Journal of The Royal Society Interface
You have accessResearch articles

Regional ejection fraction and regional area strain for left ventricular function assessment in male patients after first-time myocardial infarction

Soo-Kng Teo

Soo-Kng Teo

Department of Computing Science, Institute of High Performance Computing, Agency for Science, Technology and Research (A*STAR), Singapore, Republic of Singapore

[email protected]

Google Scholar

Find this author on PubMed

,
F. J. A. Vos

F. J. A. Vos

Department of Computing Science, Institute of High Performance Computing, Agency for Science, Technology and Research (A*STAR), Singapore, Republic of Singapore

Technical Medicine, University of Twente, Maarn, The Netherlands

Google Scholar

Find this author on PubMed

,
Ru-San Tan

Ru-San Tan

National Heart Centre Singapore, Republic of Singapore

Duke-NUS Graduate Medical School Singapore, Republic of Singapore

Google Scholar

Find this author on PubMed

,
Liang Zhong

Liang Zhong

National Heart Centre Singapore, Republic of Singapore

Duke-NUS Graduate Medical School Singapore, Republic of Singapore

Google Scholar

Find this author on PubMed

and
Yi Su

Yi Su

Department of Computing Science, Institute of High Performance Computing, Agency for Science, Technology and Research (A*STAR), Singapore, Republic of Singapore

Google Scholar

Find this author on PubMed

Abstract

In this work, we present a method to assess left ventricle (LV) regional function from cardiac magnetic resonance (CMR) imaging based on the regional ejection fraction (REF) and regional area strain (RAS). CMR scans were performed for 30 patients after first-time myocardial infarction (MI) and nine age- and sex-matched healthy volunteers. The CMR images were processed to reconstruct three-dimensional LV geometry, and the REF and RAS in a 16-segment model were computed using our proposed methodology. The method of computing the REF was tested and shown to be robust against variation in user input. Furthermore, analysis of data was feasible in all patients and healthy volunteers without any exclusions. The REF correlated well with the RAS in a nonlinear manner (quadratic fit—R2 = 0.88). In patients after first-time MI, the REF and RAS were significantly reduced across all 16 segments (REF: p < 0.05; RAS: p < 0.01). Moreover, the REF and RAS significantly decreased with the extent of transmural scar obtained from late gadolinium-enhanced CMR images. In addition, we show that the REF and RAS can be used to identify regions with compromised function in the patients with preserved global ejection fraction with reasonable accuracy (more than 78%). These preliminary results confirmed the validity of our approach for accurate analysis of LV regional function. Our approach potentially offers physicians new insights into the local characteristics of the myocardial mechanics after a MI.

1. Introduction

Myocardial infarction (MI), or more commonly known as ‘heart attack’, is a major cause of death and disability worldwide. It usually results in injuries to the myocardial tissues and alters the mechanical properties of the heart ventricles—a process known as remodelling. Each year, an estimated 635 000 Americans have a first hospitalized MI [1]. Accurate and reproducible determination of the left ventricular (LV) function in the heart is essential for the diagnosis, disease stratification and estimation of prognosis for the majority of cardiac diseases. LV remodelling after a MI is currently assessed clinically using changes in global LV volume and ejection fraction (EF). This method aggregates the regional contribution of both the infarcted (injured) and non-infarcted myocardial tissues and does not provide any specific information on the functionality or non-functionality of the infarcted regions. In addition, it has been shown that patients after MI can potentially exhibit LV EF in the ‘normal’ range, illustrating these indexes are insufficient for diagnosis, disease stratification and estimation of prognosis [2,3]. For accurate regional assessment of the myocardial tissue properties and functions, the current clinical ‘gold standard’ is cardiac magnetic resonance (CMR) imaging using delayed contrast enhancement [46]. This method has the advantage that it allows physicians to visually inspect the extent of the infarction to the myocardium tissue through the use of a contrast agent. The infarcted regions typically show higher intensity as compared with the normal myocardial tissue in the late gadolinium-enhanced (LGE) CMR images and the extent of the infarction can be semi-quantified through the use of a scar score. However, the disadvantage of this method is that the LGE scarring score is only semi-quantitative and subject to both intra- and inter-observer variability.

To overcome the above limitation of the LV EF and LGE CMR imaging, different regional indices were proposed in the last decade to enable clinicians to estimate the performance of the LV at specific regions of interest. These regional indices can quantitatively measure the localized functions of the LV and complement existing clinical standard such as LGE CMR imaging. The two most widely accepted regional indices for such localized functional assessment are (i) regional ejection fraction (REF) and (ii) regional myocardium wall strains.

The REF is usually computed using either CMR images or multi-detector computed tomography (CT) images [710]. The partitioning of the LV from the images is based on the 17-segment model that was introduced by Cerqueira et al. in 2002 [11]. This nomenclature published by the American Heart Association (AHA) was designed due to a lack of standardization in medical imaging and has been widely adopted clinically. However, the methodology of calculating the REF has not been standardized and different groups adopt slightly different methods in computing this index. For example, Zeb et al. [9] used a floating point method for partitioning the LV in order to compute the REF, whereas Masci et al. [10] used a fixed point method. The differences in both methods pertain to the definition of the basal–apical axis from the images. For the floating point method, the basal–apical axis is defined independently from the end-diastolic (ED) and end-systolic (ES) frames, respectively, whereas the fixed point method defined this axis from only the ED frame.

