence of deep learning has considerably advanced the state-of-the-art in cardiac magnetic resonance (CMR) segmentation. Many techniques have been proposed over the last few years, bringing the accuracy of automated segmentation close to human performance. However, these models have been all too often trained and validated using cardiac imaging samples from single clinical centres or homogeneous imaging protocols. This has prevented the development and validation of models that are generalizable across different clinical centres, imaging conditions or scanner vendors. To promote further research and scientific benchmarking in the field of generalizable deep learning for cardiac segmentation, this paper presents the results of the Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation (M&Ms) Challenge, which was recently organized as part of the MICCAI 2020 Conference. A total of 14 teams submitted different solutions to the problem, combining various baseline models, data augmentation strategies, and domain adaptation techniques. The obtained results indicate the importance of intensity-driven data augmentation, as well as the need for further research to improve generalizability towards unseen scanner vendors or new imaging protocols. Furthermore, we present a new resource of 375 heterogeneous CMR datasets acquired by using four different scanner vendors in six hospitals and three different countries (Spain, Canada and Germany), which we provide as open-access for the community to enable future research in the field.

Background: Coronary microvascular dysfunction (MVD) is defined by impaired flow augmentation in response to a pharmacological vasodilator in the presence of nonobstructive coronary artery disease. It is unknown whether diminished coronary vasodilator response correlates with abnormal exercise physiology or inducible myocardial ischemia. Methods: Patients with angina and nonobstructive coronary artery disease had simultaneous coronary pressure and flow velocity measured using a dual sensor-tipped guidewire during rest, supine bicycle exercise, and adenosine-mediated hyperemia. Microvascular resistance (MR) was calculated as coronary pressure divided by flow velocity. Wave intensity analysis quantified the proportion of accelerating wave energy (perfusion efficiency). Global myocardial blood flow and subendocardial:subepicardial perfusion ratio were quantified using 3-Tesla cardiac magnetic resonance imaging during hyperemia and rest; inducible ischemia was defined as hyperemic subendocardial:subepicardial perfusion ratio <1.0. Patients were classified as having MVD if coronary flow reserve <2.5 and controls if coronary flow reserve ≥2.5, with researchers blinded to the classification. Results: Eighty-five patients were enrolled (78% female, 57±10 years), 45 (53%) were classified as having MVD. Of the MVD group, 82% had inducible ischemia compared with 22% of controls ( P <0.001); global myocardial perfusion reserve was 2.01±0.41 and 2.68±0.49 ( P <0.001). In controls, coronary perfusion efficiency improved from rest to exercise and was unchanged during hyperemia (59±11% vs 65±14% vs 57±18%; P =0.02 and P =0.14). In contrast, perfusion efficiency decreased during both forms of stress in MVD (61±12 vs 44±10 vs 42±11%; both P <0.001). Among patients with a coronary flow reserve <2.5, 62% had functional MVD, with normal minimal MR (hyperemic MR<2.5 mmHg/cm/s), and 38% had structural MVD with elevated hyperemic MR. Resting MR was lower in those with functional MVD (4.2±1.0 mmHg/cm/s) than in those with structural MVD (6.9±1.7 mmHg/cm/s) or controls (7.3±2.2 mmHg/cm/s; both P <0.001). During exercise, the structural group had a higher systolic blood pressure (188±25 mmHg) than did those with functional MVD (161±27 mmHg; P =0.004) and controls (156±30 mmHg; P <0.001). Functional and structural MVD had similar stress myocardial perfusion and exercise perfusion efficiency values. Conclusion: In patients with angina and nonobstructive coronary artery disease, diminished coronary flow reserve characterizes a cohort with inducible ischemia and a maladaptive physiological response to exercise. We have identified 2 endotypes of MVD with distinctive systemic vascular responses to exercise; whether endotypes have a different prognosis or require different treatments merits further investigation.

• It is feasible to perform tracer-kinetic modelling and parameter inference with physics-informed neural networks. • This framework gives a high degree of flexibility for improving the parameter estimation performance. • The models learn purely from the observed data and are constrained by the physics of the problem, avoiding the need for a large training set.

