Reference standard, study duration
In total, we prospectively recruited 40 patients with chest pain. Of these 40 patients 6 were found to have ACS (n=3 non-ST-elevation myocardial infarction), another 6 had an underlying cardiac, non-ACS disease (n=4 hypertrophy; n=2 hypertrophic cardiomyopathy) whilst the remaining 28 were determined to have no cardiac cause of the chest pain. Patient characteristics are depicted in Table 2. Gender was evenly distributed (n=20 female; n=20 male) with a mean age of 57.1± 17.7 years.
All CMR scans were performed with a mean study time of 19.5 ± 5.3 minutes, including patient preparation and scan time.
ROC Curve Analysis
ROC curves were drawn for differentiation between cardiac illness (6 ACS patients and 6 patients with underlying cardiac disease) from non-cardiac chest pain (n=28).
Figure 2 ROC curve: identification of cardiac illness (ACS n=6/ cardiac, non-ACS n=6) (GCS-fSENC AUC:0.899, GLS-fSENC AUC: 0.882, LVLAS AUC: 0.771, GCS-FT AUC: 0.688, GLS-FT AUC: 0.740)
GCS-fSENC proved to be the strongest parameter for identification of cardiac illness within the study population (AUC: 0.899) closely followed by GLS-fSENC (AUC: 0.882). Notably, LVLAS achieved good results with an AUC of 0.771, whilst FT strain values demonstrated the weakest performance (GCS-FT AUC: 0.688, GLS-FT AUC: 0.740). GCS-FT differed significantly from GCS- and GLS-fSENC curves (GCS-FT vs. GCS-fSENC p<0.025; GCS-FT vs. GLS-fSENC p<0.035), whereas for the other parameters the difference was not significant. (Figure 2)
Triage Analysis
In a further analysis patients were triaged according to diagnosis (0: non-cardiac/ 1: ACS/ 2: cardiac-non-ACS) and the GLS and GCS (fSENC, FT) as well as LVLAS compared between the three patient groups. Results are depicted in Figure 3 and Table 3.
All strain parameters could significantly (fSENC/ FT: p<0.005; LVLAS: p<0.02) differentiate between non-cardiac (0) and underlying cardiac disease patients (2).
Whilst GCS-fSENC (p<0.0025), GLS-fSENC (p<0.025) and LVLAS (p<0.05) all allowed for significant differentiation between non-cardiac (0) and ACS patients (1), further separation between ACS (group 1) and other cardiac diseases (group 2) was only possible with GLS-fSENC (p<0.006). GCS- and GLS-FT whilst not allowing for distinction between non-cardiac (0) and ACS (1) patients, were significantly different between ACS (1) and cardiac, non-ACS (2) patients (p<0.02).
Table 2 Patient characteristics.
Total: 40
|
|
count
|
mean ( SD)
|
max/min
|
Sex
|
female
|
20
|
|
|
male
|
20
|
|
|
Age (years)
|
|
|
57.1± 17.7
|
84/23
|
BMI (kg/m2)
|
|
|
26.43.7
|
34.4/18.9
|
BP (systolic) (mmHg)
|
|
|
15823
|
204/117
|
HR (bpm)
|
|
|
7414
|
104/43
|
HEART score
|
low
|
|
14
|
|
intermediate
|
|
26
|
|
NYHA
|
1
|
32 (80%)
|
|
|
2
|
3 (7.5%)
|
|
|
3
|
5 (12.5%)
|
|
|
4
|
0 (0%)
|
|
|
EF (%)
|
|
|
72.411.6
|
|
EDV (ml)
|
|
|
114.644.1
|
|
ESV (ml)
|
|
|
32.522.7
|
|
Diabetes
|
|
2 (5%)
|
|
|
Hypertension
|
|
18 (45%)
|
|
|
Hypercholesterinemia
|
|
9 (22.5%)
|
|
|
Familial predisposition
|
|
12 (30%)
|
|
|
nicotine (py)
|
non-smoker
|
22 (55%)
|
00
|
0/0
|
past smoker
|
13 (32.5%)
|
19.515.4
|
45/2
|
smoker
|
5 (12.5%)
|
17.812.8
|
45/4
|
hscTnT 0h (ng/L)
|
|
|
10.87.0
|
32/5
|
hscTnT 1h (ng/L)
|
|
|
15.919.9
|
88/3
|
Diagnostic procedures
|
stress ECG
|
2 (5%)
|
|
|
echocardiography
|
2 (5%)
|
|
|
standard CMR
|
1 (2.5%)
|
|
|
CT angiography
|
1 (2.5%)
|
|
|
coronary angiography
|
11 (27.5%)
|
|
|
max: maximum, min: minimum, SD: standard deviation, BMI: body mass index, BP: blood pressure, HR: heart rate, NYHA: New York Heart Association, EF: ejection fraction, ESV: End-systolic volume, EDV: End-diastolic volume, py: pack years, h: hours, ACS: acute coronary syndrome, hscTNT: high-sensitive cardiac troponin T, ECG: electrocardiogram, CMR: cardiovascular magnetic resonance, CT: computed tomography
Table 3 Mean ± standard deviation (SD) with 95% confidence interval (CI) and p-values for all deformation parameters within total study population.