Similarly, the regional myocardium wall strain can be computed using echocardiography, CMR images or CT images. For example, Pourmorteza et al. used CT imaging to extract a measure that reflects local myocardial contraction. However, the potential accuracy of the method in computing the myocardium wall strains has yet to be confirmed by a first patient study [12]. Three-dimensional speckle tracking echocardiography is also widely adopted for measuring the circumferential, radial and longitudinal strains in the myocardium [13,14]. Kleijn et al. [15] determined the regional function of the myocardium based on automatic wall motion analysis using three-dimensional speckle tracking echocardiography. Generally, echocardiography showed a high accuracy for assessment of regional LV function but only when performed by an expert reader that makes data reproducibility very difficult. Bogaert et al. were among the first groups to propose using a combined positron emission tomography–magnetic resonance imaging approach to compute both the REF and myocardium wall strains and compare it against myocardial blood flow and glucose metabolism for MI patients [7,16]. However, their results were not validated against LGE scarring score as their method for partitioning the LV is non-standard.

In summary, most of the methodologies described above for computing these regional indices are defined based on either CMR images or multi-detector CT images acquired in the short-axis plane of the LV. In this study, we proposed an alternative geometrical approach for assessing regional LV functions that incorporates information from the long-axis plane of the LV. We proposed to compute the REF and regional area strain (RAS) based on the reconstructed three-dimensional LV geometry from CMR images and used these indices to assess the regional LV functions. This assessment will be validated against the LGE scarring score which is regarded as the clinical benchmark. The reconstruction of our three-dimensional LV geometry relies on standard CMR imaging that is routine in existing clinic and hospital workflow for cardiac investigation and thus our approach can be seamlessly integrated in this workflow without causing any disruptions. In addition, our approach does not require the use of contrast agent and our indices are computed non-invasively in a computationally efficient way; analysis time for a typical case is less than 5 min.

The RAS is a quantitative strain measurement that combines the endocardial wall strains in the circumferential, longitudinal and radial directions. We had first proposed the use of this index for assessing the regional functions in the right ventricle for patients with repaired tetralogy of Fallot [17]. This index is different from the myocardium wall strain measurements from three-dimensional echocardiography, CT imaging and CMR imaging as these wall strains are directional. Furthermore, the RAS can be computed in a frameless manner and is not dependent on the AHA nomenclature. In this paper, we computed the RAS using the AHA nomenclature in order to validate against the REF and LGE scarring score. We postulate that the RAS can provide additional information to physicians about the regional mechanical properties of the LV myocardial tissues. The RAS generally reflects the extent of deformation in the LV endocardial surface (the inner surface of the ventricle) during contraction and relaxation. During normal contraction, the endocardial surface deforms due to the shortening of the myofibres embedded inside the tissue. We hypothesize that the RAS should decrease at infarcted regions after MI as a result of myocardial cell death due to prolonged ischaemia (restriction to blood supply) [18,19]. Similarly, the REF should also decrease at infarcted regions after MI as the contribution of that region to the pumping efficacy of the LV is decreased due to reduced deformation of the endocardial surface. Taken together, the REF and RAS can be used for improving the diagnosis, disease stratification and estimation of prognosis after MI.

The novelty of our approach is twofold: (i) the use of the RAS and REF to assess regional LV function after MI with validation against LGE scarring score and (ii) applying the RAS and REF to discriminate MI patients with preserved global EF (more than or equal to 50%) specifically in a subgroup of patients. We demonstrate that the RAS and REF can be used to identify regions in the LV associated with compromised function (as indicated by the LGE scarring score) for this particular subgroup of patients. This identification is important as this subgroup of patients exhibit ‘normal’ global EF and could be diagnosed as healthy based on the use of global indices.

The primary objectives of this paper are therefore to: (i) validate the reproducibility of the REF using our geometrical approach, (ii) assess the correlation between REF and RAS for both the patient and control (healthy) groups, (iii) compare the differences in REF and RAS between patients and control groups, and (iv) assess the feasibility of using the REF and RAS to identify regions with compromised function in the patient subgroup with preserved global EF.

2. Material and methods

2.1. Population

This study consisted of 30 MI patients and nine age- and sex-matched healthy volunteers (table 1). All 30 patients had a first MI, with a successfully reperfused ST-elevation, weeks after the acute event. For the healthy subjects, none of them had: (i) significant valvular or congenital cardiac disease, (ii) history of MI, (iii) coronary artery lesions, or (iv) abnormal LV pressure, ED volume or EF.

