Deep Learning-based Automated Segmentation of Left Ventricular Trabeculations and Myocardium on Cardiac MR Images: A Feasibility Study

Left Ventricular Trabeculations at Cardiac MRI: Reference Ranges and Association with Cardiovascular Risk Factors in UK Biobank

Background The extent of left ventricular (LV) trabeculation and its relationship with cardiovascular (CV) risk factors is unclear. Purpose To apply automated segmentation to UK Biobank cardiac MRI scans to (a) assess the association between individual characteristics and CV risk factors and trabeculated LV mass (LVM) and (b) establish normal reference ranges in a selected group of healthy UK Biobank participants. Materials and Methods In this cross-sectional secondary analysis, prospectively collected data from the UK Biobank (2006 to 2010) were retrospectively analyzed. Automated segmentation of trabeculations was performed using a deep learning algorithm. After excluding individuals with known CV diseases, White adults without CV risk factors (reference group) and those with preexisting CV risk factors (hypertension, hyperlipidemia, diabetes mellitus, or smoking) (exposed group) were compared. Multivariable regression models, adjusted for potential confounders (age, sex, and height), were fitted to evaluate the associations between individual characteristics and CV risk factors and trabeculated LVM. Results Of 43 038 participants (mean age, 64 years ± 8 [SD]; 22 360 women), 28 672 individuals (mean age, 66 years ± 7; 14 918 men) were included in the exposed group, and 7384 individuals (mean age, 60 years ± 7; 4729 women) were included in the reference group. Higher body mass index (BMI) (β = 0.66 [95% CI: 0.63, 0.68]; P < .001), hypertension (β = 0.42 [95% CI: 0.36, 0.48]; P < .001), and higher physical activity level (β = 0.15 [95% CI: 0.12, 0.17]; P < .001) were associated with higher trabeculated LVM. In the reference group, the median trabeculated LVM was 6.3 g (IQR, 4.7-8.5 g) for men and 4.6 g (IQR, 3.4-6.0 g) for women. Median trabeculated LVM decreased with age for men from 6.5 g (IQR, 4.8-8.7 g) at age 45-50 years to 5.9 g (IQR, 4.3-7.8 g) at age 71-80 years (P = .03). Conclusion Higher trabeculated LVM was observed with hypertension, higher BMI, and higher physical activity level. Age- and sex-specific reference ranges of trabeculated LVM in a healthy middle-aged White population were established. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Kawel-Boehm in this issue.

Optimizing Deep Learning for Cardiac MRI Segmentation: The Impact of Automated Slice Range Classification

RATIONALE AND OBJECTIVES

Cardiac magnetic resonance imaging is crucial for diagnosing cardiovascular diseases, but lengthy postprocessing and manual segmentation can lead to observer bias. Deep learning (DL) has been proposed for automated cardiac segmentation; however, its effectiveness is limited by the slice range selection from base to apex.

MATERIALS AND METHODS

In this study, we integrated an automated slice range classification step to identify basal to apical short-axis slices before DL-based segmentation. We employed publicly available Multi-Disease, Multi-View & Multi-Center Right Ventricular Segmentation in Cardiac MRI data set with short-axis cine data from 160 training, 40 validation, and 160 testing cases. Three classification and seven segmentation DL models were studied. The top-performing segmentation model was assessed with and without the classification model. Model validation to compare automated and manual segmentation was performed using Dice score and Hausdorff distance and clinical indices (correlation score and Bland-Altman plots).

RESULTS

The combined classification (CBAM-integrated 2D-CNN) and segmentation model (2D-UNet with dilated convolution block) demonstrated superior performance, achieving Dice scores of 0.952 for left ventricle (LV), 0.933 for right ventricle (RV), and 0.875 for myocardium, compared to the stand-alone segmentation model (0.949 for LV, 0.925 for RV, and 0.867 for myocardium). Combined classification and segmentation model showed high correlation (0.92-0.99) with manual segmentation for biventricular volumes, ejection fraction, and myocardial mass. The mean absolute difference (2.8-8.3 mL) for clinical parameters between automated and manual segmentation was within the interobserver variability range, indicating comparable performance to manual annotation.

CONCLUSION

Integrating an initial automated slice range classification step into the segmentation process improves the performance of DL-based cardiac chamber segmentation.

3D Fractal Dimension Analysis: Prognostic Value of Right Ventricular Trabecular Complexity in Participants with Arrhythmogenic Cardiomyopathy

BACKGROUND

Arrhythmogenic cardiomyopathy (ACM) is characterized by progressive myocardial fibro-fatty infiltration accompanied by trabecular disarray. Traditionally, two-dimensional (2D) instead of 3D fractal dimension (FD) analysis has been used to evaluate trabecular disarray. However, the prognostic value of trabecular disorder assessed by 3D FD measurement remains unclear.