|
FT
|
fSENC
|
GLS (%)
|
-15.47 ± 3.63
(95% CI -16.63 - -15.47; p<0.001)
|
-17.82 ± 3.25
(95% CI -16.93 - -18.70; p<0.001)
|
GCS (%)
|
-19.11 ± 3.99
(95% CI -20.39 - -17.84; p<0.001)
|
-17.22 ± 5.53
(95% CI -15.71 - -18.73; p<0.001)
|
LVLAS
|
-13.42 ± 3.87
(95% CI -12.18 - -14.65; p<0.001)
|
Correlation
Table 4 Pearson’s correlation coefficient for all deformation parameters. ** p<0.005; * p<0.05
|
GCS-FT
|
GLS-FT
|
LVLAS
|
GCS-fSENC
|
GLS-fSENC
|
GCS-FT
|
1
|
0.754**
|
0.330*
|
0.426**
|
0.468**
|
GLS-FT
|
0.754**
|
1
|
0.476**
|
0.566**
|
0.639**
|
LVLAS
|
0.330*
|
0.476**
|
1
|
0.506**
|
0.548**
|
GCS-fSENC
|
0.426**
|
0.566**
|
0.506**
|
1
|
0.686**
|
GLS-fSENC
|
0.468**
|
0.639**
|
0.548**
|
0.686**
|
1
|
Table 5 Intraclass correlation coefficient for all deformation parameters. ** p<0.005; * p<0.05
|
GCS-FT
|
GLS-FT
|
LVLAS
|
GCS-fSENC
|
GLS-fSENC
|
GCS-FT
|
1
|
0.857**
|
0.496*
|
0.576**
|
0.633**
|
GLS-FT
|
0.857**
|
1
|
0.644**
|
0.711**
|
0.779**
|
LVLAS
|
0.496*
|
0.644**
|
1
|
0.653**
|
0.705**
|
GCS-fSENC
|
0.576**
|
0.711**
|
0.653**
|
1
|
0.806**
|
GLS-fSENC
|
0.633**
|
0.779**
|
0.705**
|
0.806**
|
1
|
Table 6 Coefficient of variation (CoV) for all deformation parameters (%).
|
CoV (%)
|
GCS-FT vs. GCS-fSENC
|
21.53
|
GLSL-FT vs. GLS-fSENC
|
18.18
|
GLS-FT vs. LVLAS
|
26.62
|
GLS-fSENC vs. LVLAS
|
22.33
|
Correlation coefficients (Pearson and ICC) for the different myocardial deformation parameters are given in Table 4, 5 and 6. Pearson’s correlation coefficient was notably strong (>0.5) between GLS-FT and GLS-fSENC as well as between GCS-/ GLS-fSENC and LVLAS. The ICC values were good (>0.75) between GLS-FT and GLS-fSENC. All correlation values were statistically significant (p<0.05). Linear regression analyses are depicted graphically in scatter plots (Figure 4) showing a weak linear relationship between GLS as derived by fSENC or FT and compared to the LVLAS (R2<0.5). The correlation was strongest, albeit weak, between GLS-fSENC and GLS-FT (R2=0.408).
Bland-Altman plots revealed similar levels of variability for GLS (fSENC, FT) and LVLAS (CoV 22.33%; CoV 26.62%). Variability was lowest between GLS as derived by fSENC compared to FT (CoV 18.18%).
Inter-observer-/ Intra-observer Variability Feature Tracking
Table 7 Pearson’s correlation coefficient/ Intraclass correlation coefficient for GCS as derived by reader 1 (intra-abserver reliability) and reader 2 (inter-observer reliability) as well as AI (artificial intelligence). ** p<0.005; * p<0.05
|
GCS
|
GCS R1
|
GCS R2
|
GCS AI
|
GCS
|
1
|
0.905**/ 0.949**
|
0.968**/ 0.984**
|
0.983**/ 0.987**
|
GCS R1
|
0.905**/ 0.949**
|
1
|
0.903**/0.947**
|
0.874**/ 0.932**
|
GCS R2
|
0.968**/0.984**
|
0.903**/ 0.947**
|
1
|
0.976**/ 0.982**
|
GCS AI
|
0.983**/ 0.987**
|
0.874**/ 0.932**
|
0.976**/ 0.982**
|
1
|
Table 8 Pearson’s correlation coefficient/ Intraclass correlation coefficient for GLS as derived by reader 1 (intra-abserver reliability) and reader 2 (inter-observer reliability) as well as AI (artificial intelligence). ** p<0.005; * p<0.05
|
GLS
|
GLS R1
|
GLS R2
|
GLS AI
|
GLS
|
1
|
0.942**/ 0.969**
|
0.937**/ 0.967**
|
0.936**/ 0.966**
|
GLS R1
|
0.942**/ 0.969**
|
1
|
0.890**/ 0.941**
|
0.923**/ 0.960**
|
GLS R2
|
0.937**/ 0.967**
|
0.890**/ 0.941**
|
1
|
0.936**/ 0.966**
|
GLS AI
|
0.936**/ 0.966**
|
0.923**/ 0.960**
|
0.936**/ 0.966**
|
1
|
Table 9 Coefficient of variation (CoV) for intra- (reader 1 R1) and inter-observer (reader 2 R2) reliability of GLS and GCS values derived by FT. Additional CoV between original GLS and GCS values and those derived by artificial intelligence (AI) tools.