Table 1.Population characteristics, for healthy subjects (n = 9) and patients (n = 30). n.s., not significant.

parameter patients healthy p
age (years) 53 ± 10 53 ± 10 n.s.
gender (M/F) 30/0 9/0 n.s.
weight (kg) 75 ± 17 72 ± 11 n.s.
body surface area (m2) 1.85 ± 0.22 1.90 ± 0.19 n.s.
systolic blood pressure (mmHg) 127 ± 20 131 ± 15 n.s.
diastolic blood pressure (mmHg) 76 ± 9 77 ± 6 n.s.
LV ED volume index (ml) 184 ± 46 133 ± 36 <0.0001
LV ES volume (ml) 103 ± 46 45 ± 20 <0.0001
LV EF (%) 46 ± 11 67 ± 6 <0.0001

All subjects underwent CMR scanning using steady-state free precession cine gradient echo sequences imaged on a 1.5 T Siemens scanner (Avanto, Siemens Medical Solutions, Erlangen). Scans for the patients were acquired one to three months after MI and three sets of CMR images were taken. The first set of images (short-axis) was taken along the plane that passes through the mitral and aortic valves of the heart. The second set of images (long-axis) was taken on planes orthogonal to the short-axis images, and oblique to each other, giving an angular cross-sectional view of the LV. The last set of images was taken on the vertical cross section (long-axis) orthogonal to and connecting with the short-axis plane images. The CMR images have a spatial resolution of 1.5 mm in-plane and 8 mm out-of-plane, acquired in a single breath hold, with 22 temporal phases per heart cycle. Of these images, those corresponding to the cardiac cycle at end-diastole (ED) and end-systole (ES) are manually segmented and used for analysis. The ED and ES frames are defined using the valve-closure and valve-opening images, respectively; the frame of the valve-closure (valve-opening) image is designated as the ED (ES) frame. To assess degree of MI, LGE CMR scans were also acquired for the MI patients. Scar scores were defined as the regional increase in CMR signal intensities 20 min after injection of 0.2 mmol kg−1 gadolinium-diethylenetriamine pentaacetic acid. All subjects were recruited without consideration of gender or ethnicity and gave informed consent. The study protocol was approved by the SingHealth Centralised Institutional Review Board.

2.2. Three-dimensional geometrical reconstruction and mesh partitioning

To perform segmentation of the endocardium, images from the CMR acquisition were processed using the LVtools plugin included in the CMRtools software package (Cardiovascular Solution, UK). Segmentation of the LV endocardial contours is manually performed on both the short- and long-axis images. Papillary and trabeculae were excluded from the myocardium during segmentation. The contours on these two sets of images are then fused to reduce registration errors during the reconstruction process. Control points are fixed on the surface of the reconstructed endocardium and these points are defined by the intersection of the short- and long-axis views. To create a more realistic reconstruction, we also use the long-axis views oriented at regular angular intervals about the LV axis. A series of B-spline curves are then generated to represent the contours of the endocardial surface.

The endocardial contours are subsequently exported from CMRtools and an in-house meshing toolkit is used to reconstruct the three-dimensional geometry of the LV endocardial surface in terms of a two-manifold structured triangle mesh (denoted as Ω). Using discrete geometrical representation of the LV endocardial surface affords ease in the extraction of various clinically meaningful indices [2023]. To facilitate analyses, the endocardial mesh is partitioned into 16 segments based on the standard published by the AHA [11]. As the standardized nomenclature was established for image-based data, we proposed a modified approach to generate the 16-segment model for three-dimensional geometrical models [20]. A typical three-dimensional LV geometry reconstructed from CMR images is as shown in figure 1. Abidance to this recommended nomenclature allows us to achieve adequate sampling of the LV without exceeding the relevant limits for clinical and research applications. It is worth noting that Segment 17 in the standardized nomenclature has been omitted in our approach because it is difficult to acquire the true apex position due to the inter-slice spacing of our CMR images (8 mm). Details of the mesh partitioning approach can be found in our previous work [20].

Figure 1.

Figure 1. (a) Regional partitioning of the LV superimposed onto the CMR images for one sample patient. Basal region: segments 1–6; mid-cavity region: segments 7–12; apical region: segments 13–16. (b) Partitioning of the reconstructed three-dimensional LV geometry with the corresponding two-dimensional bulls-eye plot. (Online version in colour.)

2.3. Calculation of regional area strain

The RAS is a dimensionless quantity that measures the magnitude of change in the surface area of the myocardial tissue as it contracts from ED to ES. This index can be regarded as a parameter that integrates the effects of deformation of the myocardium tissues in the longitudinal, circumferential and radial directions. In our previous work, we had employed the RAS in the study of patients with repaired tetralogy of Fallot [17]. To calculate the RAS of each segment, we use

Display Formula
2.1
where SAi,ED is the endocardial surface area of Segment i at ED and SAi,ES is the endocardial surface area of Segment i at ES.

2.4. Calculation of regional ejection fraction

The representation of the LV as surface meshes presents us with the convenience of calculating its volume/sub-volumes by making use of only its vertex information. First, the divergence theorem is used to reduce the volume integral to a sum of surface integrals over the individual faces of the LV mesh. Next, these surface integrals are then further reduced to a sum of line integrals using Green's theorem [24]. Finally, the volume Vi of each LV region is given by

Display Formula
2.2
where (x0, y0, z0), (x1, y1, z1) and (x2, y2, z2) are the coordinates of the vertices of a face ϕ of the LV sub-mesh Ωi representing the Segment i (1 ≤ i ≤ 16). The convention of the face vertex ordering is taken in the counter-clockwise direction with the face normal pointing away from the LV chamber. The REFi of segment i is then defined as
Display Formula
2.3
where Vi,ED is the volume of Segment i at ED and Vi,ES is the volume of Segment i at ES. This definition of the REF in our geometrical approach can be seen as an extension of the LV EF as defined by the AHA [25].

