Comparison of Manual, Semi- and Fully Automated Heart Segmentation for Assessing Global Left Ventricular Function in Multidetector Computed Tomography:

Potential for Dose Reduction in CT-Derived Left Ventricular Ejection Fraction: A Simulation Study

BACKGROUND

Measuring left ventricular ejection fraction (LVEF) is important for detecting heart failure, e.g., in treatment with potentially cardiotoxic chemotherapy. MRI is considered the reference standard for LVEF, but availability may be limited and claustrophobia or metal implants still present challenges. CT has been shown to be accurate and would be advantageous, as LVEF could be measured in conjunction with routine chest-abdomen-pelvis oncology CT. However, the use of CT is not recommended due to the excessive radiation dose. This study aimed to explore the potential for dose reduction using simulation. Using an anthropomorphic heart phantom scanned at 13 dose levels, a noise simulation algorithm was developed to introduce controlled Poisson noise. Filtered backprojection parameters were iteratively tested to minimise differences in myocardium-to-ventricle contrast/noise ratio, as well as structural similarity index (SSIM) differences between real and simulated images at all dose levels. Fifty-one clinical CT coronary angiographies, scanned with full dose through end-systolic and -diastolic phases, were located retrospectively. Using the developed algorithm, noise was introduced corresponding to 25, 10, 5 and 2% of the original dose level. LVEF was measured using clinical software (Syngo.via VB50) with papillary muscles in and excluded from the LV volume. At each dose level, LVEF was compared to the 100% dose level, using Bland-Altman analysis. The effective dose was calculated from DLP using a conversion factor of 0.026 mSv/mGycm.

RESULTS

In the clinical images, mean CTDIvol and DLP were 47.1 mGy and 771.9 mGycm, respectively (effective dose 20.0 mSv). Measurements with papillary muscles excluded did not exhibit statistically significant LVEF bias to full-dose images at 25, 10 and 5% simulated dose. At 2% dose, a significant bias of 4.4% was found. With papillary muscles included, small but significant biases were found at all simulated dose levels.

CONCLUSION

Provided that measurements are performed with papillary muscles excluded from the LV volume, the dose can be reduced by a factor of 20 without significantly affecting LVEF measurements. This corresponds to an effective dose of 1 mSv. CT can potentially be used for LVEF measurement with minimal excessive radiation.

Automated Segmentation of Left Ventricular Myocardium on Cardiac Computed Tomography Using Deep Learning

OBJECTIVE

To evaluate the accuracy of a deep learning-based automated segmentation of the left ventricle (LV) myocardium using cardiac CT.

MATERIALS AND METHODS

To develop a fully automated algorithm, 100 subjects with coronary artery disease were randomly selected as a development set (50 training / 20 validation / 30 internal test). An experienced cardiac radiologist generated the manual segmentation of the development set. The trained model was evaluated using 1000 validation set generated by an experienced technician. Visual assessment was performed to compare the manual and automatic segmentations. In a quantitative analysis, sensitivity and specificity were calculated according to the number of pixels where two three-dimensional masks of the manual and deep learning segmentations overlapped. Similarity indices, such as the Dice similarity coefficient (DSC), were used to evaluate the margin of each segmented masks.

RESULTS

The sensitivity and specificity of automated segmentation for each segment (1-16 segments) were high (85.5-100.0%). The DSC was 88.3 ± 6.2%. Among randomly selected 100 cases, all manual segmentation and deep learning masks for visual analysis were classified as very accurate to mostly accurate and there were no inaccurate cases (manual vs. deep learning: very accurate, 31 vs. 53; accurate, 64 vs. 39; mostly accurate, 15 vs. 8). The number of very accurate cases for deep learning masks was greater than that for manually segmented masks.

CONCLUSION

We present deep learning-based automatic segmentation of the LV myocardium and the results are comparable to manual segmentation data with high sensitivity, specificity, and high similarity scores.

Comparison of MRI segmentation techniques for measuring liver cyst volumes in autosomal dominant polycystic kidney disease

PURPOSE

To compare MRI segmentation methods for measuring liver cyst volumes in autosomal dominant polycystic kidney disease (ADPKD).

METHODS

Liver cyst volumes in 42 ADPKD patients were measured using region growing, thresholding and cyst diameter techniques. Manual segmentation was the reference standard.

RESULTS

Root mean square deviation was 113, 155, and 500 for cyst diameter, thresholding and region growing respectively. Thresholding error for cyst volumes below 500ml was 550% vs 17% for cyst volumes above 500ml (p<0.001).