PURPOSE

To investigate the prognostic value of right ventricular trabecular complexity in ACM patients using 3D FD analysis based on cardiac MR cine images.

STUDY TYPE

Retrospective.

POPULATION

85 ACM patients (mean age: 45 ± 17 years, 52 male).

FIELD STRENGTH/SEQUENCE

3.0T/cine imaging, T2-short tau inversion recovery (T2-STIR), and late gadolinium enhancement (LGE).

ASSESSMENT

Using cine images, RV (right ventricular) volumetric and functional parameters were obtained. RV trabecular complexity was measured with 3D fractal analysis by box-counting method to calculate 3D-FD. Cox and logistic regression models were established to evaluate the prognostic value of 3D-FD for major adverse cardiac events (MACE).

STATISTICAL TESTS

Cox regression and logistic regression to explore the prognostic value of 3D-FD. C-index, time-dependent receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) to evaluate the incremental value of 3D-FD. Intraclass correlation coefficient for interobserver variability. P < 0.05 indicated statistical significance.

RESULTS

26 MACE were recorded during the 60 month follow-up (interquartile range: 48-67 months). RV 3D-FD significantly differed between ACM patients with MACE (2.67, interquartile range: 2.51 ~ 2.81) and without (2.52, interquartile range: 2.40 ~ 2.67) and was a significant independent risk factor for MACE (hazard ratio, 1.02; 95% confidence interval: 1.01, 1.04). In addition, prognostic model fitness was significantly improved after adding 3D-FD to RV global longitudinal strain, LV involvement, and 5-year risk score separately.

DATA CONCLUSION

The myocardial trabecular complexity assessed through 3D FD analysis was found associated with MACE and provided incremental prognostic value beyond conventional ACM risk factors.

EVIDENCE LEVEL

4 TECHNICAL EFFICACY: Stage 1.

Clair N, Miller GF, Dorfman AL, Fogel MA, Ghelani S, Krishnamurthy R, Lam CZ, Quail M, Robinson JD, Schidlow D, Slesnick TC, Weigand J, Steeden JA, Rathod RH, Muthurangu V. A Deep Learning Pipeline for Assessing Ventricular Volumes from a Cardiac MRI Registry of Patients with Single Ventricle Physiology

Purpose To develop an end-to-end deep learning (DL) pipeline for automated ventricular segmentation of cardiac MRI data from a multicenter registry of patients with Fontan circulation (Fontan Outcomes Registry Using CMR Examinations [FORCE]). Materials and Methods This retrospective study used 250 cardiac MRI examinations (November 2007-December 2022) from 13 institutions for training, validation, and testing. The pipeline contained three DL models: a classifier to identify short-axis cine stacks and two U-Net 3+ models for image cropping and segmentation. The automated segmentations were evaluated on the test set (n = 50) by using the Dice score. Volumetric and functional metrics derived from DL and ground truth manual segmentations were compared using Bland-Altman and intraclass correlation analysis. The pipeline was further qualitatively evaluated on 475 unseen examinations. Results There were acceptable limits of agreement (LOA) and minimal biases between the ground truth and DL end-diastolic volume (EDV) (bias: -0.6 mL/m2, LOA: -20.6 to 19.5 mL/m2) and end-systolic volume (ESV) (bias: -1.1 mL/m2, LOA: -18.1 to 15.9 mL/m2), with high intraclass correlation coefficients (ICCs > 0.97) and Dice scores (EDV, 0.91 and ESV, 0.86). There was moderate agreement for ventricular mass (bias: -1.9 g/m2, LOA: -17.3 to 13.5 g/m2) and an ICC of 0.94. There was also acceptable agreement for stroke volume (bias: 0.6 mL/m2, LOA: -17.2 to 18.3 mL/m2) and ejection fraction (bias: 0.6%, LOA: -12.2% to 13.4%), with high ICCs (>0.81). The pipeline achieved satisfactory segmentation in 68% of the 475 unseen examinations, while 26% needed minor adjustments, 5% needed major adjustments, and in 0.4%, the cropping model failed. Conclusion The DL pipeline can provide fast standardized segmentation for patients with single ventricle physiology across multiple centers. This pipeline can be applied to all cardiac MRI examinations in the FORCE registry. Keywords: Cardiac, Adults and Pediatrics, MR Imaging, Congenital, Volume Analysis, Segmentation, Quantification Supplemental material is available for this article. © RSNA, 2023.