2.5. Myocardium scar scoring

Myocardium tissue scarring for patients was graded on a four-point scale by an experienced cardiologist for each segment based on the CMR intensity of the infarcted tissues and the size of the infarction. This scoring is performed using the LGE CMR images. Scoring criteria: 0 = no infarction, 1 = 1–25% infarction of myocardium, 2 = 26–50% infarction of myocardium, 3 = 51–75% infarction of myocardium and 4 = 76–100% infarction of myocardium, according to the AHA [26,27].

2.6. Validation

Our approach is fully automated for a given set of segmented endocardial contours and there is no variability in the computation. The only variability in our geometrical-based approach for computing the REF is the user-specified location of the anatomical landmark (pref) on the basal contour as shown in figure 1. This landmark corresponds to the anterior attachment junction of the right ventricle to the LV and is used for orientating the endocardial surface mesh. Therefore, to validate the robustness and reproducibility of our approach, we studied the effect due to noise in the pref location to mimic intra- and inter-observer variabilities. Specifically, we randomly varied the location of pref by 1.5 mm to 4.5 mm (the equivalent of 1–3 image pixels) in three independent runs and recomputed the REF for each subject. For each run, the randomization of pref for each subject was performed three times, i.e. there are three sets of recomputed REF at each of the pref variations at 1.5, 3.0 and 4.5 mm. These results were compared to the original REF using the Bland–Altman analysis to test the robustness and reproducibility of our approach.

We note that it is also possible to orientate the LV endocardial surface using either (i) the posterior attachment junction or (ii) both posterior and anterior attachment junctions. Both choices are valid and the anatomical landmark(s) is used solely for reference to orientate our mesh to the physical LV anatomy. In this paper, we had chosen to use the anterior attachment junction and have focused on the effect of its variability on our proposed geometric-based approach to compute REF. Our rationale for using one reference point as opposed to two is to minimize variability arising from additional user input.

2.7. Statistical analysis

All quantitative data are expressed as the mean ± s.d. Comparison between pairs of quantitative variables was performed by using an independent sample parametric test (Student's t-test). The association between the RAS and the REF was studied using the coefficient of determination (R2). The Bland–Altman analysis was used for the validation of our geometry-based methodology for computing the REF with a 95% CI. Comparison of means across two or more samples is performed using one-way ANOVA with the Tukey test to determine significant differences between pairs of means. All statistical tests are performed using the statistical programming language R 3.1.1 [28].

3. Results

3.1. The geometrical-based approach of computing regional ejection fraction is robust and reproducible

To test the robustness of our approach, the REF was recomputed by varying the location of the anatomical landmark (see subsection Validation) in the control group. The results of our robustness analysis are plotted in the form of linear regression and Bland–Altman plots, as shown in figure 2, using one representative set of the recomputed REF at each run. The mean difference in the REF between our original data and the recomputed data (y-axis values in the Bland–Altman plots) is approximately zero in all three runs when the landmark location was varied by 1.5 mm, 3.0 mm and 4.5 mm, respectively. This suggests that there is no systematic error in our REF computation approach associated with the landmark location. Similarly, the 95% confidence limit (representing the mean difference ±1.96 s.d., denoted by the dotted-line in the Bland–Altman plots) contains 94%, 96% and 95% of the REF variations, respectively, in the three runs. This suggests that the REF computation approach is not sensitive to the exact location of the anatomical landmark. Furthermore, the correlations between the original REF and the recomputed REF were excellent (R2 = 0.99) in all three runs, thereby validating our approach for computing the REF to be both reproducible and robust against variation in user input.

Figure 2.

Figure 2. Validation of our approach for computing REF: linear regression (a) and Bland–Altman (b) plots for illustrating the agreement between original REF and the recomputed REF by varying the location of the anatomical landmark by 1.5 mm (a(i),b(i)), 3.0 mm (a(ii),b(ii)) and 4.5 mm (a(iii),b(iii)).

In addition, we have also recomputed the REF and RAS by varying the input contours for a subset of subjects in our study (see the electronic supplementary material). Specifically, there was also no statistical difference between the original REF/RAS results and the recomputed REF/RAS results (REF: p = 0.92 > 0.05; RAS: p = 0.61 > 0.05 using a paired t-test), suggesting that our approach is indeed robust against variation in the user-segmented input contours.

3.2. There is excellent correlation between regional ejection fraction and regional area strain for both control and patient group

For the control group, the REF and RAS were calculated for all segments (9 controls × 16 segments = 144 segments) using our proposed approach. The mean global EF and area strain were 69.7% and 73.5%, respectively (table 2). Both the REF and RAS demonstrated an increase from basal level towards the apical part of the ventricle. Functional non-uniformity was also noted between the inferior and anterior regions of the LV. The contribution to the REF of the inferior region (segments 4, 10 and 15) was significantly higher compared with the anterior region (segments 1, 7 and 13) throughout the entire ventricle (p < 0.01). No such differences were found between the lateral and septal side.

