Introduction of a cascaded segmentation pipeline for parametric T1 mapping in cardiovascular magnetic resonance to improve segmentation performance.

The manual and often time-consuming segmentation of the myocardium in cardiovascular magnetic resonance is increasingly automated using convolutional neural networks (CNNs). This study proposes a cascaded segmentation (CASEG) approach to improve automatic image segmentation quality. First, an object detection algorithm predicts a bounding box (BB) for the left ventricular myocardium whose 1.5 times enlargement defines the region of interest (ROI). Then, the ROI image section is fed into a U-Net based segmentation. Two CASEG variants were evaluated: one using the ROI cropped image solely (cropU) and the other using a 2-channel-image additionally containing the original BB image section (crinU). Both were compared to a classical U-Net segmentation (refU). All networks share the same hyperparameters and were tested on basal and midventricular slices of native and contrast enhanced (CE) MOLLI T1 maps. Dice Similarity Coefficient improved significantly (p < 0.05) in cropU and crinU compared to refU (81.06%, 81.22%, 72.79% for native and 80.70%, 79.18%, 71.41% for CE data), while no significant improvement (p < 0.05) was achieved in the mean absolute error of the T1 time (11.94 ms, 12.45 ms, 14.22 ms for native and 5.32 ms, 6.07 ms, 5.89 ms for CE data). In conclusion, CASEG provides an improved geometric concordance but needs further improvement in the quantitative outcome.

Isotropic 3D compressed sensing (CS) based sequence is comparable to 2D-LGE in left ventricular scar quantification in different disease entities.

The goal of this study was to evaluate a three-dimensional compressed sensing (3D-CS) LGE prototype sequence for the detection and quantification of myocardial fibrosis in patients with chronic myocardial infarction (CMI) and myocarditis (MYC) compared with a 2D-LGE standard. Patients with left-ventricular LGE due to CMI (n = 33) or MYC (n = 20) were prospectively recruited. 2D-LGE and 3D-CS images were acquired in random order at 1.5 Tesla. 3D-CS short axis (SAX) images were reconstructed corresponding to 2D SAX images. LGE was quantitatively assessed on patient and segment level using semi-automated threshold methods. Image quality (4-point scoring system), Contrast-ratio (CR) and acquisition times were compared. There was no significant difference between 2D and 3D sequences regarding global LGE (%) (CMI [2D-LGE: 11.4 ± 7.5; 3D-LGE: 11.5 ± 8.5; p = 0.99]; MYC [2D-LGE: 27.0 ± 15.7; 3D-LGE: 26.2 ± 13.1; p = 0.70]) and segmental LGE-extent (p = 0.63). 3D-CS identified papillary infarction in 5 cases which was not present in 2D images. 2D-LGE acquisition time was shorter (2D: median: 06:59 min [IQR: 05:51-08:18]; 3D: 14:48 min [12:45-16:57]). 3D-CS obtained better quality scores (2D: 2.06 ± 0.56 vs. 3D: 2.29 ± 0.61). CR did not differ (p = 0.63) between basal and apical regions in 3D-CS images but decreased significantly in 2D apical images (CR basal: 2D: 0.77 ± 0.11, 3D: 0.59 ± 0.10; CR apical: 2D: 0.64 ± 0.17, 3D: 0.53 ± 0.11). 3D-LGE shows high congruency with standard LGE and allows better identification of small lesions. However, the current 3D-CS LGE sequence did not provide PSIR reconstruction and acquisition time was longer.