Evaluation of in-line automated deep learning-based slice prescription for cardiac MRI

Author

Lu Mao, Geng Li

Published

October 23, 2025

Statistical analysis

The Bland-Altman analysis was used to assess the agreement of cardiac function parameters between different prescription methods. Intra-operator repeatability and inter-operator reproducibility were evaluated using the repeatability coefficient (RC) and intraclass correlation coefficient (ICC) with 95% confidence intervals (CIs) under a two-way random-effects model for absolute agreement. Quality ratings between different prescription methods and between radiologists were compared using cumulative link mixed models (CLMM) with random effects for TeslaFlow and fixed effects for radiologists. P-values < 0.05 were considered statistically significant. All statistical analyses were performed using R software (version 4.2.2; R Foundation for Statistical Computing, Vienna, Austria).

Human Analyst (Old)

Bland-Altman Analysis

Table 1 below shows the Bland-Altman analysis of cardiac function parameters between different prescription methods (Table 2 in manuscript).

Table 1: Bland-Altman analysis of cardiac function parameters between different prescription methods.
Comparison Parameter lower LOA mean diff upper LOA LOA width
TF vs M EDV -12.5 2.9 18.3 30.7
ESV -15.0 -2.0 10.9 25.8
SV -13.3 5.0 23.2 36.5
CO -1.2 0.4 1.9 3.1
EF -6.6 1.8 10.2 16.8
MyoMass_diast -7.9 -0.4 7.2 15.1
MyoMass_syst -8.2 -0.4 7.3 15.4
HR -7.1 0.5 8.0 15.1
TF+M vs M EDV -16.8 -0.7 15.5 32.3
ESV -12.7 -1.9 8.9 21.6
SV -17.4 1.2 19.8 37.2
CO -1.4 0.1 1.5 2.9
EF -6.8 0.9 8.6 15.5
MyoMass_diast -7.1 0.6 8.2 15.2
MyoMass_syst -6.9 0.5 7.9 14.9
HR -6.3 -0.1 6.0 12.3

The figures below show the Bland-Altman plots for different cardiac function parameters between different prescription methods (“Table 3” in manuscript). (Right click on figure to save.)

Intra-Operator repeatability and Inter-Operator Reproducibility

Table 2 and Table 3 below summarize the intra-operator repeatability and inter-operator reproducibility of cardiac function parameters for different prescription methods (Table 4 in manuscript).

  • Please update Table 4 in manuscript with the following for consistency.
Table 2: Intra-operator repeatability of cardiac function parameters for different prescription methods.
Parameter Prescription RC ICC 95% CI Bt-Subj SD Wt-Subj SD
EDV M 14.9 0.973 (0.913, 0.992) 31.9 5.4
TF 22.2 0.948 (0.829, 0.985) 34.2 8.0
TF+M 14.7 0.973 (0.914, 0.992) 31.8 5.3
ESV M 12.3 0.934 (0.8, 0.98) 16.8 4.4
TF 8.2 0.970 (0.898, 0.991) 16.7 3.0
TF+M 9.5 0.960 (0.874, 0.988) 16.8 3.4
SV M 14.9 0.924 (0.77, 0.976) 18.7 5.4
TF 26.2 0.798 (0.434, 0.938) 18.8 9.4
TF+M 13.9 0.926 (0.776, 0.977) 17.7 5.0
MyoMass_diast M 7.2 0.987 (0.959, 0.996) 22.9 2.6
TF 7.5 0.987 (0.957, 0.996) 23.8 2.7
TF+M 6.6 0.989 (0.965, 0.997) 22.7 2.4
Table 3: Inter-operator reproducibility of cardiac function parameters for different prescription methods.
Parameter Prescription ICC 95% CI Bt-Subj SD Wt-Subj SD
EDV M 0.971 (0.921, 0.989) 35.7 6.2
TF 0.932 (0.819, 0.975) 34.4 9.3
TF+M 0.973 (0.925, 0.99) 36.1 6.0
ESV M 0.892 (0.722, 0.959) 19.4 6.7
TF 0.878 (0.687, 0.953) 18.2 6.8
TF+M 0.913 (0.764, 0.967) 19.5 6.0
SV M 0.829 (0.578, 0.934) 19.3 8.7
TF 0.695 (0.315, 0.877) 18.0 11.9
TF+M 0.862 (0.642, 0.948) 20.2 8.1
MyoMass_diast M 0.953 (0.872, 0.982) 23.4 5.2
TF 0.975 (0.93, 0.991) 23.5 3.8
TF+M 0.951 (0.864, 0.982) 23.3 5.3

Radiologist rating: quality of prescription

Table 4 below shows the p-values for comparisons of quality ratings between different prescription methods and between radiologists.

Table 4: P-values for comparisons of quality ratings between different prescription methods and between radiologists.
parameter M vs TF+M M vs TF TF+M vs TF Radiologist
SAX 0.688 0.048 0.028 0.004
4CH 0.084 0.341 0.009 <0.001
3CH <0.001 0.239 <0.001 <0.001
2CH 0.001 <0.001 0.422 <0.001

AI Analyst (New)

Bland-Altman Analysis

There are duplicate records in Teslaflow 21, Teslaflow 30 for AI analyst data, we only keep the first record for each Patient, Exam, and prescription combination.

Table 5 below shows the Bland-Altman analysis of cardiac function parameters between different prescription methods for AI analyst.