Table 2.Table of normal reference values for REF and RAS for healthy subjects (n = 9).

segments REF (%) RAS (%)
(1) basal anterior 56.5 ± 8.7 56.4 ± 10.5
(2) basal anterior septal 55.7 ± 6.1 51.2 ± 8.1
(3) basal inferior septal 63.2 ± 6.4 54.4 ± 8.5
(4) basal inferior 67.0 ± 7.0 58.2 ± 9.9
(5) basal inferior lateral 64.0 ± 6.4 57.3 ± 11.6
(6) basal anterior lateral 59.9 ± 8.4 59.1 ± 14.5
(7) mid anterior 69.2 ± 7.0 75.1 ± 15.1
(8) mid anterior septal 70.1 ± 8.5 72.4 ± 15.0
(9) mid inferior septal 74.6 ± 9.0 75.5 ± 15.4
(10) mid inferior 78.4 ± 7.0 84.1 ± 19.0
(11) mid inferior lateral 77.4 ± 6.6 86.8 ± 22.0
(12) mid anterior lateral 72.4 ± 8.9 80.4 ± 21.3
(13) apical anterior 73.2 ± 8.5 84.4 ± 12.8
(14) apical septal 74.1 ± 10.3 85.7 ± 12.0
(15) apical inferior 80.4 ± 5.9 97.7 ± 14.2
(16) apical lateral 79.5 ± 7.8 97.2 ± 15.9
aggregrating over the basal, mid-cavity and apical regions
regions REF RAS
(i) basal 61.0 ± 8.0 56.1 ± 10.6
(ii) mid-cavity 73.7 ± 8.3 79.0 ± 18.1
(iii) apical 76.8 ± 8.5 91.3 ± 14.6
aggregating over the entire LV
EF AS
global 69.7 ± 10.7 73.5 ± 20.5

For the patient group, the REF and RAS were also calculated for all segments (30 patients × 16 segments = 480 segments). No exclusion of segments was necessary (table 3). To assess the correlation between the REF and RAS, a scatter plot was generated combining data from both the control and patient group (figure 3). It can be seen from the plot that the RAS has an excellent correlation with the REF. Using a quadratic fit, the R2 is 0.88 for the combined data (n = 624). We observe that the slope of the quadratic fit at higher REF (more than 40%) is larger compared with that at lower REF. This observation suggests that the RAS could potentially be more sensitive at functional assessment in segments with higher REF. In summary, the RAS reflects the deformation in the endocardial surface and can be used to measure the regional LV contraction performance, whereas the REF reflects the regional contribution to the overall pumping efficacy of the LV.

Table 3.Patient database overview (n = 30) of all segments for different ventricular levels, based on the LGE scar score. Segments are grouped together based on their LGE CMR scarring score into four categories: (i) LGE3-4, (ii) LGE1-2, (iii) border (LGE0), and (iv) remote (LGE0). Refer to main text for details of grouping.

LGE0
LGE1-2 LGE3-4 total
remote border
basal 54 84 28 14 180
mid-cavity 21 91 37 31 180
apex 16 23 46 35 120
total 91 198 111 80 480
Figure 3.

Figure 3. Correlation between the REF and RAS for the combined data from normal control and patient groups. The RAS has an excellent correlation with the REF. Fitting to the data is performed using a quadratic functional. (Online version in colour.)

3.3. There is a significant drop in the regional ejection fraction and regional area strain after myocardial infarction

A comparison was made between the control and patient group using both the REF and RAS. An overview is shown in figure 4. We observed a significant decrease for both the REF and RAS across all 16 segments in the patient group after MI (REF: p < 0.05; RAS: p < 0.01). A decrease of the REF to below 44% for patients is visible across the basal, mid-cavity and apical regions (table 4). Similarly, the RAS in these regions also decreased to less than 38%. Due to necrosis in the infarcted regions, cells will not be able to deform and play a part in the contraction of the LV. As a result, less blood will be properly transported through the heart, resulting in lower REF and correspondingly lower global EF.

Table 4.Comparison of the REF and RAS for control and patient groups at the base, mid-cavity and apex level.

level REF (%)
RAS (%)
control patient p control patient p
basal 61.0 ± 8.0 43.2 ± 15.1 <0.01 56.1 ± 10.6 34.7 ± 17.3 <0.01
mid-cavity 73.7 ± 8.3 43.9 ± 19.8 <0.01 79.0 ± 18.1 37.4 ± 21.5 <0.01
apex 76.8 ± 8.5 39.6 ± 25.6 <0.01 91.3 ± 14.6 36.9 ± 23.7 <0.01
Figure 4.

Figure 4. Comparison of REF (a) and RAS (b) between control and patient groups. There is a significant decrease in both the REF and RAS in the patient group across all 16 segments as compared with the control group (REF: p < 0.05; RAS: p < 0.01).

3.4. There is a significant drop in the regional ejection fraction and regional area strain for infarcted segments as compared with non-infarct segments

The REF and RAS were correlated to the LGE CMR scarring scores in the patient group as LGE CMR is the current ‘gold standard’ for LV function assessment. For this analysis, segments are grouped together based on their LGE CMR scarring score into four categories: (i) High LGE scarring score of 3 or 4 (denoted as LGE3-4), (ii) Low LGE scarring score of 1 or 2 (denoted as LGE1-2), (iii) LGE scarring score of 0 but adjacent to another infarcted segment (denoted as Border), and (iv) LGE scarring score of 0 with all adjacent segments also having LGE scarring score of 0 (denoted as Remote). The rationale for this grouping is twofold: (i) to study the effect of scarring extent on regional functions and (ii) to study if regional functions are compromised in segments that are adjacent to infarct segments. The results of our analysis are shown in figure 5.

