Unlocking cardiac motion: assessing software and machine learning for single-cell and cardioid kinematic insights

Figures

• 

  Figure 1. Example of normalized contraction profile (grey trace) and its first derivative named θ (orange trace) obtained with MUSCLEMOTION (benchmark). This represents a qualitative graphical description of the kinematic parameters calculated and compared during the experiments. The time between points A and E is defined as beat duration. The time between points B and D is the distance between the peaks of contraction and relaxation (t[θc−r]). The time between points A and B corresponds to the interval of maximum speed during the contraction phase (t[θc]). Lastly, the time between C and D is defined as the interval of maximum speed during the relaxation phase (t[θr]).

• 

  Figure 2. Comparison of the three open-source software programs with in-silico cardiomyocytes with 200% time-to-peak. (a) The first experiment with “grey” patterned cardiomyocyte. The plot on the left side of the panel represents the speed profile evaluated by the three software programs. The grey profile is the benchmark considered in this study (MUSCLEMOTION), the cyan profile derives from CONTRACTIONWAVE evaluation, and the orange profile is the ViKiE estimation. The histograms on the right side represent the four parameters extracted by the profiles and measured by the three methodologies. (b) The second experiment with “fog” patterned cardiomyocyte. As in panel (a), the profiles and histograms of the software evaluations are reported to address the ground-truth comparison.

• 

  Figure 3. The frequency response of single cells was measured with the three open-source software programs. Grey: MM, Cyan: CW, Orange: ViKiE. (a) Beat duration of the untreated cells, for each algorithm. X-axis: frequency of stimulation in Hz (s−1). Y-axis: time occurred in milliseconds (ms) (b) The estimation of the time occurring between the double speed peaks. (c) The maximum speed reached during the contraction phase is reported in this panel. (d) The maximum speed reached during the relaxation phase t[θr]. Two-way ANOVA was performed with Tukey statistical hypothesis for multiple comparisons, with significance levels defined *p < 0.05, **p < 0.01, ***p < 0.001 (n = 3).

• 

  Figure 4. Kinematic analysis of single cells treated with caffeine and KCl, normalized for the control mean value. Vertical-themed bars, 10 mmol/L KCl effect. Oblique-themed bars, 10 mmol/L caffeine-treated cells. X-axis: differences of mean values of the parameters. Grey: MM, Cyan: CW, Orange: ViKiE. (a) The four parameters measured at 0.5 Hz pacing frequency. (b) Parameters assessed at 1 Hz pacing. (c) The same measurements at 2 Hz. Two-way ANOVA was performed with Tukey statistical hypothesis for multiple comparison, significance levels defined *p < 0.05, **p < 0.01, ***p < 0.001 (n = 3).

• 

  Figure 5. Frequency response of cardioids measured with the three open-source software programs. Grey: MM, Cyan: CW, Orange: ViKiE. X-axis: the frequency of stimulation in Hz (s−1). Y-axis: time occurred in milliseconds. (a) Beat duration of the untreated cardioids for each algorithm. (b) The estimation of the time occurring between the speed peaks at both contraction and relaxation. (c) The maximum speed reached during the contraction phase. (d) The maximum speed reached during the relaxation phase. Two-way ANOVA was performed with Tukey statistical hypothesis for multiple comparisons, with significance levels defined *p < 0.05, **p < 0.01, ***p < 0.001 (n = 3).

• 

  Figure 6. Kinematic analysis of cardioids treated with caffeine and KCl, normalized with the control mean value. Vertical-themed bars, 10 mmol/L KCl effect. Oblique-themed bars, 10 mmol/L caffeine-treated cells. Grey: MM, Cyan: CW, Orange: ViKiE. (a) All the kinematic parameters evaluated at 0.5 Hz pacing, (b) all the parameters evaluated with cardioids paced at 0.75 Hz. (c) Parameters measured at 1 Hz paced cardioids (n = 3).

• 

  Figure 7. Comparison of performance metrics (Accuracy and TPR) using three different machine learning (ML) algorithms. (a) and (b) ML was applied on single beat parameters from three motion tracking software (MUSCLEMOTION, CONTRACTIONWAVE and ViKiE) in single cells, at the following stimulation frequencies (0.5 Hz, 1 Hz, 2 Hz), and after 10 mmol/L caffeine administration. (c) and (d): same as (a) and (b) but for cardiac spheroids at the following stimulation frequencies (0.5 Hz, 0.75 Hz, 1 Hz). TPR true positive rate. Grey: support vector machine (linear kernel), Cyan: support vector machine (polynomial kernel), Orange: random forest.

• 

  Figure 8. Comparison of performance metrics (Accuracy and TPR) using three different machine learning (ML) algorithms. (a) and (b) ML was applied on single beat parameters from three motion tracking software (MUSCLEMOTION, CONTRACTIONWAVE and ViKiE) in single cells, at the following stimulation frequencies (0.5 Hz, 1 Hz, 2 Hz), and after 10 mmol/L KCl administration. (c) and (d): same as (a) and (b) but for cardiac spheroids at the following stimulation frequencies (0.5 Hz, 0.75 Hz, 1 Hz). TPR true positive rate. Grey: support vector machine (linear kernel), Cyan: support vector machine (polynomial kernel), Orange: random forest.