Table 5: Bland-Altman analysis of cardiac function parameters between different prescription methods for AI analyst.
Comparison Parameter lower LOA mean diff upper LOA LOA width
TF vs M EDV -13.0 1.1 15.2 28.2
ESV -17.1 -2.6 11.9 29.0
SV -14.6 3.7 21.9 36.5
CO -1.3 0.3 1.9 3.2
EF -7.8 1.6 11.0 18.8
MyoMass_diast -10.8 -0.3 10.2 20.9
MyoMass_syst -15.7 -2.8 10.0 25.7
HR -7.5 0.6 8.6 16.1
TF+M vs M EDV -16.7 0.8 18.3 35.0
ESV -12.8 -1.7 9.5 22.2
SV -16.9 2.4 21.8 38.7
CO -1.5 0.1 1.7 3.2
EF -6.5 1.1 8.8 15.3
MyoMass_diast -11.4 0.5 12.5 24.0
MyoMass_syst -10.3 -0.6 9.1 19.4
HR -6.7 0.0 6.7 13.5

The figures below show the Bland-Altman plots for different cardiac function parameters between different prescription methods for AI analyst.

Intra-Operator repeatability and Inter-Operator Reproducibility

Table 6 and Table 7 below summarize the intra-operator repeatability and inter-operator reproducibility of cardiac function parameters for different prescription methods for AI analyst.

Table 6: Intra-operator repeatability of cardiac function parameters for different prescription methods for AI analyst.
Parameter Prescription RC ICC 95% CI Bt-Subj SD Wt-Subj SD
EDV M 16.8 0.967 (0.895, 0.99) 32.7 6.1
TF 19.5 0.957 (0.859, 0.988) 33.4 7.0
TF+M 14.8 0.973 (0.913, 0.992) 31.8 5.3
ESV M 15.6 0.905 (0.719, 0.97) 17.4 5.6
TF 15.2 0.898 (0.685, 0.97) 16.3 5.5
TF+M 10.9 0.948 (0.839, 0.984) 16.8 3.9
SV M 20.4 0.846 (0.568, 0.951) 17.3 7.4
TF 19.9 0.876 (0.624, 0.963) 19.1 7.2
TF+M 15.6 0.912 (0.738, 0.972) 18.1 5.6
MyoMass_diast M 7.6 0.988 (0.962, 0.996) 25.3 2.7
TF 12.0 0.970 (0.898, 0.991) 24.5 4.3
TF+M 8.7 0.983 (0.945, 0.995) 23.7 3.1
Table 7: Inter-operator reproducibility of cardiac function parameters for different prescription methods for AI analyst.
Parameter Prescription ICC 95% CI Bt-Subj SD Wt-Subj SD
EDV M 0.981 (0.945, 0.993) 34.6 4.9
TF 0.933 (0.815, 0.975) 36.5 9.8
TF+M 0.961 (0.887, 0.986) 37.6 7.6
ESV M 0.911 (0.76, 0.967) 18.8 5.9
TF 0.942 (0.839, 0.978) 17.7 4.4
TF+M 0.956 (0.872, 0.984) 20.4 4.4
SV M 0.875 (0.671, 0.953) 18.4 7.0
TF 0.812 (0.528, 0.928) 20.0 9.6
TF+M 0.855 (0.612, 0.946) 19.3 8.0
MyoMass_diast M 0.972 (0.922, 0.99) 24.6 4.2
TF 0.970 (0.916, 0.989) 24.3 4.3
TF+M 0.970 (0.911, 0.989) 25.8 4.6

Agreement between Human and AI Analysts

Table 8 below summarizes the inter-analyst reproducibility of cardiac function parameters between Human and AI analysts.

Table 8: Inter-analyst reproducibility of cardiac function parameters between Human and AI analysts.
Parameter Prescription ICC 95% CI Bt-Subj SD Wt-Subj SD
EDV M 0.972 (0.942, 0.987) 35.3 6.0
TF 0.974 (0.945, 0.988) 37.0 6.1
TF+M 0.975 (0.948, 0.988) 36.8 5.9
ESV M 0.953 (0.902, 0.977) 19.9 4.4
TF 0.962 (0.921, 0.982) 19.0 3.8
TF+M 0.966 (0.929, 0.984) 19.4 3.6
SV M 0.877 (0.757, 0.94) 18.8 7.0
TF 0.928 (0.852, 0.966) 21.1 5.9
TF+M 0.917 (0.832, 0.96) 20.1 6.0
CO M 0.907 (0.813, 0.955) 1.4 0.5
TF 0.941 (0.878, 0.972) 1.6 0.4
TF+M 0.919 (0.836, 0.961) 1.4 0.4
EF M 0.727 (0.5, 0.861) 5.0 3.1
TF 0.776 (0.575, 0.889) 4.5 2.4
TF+M 0.818 (0.65, 0.91) 4.7 2.2
MyoMass_diast M 0.929 (0.856, 0.966) 24.8 6.8
TF 0.933 (0.861, 0.968) 24.2 6.5
TF+M 0.939 (0.875, 0.971) 25.4 6.5
MyoMass_syst M 0.901 (0.802, 0.952) 25.8 8.5
TF 0.927 (0.851, 0.966) 25.4 7.1
TF+M 0.911 (0.822, 0.957) 25.7 8.0
HR M 0.998 (0.996, 0.999) 12.0 0.5
TF 1.000 (1, 1) 12.4 0.0
TF+M 1.000 (1, 1) 12.0 0.0