Figure 5.

Figure 5. Box-whisker plot correlating the LGE scar score with REF (a) and RAS (b) for the entire LV (a(i),b(i)), basal region (a(ii),b(ii)), mid-cavity region (a(iii),b(iii)) and apical region (a(iv),b(iv)). On each box, the central mark (in red) is the median, the edges of the box are the 25th and 75th percentiles and the whiskers extend to approximately ±2.7 s.d. coverage if the data are normally distributed. Segments are grouped together based on their LGE CMR scarring score into four categories: LGE3-4, LGE1-2, border (LGE0) and remote (LGE0). Refer to main text for details of grouping (*p < 0.05, **p < 0.01). (Online version in colour.)

Figure 5a(i),b(i) shows the correlation for all 16 segments (global analysis). Both REF and RAS for segments in the Remote group are significantly different from those segments in the Border group (p < 0.05), LGE1-2 group (p < 0.01) and LGE3-4 group (p < 0.01). Similarly, both REF and RAS for segments in the Border group are significantly different from those segments in the LGE1-2 group (p < 0.01) and LGE3-4 group (p < 0.01). The subsequent panels in figure 5 show the correlation for segments aggregated over the basal, mid-cavity and apical regions, respectively. A similar trend can also be seen:

Basal region. Both REF and RAS for segments in the Remote group are significantly different from those segments in the Border group (p < 0.01), LGE1-2 group (p < 0.01) and LGE3-4 group (p < 0.01).

Mid-cavity region. Both REF and RAS for segments in the Remote group are significantly different from those segments in the LGE1-2 group (p < 0.01) and LGE3-4 group (p < 0.01).

Apical region. Both REF and RAS for segments in the Remote group are significantly different from those segments in the LGE1-2 group (REF: p < 0.01; RAS: p < 0.05) and LGE3-4 group (p < 0.01).

Summarizing our observation, the REF and RAS in the patient group were significantly different between segments in the Remote group and infarct segments (LGE1-2 and LGE3-4 group), suggesting that both indices can potentially be used for the identification of the infarct areas. This difference was observed across the basal, mid-cavity and apical regions, showing that our method of discrimination is robust and independent of the location of the infarct regions. Furthermore, we also observe that both REF and RAS for segments in the Remote group are higher as compared with the Border group. We hypothesize that segments in the Border group exhibit some degree of compromised function arising from their proximity to the adjacent infarcted segments.

3.5. It is feasible to identify segments with compromised function for patients with ‘normal’ global ejection fraction using the regional ejection fraction and regional area strain

From the patient group, we identified a subgroup of patients with preserved global EF (more than or equal to 50%) and show that the REF and RAS can be used to identify segments with compromised function in this subgroup. Here, we define segments with compromised function as segments in the Border, LGE1-2 or LGE3-4 groups (see above section). The identification was performed by considering the basal, mid-cavity and apical regions independently for each patient with the threshold values for REF and RAS obtained using (mean − 1.0 × s.d.) from the normal control group. The results of our analysis are shown in table 5.

Table 5.Identification of segments with compromised functions for patients with ‘normal’ EF using the REF and RAS. True positive (TP) are segments in the Border, LGE1-2 or LGE3-4 groups correctly identified as segments with compromised function; true negative (TN) are segments in the Remote group correctly identified as segments with no compromised function; false negative (FN) are segments in the Border, LGE1-2 or LGE3-4 groups that are not identified as segments with compromised function; false positive (FP) are segments in Remote group mis-identified as segments with compromised function. Sensitivity (Se) is defined as 100 × (TP)/(TP + FN); specificity (Sp) is defined as 100 × (TN)/(TN + FP); positive predictive value (+P) is defined as 100 × (TP)/(TP + FP); and negative predictive value (−P) is defined as 100 × (TN)/(TN + FN). Accuracy (Acc) is the average of Se and +P. No. segments refer to the total number of segments with compromised function, i.e. segments in the Border, LGE1-2 or LGE3-4 groups.

ID global EF no. segments REF
RAS
Se +P Sp −P Acc Se +P Sp −P Acc
1 50.0 13 84.6 78.6 0.0 0.0 81.6 100.0 81.3 0.0 n.a. 90.6
2 50.0 15 73.3 100.0 100.0 20.0 86.7 93.3 100.0 100.0 50.0 96.7
3 50.0 11 100.0 68.8 0.0 n.a. 84.4 100.0 73.3 20.0 100.0 86.7
4 51.7 11 90.9 71.4 20.0 50.0 81.2 81.8 64.3 0.0 0.0 73.1
5 52.3 11 90.9 90.9 80.0 80.0 90.9 81.8 75.0 40.0 50.0 78.4
6 54.5 14 85.7 92.3 50.0 33.3 89.0 78.6 91.7 50.0 25.0 85.1
7 55.4 13 76.9 100.0 100.0 50.0 88.5 69.2 100.0 100.0 42.9 84.6
8 56.1 14 78.6 100.0 100.0 40.0 89.3 85.7 100.0 100.0 50.0 92.9
9 56.9 14 50.0 100.0 100.0 22.2 75.0 57.1 100.0 100.0 25.0 78.6
10 57.2 10 100.0 90.9 83.3 100.0 95.5 90.0 81.8 66.7 80.0 85.9
11 58.1 10 70.0 77.8 66.7 57.1 73.9 70.0 77.8 66.7 57.1 73.9
12 62.3 10 70.0 87.5 83.3 62.5 78.8 20.0 66.7 83.3 38.5 43.3
13 63.9 10 60.0 100.0 100.0 60.0 80.0 70.0 77.8 66.7 57.1 73.9
14 67.6 12 0.0 0.0 100.0 25.0 0.0 8.3 100.0 100.0 26.7 54.2
gross average 73.6 82.7 70.2 46.2 78.2 71.9 85.0 63.8 46.3 78.4