Background: Quantitative myocardial perfusion cardiac MRI can provide a fast and robust assessment of myocardial perfusion status for the noninvasive diagnosis of myocardial ischemia while being more objective than visual assessment. However, it currently has limited use in clinical practice due to the challenging postprocessing required, particularly the segmentation.Purpose: To evaluate the efficacy of an automated deep learning (DL) pipeline for image processing prior to quantitative analysis.Study Type: Retrospective.Population: In all, 175 (350 MRI scans; 1050 image series) clinical patients under both rest and stress conditions (135/10/30 training/validation/test).Field Strength/Sequence: 3.0T/2D multislice saturation recovery T₁‐weighted gradient echo sequence.Assessment: Accuracy was assessed, as compared to the manual operator, through the mean square error of the distance between landmarks and the Dice similarity coefficient of the segmentation and bounding box detection. Quantitative perfusion maps obtained using the automated DL‐based processing were compared to the results obtained with the manually processed images.Statistical Tests: Bland–Altman plots and intraclass correlation coefficient (ICC) were used to assess the myocardial blood flow (MBF) obtained using the automated DL pipeline, as compared to values obtained by a manual operator.Results: The mean (SD) error in the detection of the time of peak signal enhancement in the left ventricle was 1.49 (1.4) timeframes. The mean (SD) Dice similarity coefficients for the bounding box and myocardial segmentation were 0.93 (0.03) and 0.80 (0.06), respectively. The mean (SD) error in the RV insertion point was 2.8 (1.8) mm. The Bland–Altman plots showed a bias of 2.6% of the mean MBF between the automated and manually processed MBF values on a per‐myocardial segment basis. The ICC was 0.89, 95% confidence interval = [0.87, 0.90].Data Conclusion: We showed high accuracy, compared to manual processing, for the DL‐based processing of myocardial perfusion data leading to quantitative values that are similar to those achieved with manual processing.Level of Evidence: 3Technical Efficacy Stage: 1J. Magn. Reson. Imaging 2020;51:1689–1696.

Clinical evaluation of stress perfusion cardiovascular magnetic resonance (CMR) is currently based on visual assessment and has shown high diagnostic accuracy in previous clinical trials, when performed by expert readers or core laboratories. However, these results may not be generalizable to clinical practice, particularly when less experienced readers are concerned. Other factors, such as the level of training, the extent of ischemia, and image quality could affect the diagnostic accuracy. Moreover, the role of rest images has not been clarified. The aim of this study was to assess the diagnostic accuracy of visual assessment for operators with different levels of training and the additional value of rest perfusion imaging, and to compare visual assessment and automated quantitative analysis in the assessment of coronary artery disease (CAD). We evaluated 53 patients with known or suspected CAD referred for stress-perfusion CMR. Nine operators (equally divided in 3 levels of competency) blindly reviewed each case twice with a 2-week interval, in a randomised order, with and without rest images. Semi-automated Fermi deconvolution was used for quantitative analysis and estimation of myocardial perfusion reserve as the ratio of stress to rest perfusion estimates. Level-3 operators correctly identified significant CAD in 83.6% of the cases. This percentage dropped to 65.7% for Level-2 operators and to 55.7% for Level-1 operators (p < 0.001). Quantitative analysis correctly identified CAD in 86.3% of the cases and was non-inferior to expert readers (p = 0.56). When rest images were available, a significantly higher level of confidence was reported (p = 0.022), but no significant differences in diagnostic accuracy were measured (p = 0.34). Our study demonstrates that the level of training is the main determinant of the diagnostic accuracy in the identification of CAD. Level-3 operators performed at levels comparable with the results from clinical trials. Rest images did not significantly improve diagnostic accuracy, but contributed to higher confidence in the results. Automated quantitative analysis performed similarly to level-3 operators. This is of increasing relevance as recent technical advances in image reconstruction and analysis techniques are likely to permit the clinical translation of robust and fully automated quantitative analysis into routine clinical practice.

Background: Among patients with angina and nonobstructive coronary artery disease, those with coronary microvascular dysfunction have a poor outcome. Coronary microvascular dysfunction is usually diagnosed by assessing flow reserve with an endothelium-independent vasodilator like adenosine, but the optimal diagnostic threshold is unclear. Furthermore, the incremental value of testing endothelial function has never been assessed before. We sought to determine what pharmacological thresholds correspond to exercise pathophysiology and myocardial ischemia in patients with coronary microvascular dysfunction. Methods: Patients with angina and nonobstructive coronary artery disease underwent simultaneous acquisition of coronary pressure and flow during rest, supine bicycle exercise, and pharmacological vasodilatation with adenosine and acetylcholine. Adenosine and acetylcholine coronary flow reserve were calculated as vasodilator/resting coronary blood flow (CFR and AchFR, respectively). Coronary wave intensity analysis was used to quantify the proportion of accelerating wave energy; a normal exercise response was defined as an increase in accelerating wave energy from rest to peak exercise. Ischemia was assessed by quantitative 3-Tesla stress perfusion cardiac magnetic resonance imaging and dichotomously defined by a hyperemic endo-epicardial gradient <1.0. Results: Ninety patients were enrolled (58±10 years, 77% female). Area under the curve using receiver-operating characteristic analysis demonstrated optimal CFR and AchFR thresholds for identifying exercise pathophysiology and ischemia as 2.6 and 1.5, with positive and negative predictive values of 91% and 86%, respectively. Fifty-eight percent had an abnormal CFR (of which 96% also had an abnormal AchFR). Of those with a normal CFR, 53% had an abnormal AchFR, and 47% had a normal AchFR; ischemia rates were 83%, 63%, and 14%, respectively. Conclusions: The optimal CFR and AchFR diagnostic thresholds are 2.6 and 1.5, with high-positive and negative predictive values, respectively. A normal CFR value should prompt the measurement of AchFR. A stepwise algorithm incorporating both vasodilators can accurately identify an ischemic cause in patients with nonobstructive coronary artery disease.