|
CoV (%)
|
GLS-FT vs. GLS-FT-R1
|
8.16
|
GLS-FT vs. GLS-FT-R2
|
8.35
|
GLS-FT vs. GLS-FT-AI
|
8.47
|
GCS-FT vs. GCS-FT-R1
|
10.99
|
GCS-FT vs. GCS-FT-R2
|
6.36
|
GCS-FT vs. GCS-FT-AI
|
5.32
|
Intra- and inter-observer reliability was excellent for FT (Pearson and ICC >0.85). Strain values derived by automated contours using AI tools were similarly reproducible (Pearson and ICC >0.85). All data were statistically highly significant (p<0.005). Correlation, limits of agreement (LoA), biases and coefficient of variation (CoV) are depicted in Figure 5, 6 and Table 7-9. Variation was lowest between GCS-FT compared to GCS-FT-AI (CoV 5.32%).
Inter-observer-/ Intra-observer Variability fSENC
Table 10 Pearson’s correlation coefficient/ Intraclass correlation coefficient for GCS as derived by reader 1 (intra-abserver reliability) and reader 2 (inter-observer reliability). ** p<0.005; * p<0.05
|
GCS
|
GCS R1
|
GCS R2
|
GCS
|
1
|
0.876**/ 0.934**
|
0.954**/ 0.965**
|
GCS R1
|
0.876**/ 0.934**
|
1
|
0.859**/ 0.914**
|
GCS R2
|
0.954**/ 0.965**
|
0.859**/ 0.914**
|
1
|
Table 11 Pearson’s correlation coefficient/ Intraclass correlation coefficient for GLS as derived by reader 1 (intra-abserver reliability) and reader 2 (inter-observer reliability). ** p<0.005; * p<0.05
|
GLS
|
GLS R1
|
GLS R2
|
GLS
|
1
|
0.889**/ 0.927**
|
0.955**/ 0.971**
|
GLS R1
|
0.889**/ 0.927**
|
1
|
0.818**/ 0.898**
|
GLS R2
|
0.955**/ 0.971**
|
0.818**/ 0.898**
|
1
|
Table 12 Coefficient of variation (CoV) for intra- (reader 1 R1) and inter-observer (reader 2 R2) reliability of GLS and GCS values derived by fSENC.
|
CoV (%)
|
GLS-fSENC vs. GLS-fSENC-R1
|
6.80
|
GLS-fSENC vs. GLS-fSENC-R2
|
4.36
|
GCS-fSENC vs. GCS-fSENC-R1
|
12.43
|
GCS-fSENC vs. GCS-fSENC-R2
|
8.29
|
Intra- and inter-observer reliability for fSENC was excellent and comparable to that of FT (Pearson and ICC >0.8). All data were statistically highly significant (p<0.005). Correlation, limits of agreement (LoA), biases and coefficient of variation (CoV) are depicted in Figure 7 and Table 10-12. Variation was lowest for GLS-values (vs. R1 6.80%; vs. R2 4.36%).
Inter-observer-/ Intra-observer Variability LVLAS
Table 13 Pearson’s correlation coefficient/ Intraclass correlation coefficient for LVLAS as derived by reader 1 (intra-abserver reliability) and reader 2 (inter-observer reliability). ** p<0.005; * p<0.05
|
LVLAS
|
LVLAS R1
|
LVLAS R2
|
LVLAS
|
1
|
0.750**/ 0.850**
|
0.686**/ 0.804**
|
LVLAS R1
|
0.750**/ 0.850**
|
1
|
0.938**/ 0.968**
|
LVLAS R2
|
0.686**/ 0.804**
|
0.938**/ 0.968**
|
1
|
Table 14 Coefficient of variation (CoV) for intra- (reader 1 R1) and inter-observer (reader 2 R2) reliability of LVLAS.
|
CoV (%)
|
LVLAS vs. LVLAS-R1
|
25.19
|
LVLAS vs. LVLAS-R2
|
29.57
|
Whilst intra-observer reliability was slightly higher (Pearson >0.7, ICC>0.8) than inter-observer reliability (Pearson>0.65, ICC>0.8), the LVLAS showed lower levels of correlation than FT or fSENC-derived strain values. The correlation data was highly significant (p<0.005). Correlation, limits of agreement (LoA), biases and coefficient of variation (CoV) are depicted in Figure 8 and Table 13 and 14. It is evident that variation was markedly higher for LVLAS as compared to FT or fSENC with CoV > 25%.