For this subgroup of patients, our approach has an accuracy of approximately 78% (REF—78.4%; RAS—78.2%) for identifying segments with compromised function. Generally, the accuracy of our approach decreases with increasing global EF in this patient subgroup. For patients with global EF < 60%, our approach has an average accuracy of 84.2% (REF) and 85.1% (RAS). Conversely, for patient 14 with global EF of 67.6% (grey highlight in table 5), our approach either failed (based on REF) or was able to identify 1 segment out of the 12 segments with compromised function (based on RAS). Nonetheless, we had shown that it is feasible to use the REF and RAS to identify regions with compromised function in patients with preserved global EF. The RAS across all 16 segments for patient 8 are shown in figure 6 to illustrate our approach graphically. For this particular patient, there are two false negative (FN) segments (segments 3 and 6), i.e. segments in the Border, LGE1-2 or LGE3-4 groups that are not identified as segments with compromised function.

Figure 6.

Figure 6. Bar chart of the RAS for all 16 segments for a typical patient with preserved global EF. The detection thresholds for identifying segments with compromised function are denoted by the red horizontal lines. Different thresholds were used for the basal, mid-cavity and apical regions. Segments with RAS below the threshold are identified as segments with compromised function. B and R denote Border and Remote segments, respectively (both groups have LGE scarring score of 0). (Online version in colour.)

4. Discussion

This paper presents a validated computer-based approach to determine the REF and RAS for patients after a first-time MI. After MI, while only a selected part of the LV is injured, the global LV performance is affected. In daily clinical practice, global indicators are still used for the assessment of the LV function despite their limitation in providing regional information [2,3]. To overcome this limitation, regional indices such as REF and regional myocardium wall strains were introduced to estimate localized functions at specific regions of interest. However, there are still no standardized approaches for computing the REF from imaging and the measurement of myocardium wall strains from echocardiography is still heavily dependent on the skills of the operator. In this paper, we proposed an alternative geometrical approach for assessing regional LV functions that incorporates information from both the short- and long-axis CMR images of the LV. We computed the REF and RAS based on the reconstructed three-dimensional LV geometry and showed that both indices correlate well with the LGE CMR scarring score (see figure 5). We also showed that both indices can be used to identify segments with compromised function with reasonable accuracy in patients with preserved global EF (see table 5).

Our approach for computing the REF depends on the segmented contours from the CMR images and a user-specified location of an anatomical landmark (anterior attachment junction of the right ventricle to the LV). We had shown that the REF computation is indeed robust to variations in the location of the anatomical landmark. The reproducibility of the REF to variations in the segmented contours is not considered in this study as it has been shown that LV parameters are highly reproducible [29,30]. As such, there is no reason to doubt that our endocardial contour segmentation protocols will induce any significant variations to the computed REF or RAS values. We acknowledge that the values of the computed REF and RAS are dependent on both the segmented contours and location of the user-defined anatomical landmark. However, it is highly unlikely to change the trend and interpretation of our reported results.

We used the valve-closure (valve-opening) time to define the ED and ES frames in our approach for computing the REF and RAS. It is possible that the time to reach peak REF for certain segments might not coincide with the ES frame as defined above. A more accurate approach to track the time to reach peak REF is then to compute the volume against time and use the frame of minimum volume to compute the REF. This is the approach proposed by Suinesiaputra et al. [31] for assessing intraventricular dyssynchrony for the LV. However, the main objective of our study is to study the differences in REF and RAS between patients and control groups and not to measure any potential LV intraventricular dyssynchrony. As such, using the valve-opening time to define ES is still valid and justifiable as this provides a consistent basis for comparing across subjects in both the patient and control groups.

For the normal control group, we observed that the REF and RAS demonstrated an increase from basal level towards the apical part of the ventricle. We postulate that this observation is a result of the increased LV contraction and deformation towards the apex as compared with the basal region. This can be explained by the increase in twisting motion at the apex regions as the myofibres from the endocardial surface run in an almost opposite direction to the myofibres on the epicardial surface. For the patient group, we did not observe any such trend in the REF and RAS (see table 4). This suggests that after MI, the LV contraction and deformation deviates from the trend seen in normal controls as certain regions lose contractile function.