CONCLUSION

For measuring volume of a small number of cysts, cyst diameter and manual segmentation methods are recommended. For severe disease with numerous, large hepatic cysts, thresholding is an acceptable alternative.

Late enhanced computed tomography in Hypertrophic Cardiomyopathy enables accurate left-ventricular volumetry

OBJECTIVES

Late enhancement (LE) multi-slice computed tomography (leMDCT) was introduced for the visualization of (intra-) myocardial fibrosis in Hypertrophic Cardiomyopathy (HCM). LE is associated with adverse cardiac events. This analysis focuses on leMDCT derived LV muscle mass (LV-MM) which may be related to LE resulting in LE proportion for potential risk stratification in HCM.

METHODS

N=26 HCM-patients underwent leMDCT (64-slice-CT) and cardiovascular magnetic resonance (CMR). In leMDCT iodine contrast (Iopromid, 350 mg/mL; 150mL) was injected 7 minutes before imaging. Reconstructed short cardiac axis views served for planimetry. The study group was divided into three groups of varying LV-contrast. LeMDCT was correlated with CMR.

RESULTS

The mean age was 64.2 ± 14 years. The groups of varying contrast differed in weight and body mass index (p < 0.05). In the group with good LV-contrast assessment of LV-MM resulted in 147.4 ± 64.8 g in leMDCT vs. 147.1 ± 65.9 in CMR (p > 0.05). In the group with sufficient contrast LV-MM appeared with 172 ± 30.8 g in leMDCT vs. 165.9 ± 37.8 in CMR (p > 0.05). Overall intra-/inter-observer variability of semiautomatic assessment of LV-MM showed an accuracy of 0.9 ± 8.6 g and 0.8 ± 9.2 g in leMDCT. All leMDCT-measures correlated well with CMR (r > 0.9).

CONCLUSIONS

LeMDCT primarily performed for LE-visualization in HCM allows for accurate LV-volumetry including LV-MM in > 90% of the cases.

KEY POINTS

• LeMDCT of relatively low contrast allows for LV planimetry in HCM. • The correlation of leMDCT-based LV volumetry with gold-standard CMR was excellent (r > 0.9). • LeMDCT requires approximately 2.0mL/kgBW of dye to achieve acceptable contrast.

Model-based automatic segmentation algorithm accurately assesses the whole cardiac volumetric parameters in patients with cardiac CT angiography: a validation study for evaluating the accuracy of the workstation software and establishing the reference values

RATIONALE AND OBJECTIVES

The cardiac chamber volumes and functions can be assessed manually and automatically using the current computed tomography (CT) workstation system. We aimed to evaluate the accuracy and precision and to establish the reference values for both segmentation methods using cardiac CT angiography (CTA).

MATERIALS AND METHODS

A total of 134 subjects (mean age 55.3 years, 72 women) without heart disease were enrolled in the study. The cardiac four-chamber volumes, left ventricular (LV) mass, and biventricular functions were measured with manual, semiautomatic, and model-based fully automatic approaches. The accuracies of the semiautomated and fully automated approaches were validated by comparing them with manual segmentation as a reference. The precision error was determined and compared for both manual and automatic measurements.

RESULTS

No significant difference was found between the manual and semiautomatic assessments for the assessment of all functional parameters (P > .05). Using the manual method as a reference, the automatic approach provided a similar value in LV ejection fraction and left atrial volumes in both genders and right ventricular (RV) stroke volume in women (P > .05), with some underestimation of RV volume (P < .001) and overestimation of all remaining parameters (P < .05) in both genders. In addition, a significantly higher precision with a considerable association in intermeasurement (reproducibility) was observed using the automated approach.

CONCLUSIONS

The model-based fully automatic segmentation algorithm can help with the assessment of the cardiac four-chamber volume and function. This may help in establishing reference values of functional parameters in patients who undergo cardiac CTA.

Time efficiency and diagnostic accuracy of new automated myocardial perfusion analysis software in 320-row CT cardiac imaging

Objective

We aimed to evaluate the time efficiency and diagnostic accuracy of automated myocardial computed tomography perfusion (CTP) image analysis software.

Materials and Methods

320-row CTP was performed in 30 patients, and analyses were conducted independently by three different blinded readers by the use of two recent software releases (version 4.6 and novel version 4.71GR001, Toshiba, Tokyo, Japan). Analysis times were compared, and automated epi- and endocardial contour detection was subjectively rated in five categories (excellent, good, fair, poor and very poor). As semi-quantitative perfusion parameters, myocardial attenuation and transmural perfusion ratio (TPR) were calculated for each myocardial segment and agreement was tested by using the intraclass correlation coefficient (ICC). Conventional coronary angiography served as reference standard.