Anatomical Prior-Based Automatic Segmentation for Cardiac Substructures from Computed Tomography Images

Cardiac substructure segmentation is a prerequisite for cardiac diagnosis and treatment, providing a basis for accurate calculation, modeling, and analysis of the entire cardiac structure. CT (computed tomography) imaging can be used for a noninvasive qualitative and quantitative evaluation of the cardiac anatomy and function. Cardiac substructures have diverse grayscales, fuzzy boundaries, irregular shapes, and variable locations. We designed a deep learning-based framework to improve the accuracy of the automatic segmentation of cardiac substructures. This framework integrates cardiac anatomical knowledge; it uses prior knowledge of the location, shape, and scale of cardiac substructures and separately processes the structures of different scales. Through two successive segmentation steps with a coarse-to-fine cascaded network, the more easily segmented substructures were coarsely segmented first; then, the more difficult substructures were finely segmented. The coarse segmentation result was used as prior information and combined with the original image as the input for the model. Anatomical knowledge of the large-scale substructures was embedded into the fine segmentation network to guide and train the small-scale substructures, achieving efficient and accurate segmentation of ten cardiac substructures. Sixty cardiac CT images and ten substructures manually delineated by experienced radiologists were retrospectively collected; the model was evaluated using the DSC (Dice similarity coefficient), Recall, Precision, and the Hausdorff distance. Compared with current mainstream segmentation models, our approach demonstrated significantly higher segmentation accuracy, with accurate segmentation of ten substructures of different shapes and sizes, indicating that the segmentation framework fused with prior anatomical knowledge has superior segmentation performance and can better segment small targets in multi-target segmentation tasks.

Papillary-Muscle-Derived Radiomic Features for Hypertrophic Cardiomyopathy versus Hypertensive Heart Disease Classification

Purpose: This study aimed to assess the value of radiomic features derived from the myocardium (MYO) and papillary muscle (PM) for left ventricular hypertrophy (LVH) detection and hypertrophic cardiomyopathy (HCM) versus hypertensive heart disease (HHD) differentiation. Methods: There were 345 subjects who underwent cardiovascular magnetic resonance (CMR) examinations that were analyzed. After quality control and manual segmentation, the 3D radiomic features were extracted from the MYO and PM. The data were randomly split into training (70%) and testing (30%) datasets. Feature selection was performed on the training dataset. Five machine learning models were evaluated using the MYO, PM, and MYO+PM features in the detection and differentiation tasks. The optimal differentiation model was further evaluated using CMR parameters and combined features. Results: Six features were selected for the MYO, PM, and MYO+PM groups. The support vector machine models performed best in both the detection and differentiation tasks. For LVH detection, the highest area under the curve (AUC) was 0.966 in the MYO group. For HCM vs. HHD differentiation, the best AUC was 0.935 in the MYO+PM group. Comparing the radiomics models to the CMR parameter models for the differentiation tasks, the radiomics models achieved significantly improved the performance (p = 0.002). Conclusions: The radiomics model with the MYO+PM features showed similar performance to the models developed from the MYO features in the detection task, but outperformed the models developed from the MYO or PM features in the differentiation task. In addition, the radiomic models performed better than the CMR parameters' models.

Fully automated intracardiac 4D flow MRI post-processing using deep learning for biventricular segmentation

OBJECTIVES

4D flow MRI allows for a comprehensive assessment of intracardiac blood flow, useful for assessing cardiovascular diseases, but post-processing requires time-consuming ventricular segmentation throughout the cardiac cycle and is prone to subjective errors. Here, we evaluate the use of automatic left and right ventricular (LV and RV) segmentation based on deep learning (DL) network that operates on short-axis cine bSSFP images.

METHODS

A previously published DL network was fine-tuned via retraining on a local database of 106 subjects scanned at our institution. In 26 test subjects, the ventricles were segmented automatically by the network and manually by 3 human observers on bSSFP MRI. The bSSFP images were then registered to the corresponding 4D flow images to apply the segmentation to 4D flow velocity data. Dice coefficients and the relative deviation between measurements (automatic vs. manual and interobserver manual) of various hemodynamic parameters were assessed.

RESULTS

The automated segmentation resulted in similar Dice scores (LV: 0.92, RV: 0.86) and lower relative deviations from manual segmentation in left ventricular (LV) average kinetic energy (KE) (8%) and RV KE (15%) than the Dice scores (LV: 0.91, RV: 0.87) and relative deviations between manual segmentation observers (LV KE: 11%, p = 0.01; RV KE: 19%, p = 0.03).

CONCLUSIONS

The automated post-processing method using deep learning resulted in hemodynamic measurements that differ from a manual observer's measurements equally or less than the variation between manual observers. This approach can be used to decrease post-processing time on intraventricular 4D flow data and mitigate interobserver variability.

KEY POINTS

• Our proposed method allows for fully automated post-processing of intraventricular 4D flow MRI data. • Our method resulted in hemodynamic measurements that matched those derived from manual segmentation equally as well as interobserver variability. • Our method can be used to greatly accelerate intraventricular 4D flow post-processing and improve interobserver repeatability.