Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Guala A. received funding from the Spanish Ministry of Science, Innovation and Universities Background Phase-contrast (PC) enhanced magnetic resonance (MR) angiography (MRA) is a class of angiogram exploiting velocity data to increase the signal-to-noise ratio, thus avoiding the administration of external contrast agent, normally used to segment 4D flow MR data. To train deep-learning algorithms to segment PC-MRA a large amount of manually annotated data is needed: however, the relatively novelty of the sequence, its rapid evolution and the extensive time needed to manually segment data limit its availability. Purpose The aim of this study was to test a deep learning algorithm in the segmentation of multi-center and multi-vendor PC-MRA and to test if transfer learning (TL) improves performance. Methods A large dataset (LD) of 262 and a small one (SD) of 22 PC-MRA, acquired without contrast agent at 1.5 T in a General Electric and a Siemens scanner, respectively, were manually annotated and divided into training (232 and 15 cases) and testing (30 and 7) sets. They both included PC-MRA of healthy subjects and patients with aortic diseases (excluding dissections) and native aorta. A convolutional neural networks (CNN) based on nnU-Net framework [1] was trained in the LD and another in the SD. The left ventricle was removed semi-automatically from the DL segmentations of the LD as it was not relevant for this application. Networks were then tested on the test sets of the dataset there were trained and the other dataset to assess generalizability. Finally, a fine-tuning transfer learning approach was applied to LD network and the performance on both test sets were tested. Dice score, Hausdorff distance, Jaccard score and Average Symmetrical Surface Distance were used as segmentation quality metrics. Results LD network achieved good performance in LD test set, with a DS of 0.904, ASSD of 1.47, J of 0.827 and HD of 6.35, which further improve after removing the left ventricle in the post-processing to a DS of 0.942, ASSD of 0.93, J of 0.892 and HD of 3.32. SD network results in an average DS of 0.895, ASSD of 0.59, J of 0.812 and HD of 2.05. Once tested on the testing set of the other dataset, LD network resulted in a DS of 0.612 while SD network in DS of 0.375, thus showing limited generalizability. However, the application of transfer learning to LD network resulted in the improvement of the evaluation metrics on the SD from a DS of 0.612 to 0.858, while slightly worsening in the first one without post-processing to 0.882. Conclusions nnU-net framework is effective for fast automatic segmentation of the aorta from multi-center and multi-vendor PC-MRA, showing performance comparable with the state of the art. The application of transfer learning allows for increased generalization to data from center not included in the original training. These results unlock the possibility for fully-automatic analysis of multi-vendor multi-center 4D flow MR.

No abstract

Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Heart Foundation (NHF) of New Zealand Health Research Council (HRC) of New Zealand Artificial intelligence shows considerable promise for automated analysis and interpretation of medical images, particularly in the domain of cardiovascular imaging. While application to cardiac magnetic resonance (CMR) has demonstrated excellent results, automated analysis of 3D echocardiography (3D-echo) remains challenging, due to the lower signal-to-noise ratio (SNR), signal dropout, and greater interobserver variability in manual annotations. As 3D-echo is becoming increasingly widespread, robust analysis methods will substantially benefit patient evaluation.  We sought to leverage the high SNR of CMR to provide training data for a convolutional neural network (CNN) capable of analysing 3D-echo. We imaged 73 participants (53 healthy volunteers, 20 patients with non-ischaemic cardiac disease) under both CMR and 3D-echo (<1 hour between scans). 3D models of the left ventricle (LV) were independently constructed from CMR and 3D-echo, and used to spatially align the image volumes using least squares fitting to a cardiac template. The resultant transformation was used to map the CMR mesh to the 3D-echo image. Alignment of mesh and image was verified through volume slicing and visual inspection (Fig. 1) for 120 paired datasets (including 47 rescans) each at end-diastole and end-systole. 100 datasets (80 for training, 20 for validation) were used to train a shallow CNN for mesh extraction from 3D-echo, optimised with a composite loss function consisting of normalised Euclidian distance (for 290 mesh points) and volume. Data augmentation was applied in the form of rotations and tilts (<15 degrees) about the long axis. The network was tested on the remaining 20 datasets (different participants) of varying image quality (Tab. I). For comparison, corresponding LV measurements from conventional manual analysis of 3D-echo and associated interobserver variability (for two observers) were also estimated. Initial results indicate that the use of embedded CMR meshes as training data for 3D-echo analysis is a promising alternative to manual analysis, with improved accuracy and precision compared with conventional methods. Further optimisations and a larger dataset are expected to improve network performance. (n = 20) LV EDV (ml) LV ESV (ml) LV EF (%) LV mass (g) Ground truth CMR 150.5 ± 29.5 57.9 ± 12.7 61.5 ± 3.4 128.1 ± 29.8 Algorithm error -13.3 ± 15.7 -1.4 ± 7.6 -2.8 ± 5.5 0.1 ± 20.9 Manual error -30.1 ± 21.0 -15.1 ± 12.4 3.0 ± 5.0 Not available Interobserver error 19.1 ± 14.3 14.4 ± 7.6 -6.4 ± 4.8 Not available Tab. 1. LV mass and volume differences (means ± standard deviations) for 20 test cases. Algorithm: CNN – CMR (as ground truth). Abstract Figure. Fig 1. CMR mesh registered to 3D-echo.