Results

The analysis time was significantly reduced with the novel automated software version as compared with the former release (Reader 1: 43:08 ± 11:39 min vs. 09:47 ± 04:51 min, Reader 2: 42:07 ± 06:44 min vs. 09:42 ± 02:50 min and Reader 3: 21:38 ± 3:44 min vs. 07:34 ± 02:12 min; p < 0.001 for all). Epi- and endocardial contour detection for the novel software was rated to be significantly better (p < 0.001) than with the former software. ICCs demonstrated strong agreement (≥ 0.75) for myocardial attenuation in 93% and for TPR in 82%. Diagnostic accuracy for the two software versions was not significantly different (p = 0.169) as compared with conventional coronary angiography.

Conclusion

The novel automated CTP analysis software offers enhanced time efficiency with an improvement by a factor of about four, while maintaining diagnostic accuracy.

Informatics in radiology: postprocessing pitfalls in using CT for automatic and semiautomatic determination of global left ventricular function

Recent advances in technical capabilities of computed tomographic (CT) scanners, including an increasing number of detector rows, improved spatial and temporal resolution, and the development of retrospective gating, have allowed the acquisition of four-dimensional (4D) datasets of the beating heart. As a result, the heart can be visualized in different phases and CT datasets can be used to assess cardiac function. Many software packages currently exist that allow automatic or semiautomatic evaluation of left ventricular function on the basis of 4D CT datasets. The level of automation varies from extensive, completely manual segmentation by the user to fully automatic evaluation of left ventricular function without any user interaction. Although the reproducibility of functional parameter assessment is reported to be high and intersoftware variability low for larger groups of patients, significant differences can exist among measurements obtained with different software tools from the same dataset. Thus, careful review of automatically or semiautomatically obtained results is required.

Accuracy of different reconstruction intervals to quantify left ventricular function and mass in cardiac computed tomography examinations

PURPOSE

To compare the accuracy of cardiac dual-source CT (DSCT) reconstructions obtained at 5% and 10% of the cardiac cycle and MRI for quantifying global left ventricular (LV) function and mass in heart transplant recipients.

MATERIAL AND METHODS

We prospectively included 23 heart transplant recipients (21 male, mean age 60±11.7 years) who underwent cardiac DSCT and MRI examinations. We compared LV parameters on cardiac DSCT reconstructions obtained at 5% (0%-95%) and 10% (0%-90%) intervals of the cardiac cycle and on double-oblique short-axis MR images. We determined ejection fraction (EF), end-diastolic volume (EDV), end-systolic volume (ESV), stroke volume (SV), and myocardial mass using commercially available semiautomated segmentation analysis software for DSCT datasets and conventional manual contour tracing for MR studies.

RESULTS

Using different reconstruction intervals to quantify LV parameters at DSCT resulted in non-significant differences (P>.05). Compared to MRI, DSCT slightly overestimated LV-EDV, ESV, and mass when both 5% (11.5±25.1ml, 6.8±10.9ml, and 28.3±21.6g, respectively) and 10% (mean difference 15.3±26.3ml, 7.4±11.5ml, and 29.3±18.7g, respectively) reconstruction intervals were used. DSCT and MRI estimates of EF and SV were not significantly different.

CONCLUSION

In heart transplant recipients, DSCT allows reliable quantification of LV function and mass compared with MRI, even using 10% interval reconstructions.

Segmentation of the heart and great vessels in CT images using a model-based adaptation framework

Recently, model-based methods for the automatic segmentation of the heart chambers have been proposed. An important application of these methods is the characterization of the heart function. Heart models are, however, increasingly used for interventional guidance making it necessary to also extract the attached great vessels. It is, for instance, important to extract the left atrium and the proximal part of the pulmonary veins to support guidance of ablation procedures for atrial fibrillation treatment. For cardiac resynchronization therapy, a heart model including the coronary sinus is needed. We present a heart model comprising the four heart chambers and the attached great vessels. By assigning individual linear transformations to the heart chambers and to short tubular segments building the great vessels, variable sizes of the heart chambers and bending of the vessels can be described in a consistent way. A configurable algorithmic framework that we call adaptation engine matches the heart model automatically to cardiac CT angiography images in a multi-stage process. First, the heart is detected using a Generalized Hough Transformation. Subsequently, the heart chambers are adapted. This stage uses parametric as well as deformable mesh adaptation techniques. In the final stage, segments of the large vascular structures are successively activated and adapted. To optimize the computational performance, the adaptation engine can vary the mesh resolution and freeze already adapted mesh parts. The data used for validation were independent from the data used for model-building. Ground truth segmentations were generated for 37 CT data sets reconstructed at several cardiac phases from 17 patients. Segmentation errors were assessed for anatomical sub-structures resulting in a mean surface-to-surface error ranging 0.50-0.82mm for the heart chambers and 0.60-1.32mm for the parts of the great vessels visible in the images.