Also, the correlation between the REF and RAS (see figure 3) appears to be nonlinear. We used a quadratic functional to fit the scatter plot of the REF against RAS and obtained a coefficient of determination (R2) of 0.88. The slope of the quadratic fit at higher REF (more than 40%) is larger compared with that at lower REF, suggesting that the RAS (REF) could potentially be more sensitive at function assessment in segments with higher (lower) REF. Combining both indices for regional function assessment and discrimination of patients will be advantageous and could potentially increase the accuracy of the assessment.

For identifying segments with compromised function in patients with preserved EF, we used a detection threshold of (mean − 1.0 × s.d.) derived from the normal control group. Our results show that the accuracy of our detection generally decreases with increasing global EF. For one particular patient in this study (patient 14 in table 5), the infarcted segments exhibit REF and RAS that are comparable to normal controls. This observation is interesting and could warrant further investigation to ascertain if infarcted segments after MI are able to retain any contractile functions. It could also be possible that for this particular patient, the size of the infarct is small compared with the size of the segments and hence the effect of the infarction on the functionality of the segment is minimal. We can also further increase the accuracy of our approach by either (i) varying our detection threshold or (ii) increasing the number of normal controls. We acknowledge that the number of patient and normal controls in this study is limited and reiterate that this is a proof-of-concept study.

5. Clinical applicability

CMR imaging is essential to assess the degree of viability of the LV myocardial tissues. If available, LGE CMR imaging and single-photon emission computed tomography are the preferred options for such regional assessment. However, both techniques require a contrast agent or radionuclide tracer to improve the image quality for infarct localization. All REF and RAS data presented in this article are obtained from CMR images without the use of any contrast agent. Furthermore, the RAS showed an excellent correlation with the REF, making the RAS another potential candidate to estimate the extent of MI on a regional level. In daily practice, our proposed indices can provide information to physicians regarding (i) the pumping efficiency of all 16 segments of the LV (through the REF) to enable localization of significant coronary artery stenosis and (ii) the contractility of the myocardium (through RAS) that can elucidate the alterations in the mechanical properties of the myocardial tissues (figure 6). This approach potentially offers the physician with new insights into the local characteristics of the myocardial tissues after an infarction, which are essential for diagnosis and disease stratification.

6. Limitations

Our approach relies on the availability of segmented LV endocardium CMR images that are performed manually, which is considered time-consuming, as well as the availability of the magnetic resonance imaging acquisition sequence. Automatic image segmentation techniques [32,33] could be explored to speed up the manual segmentation process. Other segmentation techniques such as shape-based interpolation [34] and super-resolution [35,36] can also be potentially used to automate and increase the accuracy of the cardiac segmentation. The deployment of such an automated segmentation algorithm will allow our computation of the REF and RAS to be further speeded up as compared with our current approach. Furthermore, as clinical CMR images are highly anisotropic with inter-slice spacing in the range of 5–10 mm, geometrical interpolation between image slices needs to be performed and this may affect the accuracy of the geometry reconstruction. A 16-segment model was used to partition different regions of the LV. The adoption of such a nomenclature allows us to achieve adequate sampling of the LV without exceeding the relevant limits for clinical and research purposes. In contrast to standardized nomenclature, a minor adjustment has been made. Due to the generally intricate shape of the apical region of the LV, and the fact that the apex is below the bottom-most apical short-axis slice, it is difficult to locate the exact location of the apex. Therefore, region 17 is completely excluded from consideration. For partitioning, different planes were specified to split the mesh into 16 segments based on two centroid points at the basal and apical contours of the LV. This two-point method does not allow the partitioning to take into account the unique curvature of the LV. Currently, more work is being done to improve this method.

Finally, we acknowledge that the number of patients in our study is relatively limited (30 patients). As such, the statistical power of our results has to be further substantiated with additional clinical trials involving more patients. The two main objectives of this paper are to prove the feasibility of (i) using the RAS and REF to assess regional LV function after MI with validation against LGE scarring score and (ii) applying the RAS and REF to discriminate MI patients with preserved global EF (more than or equal to 50%). As a proof-of-concept, we have shown that our approach is able to meet both objectives successfully in the group of patients recruited for this study. As the average age of the patient group is 53 years, we were also limited in our recruitment of healthy normal controls that are age-matched to the patient group because it is difficult to recruit older subjects without any history of cardiac diseases. Using normal controls from a younger age group in this study will not be feasible as that will skew the REF and RAS of the normal control group to a higher value. This is because it will be reasonable to assume that myocardium contractility and function are higher in younger subjects as compared to older subjects.

7. Conclusion

In this work, we had presented a method to compute the REF and RAS of the LV from CMR images. We had shown that the approach was robust and reproducible with respect to variation in user input. The REF had an excellent correlation with the RAS for both the patient and control groups. There was also a significant decrease in the REF and RAS in patients after MI, which is consistent with the extent of myocardial scarring. Finally, both the REF and RAS can potentially be used to identify segments with compromised function for patients with preserved global EF after MI. Taken together, our results showed that the REF and RAS are reliable indices that can provide accurate assessment of the regional LV function in patients after first-time MI and offers physicians new insights into the local characteristics of the myocardial tissues after MI.

Funding statement

This work was supported in part by research grants from the Agency for Science, Technology and Research (A*STAR), SERC Biomedical Engineering Programme grant, 092 148 0071 and 132 148 0012.

Footnotes

References