Comparison of (semi-)automatic and manually adjusted measurements of left ventricular function in dual source computed tomography using three different software tools

To assess the accuracy of (semi-)automatic measurements of left ventricular (LV) functional parameters in cardiac dual-source computed tomography (DSCT) compared to manually adjusted measurements in three different workstations. Forty patients, who underwent cardiac DSCT, were included (31 men, mean age 58 ± 14 years). Multiphase reconstructions were made with ten series at every 10% of the RR-interval. LV function analysis was performed on three different, commercially available workstations. On all three workstations, end-systolic volume (ESV), end-diastolic volume (EDV), LV ejection fraction (LVEF) and myocardial mass (MM) were calculated as automatically as possible. With the same DSCT datasets, LV functional parameters were also calculated with as many manual adjustments as needed for accurate assessment for all three software tools. For both semi-automatic as well as manual methods, time needed for evaluation was recorded. Paired t-tests were employed to calculate differences in LV functional parameters. Repeated measurements were performed to determine intra-observer and inter-observer variability. (Semi-)automatic measurements revealed a good correlation with manually adjusted measurements for Vitrea (LVEF r = 0.93, EDV r = 0.94, ESV r = 0.98 and MM r = 0.94) and Aquarius (LVEF r = 0.96, EDV r = 0.94, ESV r = 0.98 and MM r = 0.96). Also, good correlation was obtained for Circulation, except for LVEF (LVEF r = 0.45, EDV r = 0.93, ESV r = 0.92 and MM r = 0.86). However, statistically significant differences were found between (semi-)automatically and manually adjusted measurements for LVEF (P < 0.05) and ESV (P < 0.001) in Vitrea, all LV functional parameters in Circulation (P < 0.001) and EDV, ESV and MM (<0.001) in Aquarius Workstation. (Semi-)automatic measurement of LV functional parameters is feasible, but significant differences were found for at least two different functional parameters in all three workstations. Therefore, expert manual correction is recommended at all times.

Assessment of left atrial volumes and function in orthotopic heart transplant recipients by dual-source CT: comparison with MRI

INTRODUCTION

To compare left atrial performance with dual-source CT (DSCT) with respect to magnetic resonance imaging (MRI) in orthotopic heart transplant recipients.

METHODS

Twenty-nine consecutive heart transplant recipients (27 male; mean age 64.1 +/- 13 years; mean time from transplantation 122.8 +/- 69.7 months) referred for exclusion of cardiac allograft vasculopathy underwent cardiac DSCT and MRI. Standard biatrial technique was employed in 13 subjects whereas 16 were transplanted after the bicaval technique. Axial 5-mm slice-thickness DSCT datasets reconstructed in 5% steps of the cardiac cycle and axial 5-mm SSFP-MRI images were analyzed. Two blinded readers manually traced left atrial contours in random order to estimate end-diastolic volume (EDV), end-systolic volume (ESV), and ejection fraction (EF). Parameters were compared with a paired sample Student t test. Concordance correlation coefficient (CCC) was calculated to determine measurement agreement between techniques and observers.

RESULTS

Left atrial volumes were significantly higher with cardiac DSCT (EDV: 170.9 +/- 78.1 mL; ESV: 139.5 +/- 76.6 mL) than with MRI (EDV: 158.2 +/- 72.5 mL; ESV: 124.2 +/- 68.2 mL), whereas left atrial EF was lower with DSCT (EF: 20.8% +/- 7.5% vs. 23.6% +/- 7.7%) (P < 0.05). Measurement agreement between DSCT and MRI was excellent for all parameters (CCC > or =0.82). Individuals operated with the biatrial anastomosis technique presented significantly higher left atrial volumes and lower EF compared with subjects with bicaval anastomosis. Interobserver agreement was excellent for all parameters (CCC > or =0.80).

CONCLUSION

Even if DSCT slightly overestimates left atrial volumes with respect to MRI, results remain clinically valid. Bicaval surgical technique offers improved left atrial performance compared with standard biatrial anastomosis. DSCT may be used as a reliable tool to estimate left atrial parameters in orthotopic heart transplant recipients.