Open Data and Sources
1) You can download the data used in this study from http://neuropoly.pub/pigHeartsData
- Raw DICOM data (.dcm)
- Data in native Octave/MATLAB (.mat) format
2) You can let MATLAB download the data and repeat the statistical analysis.
Access source code on GitHub: http://neuropoly.pub/pigHeartsSrc
Simply run pigExvivo.m (under MATLABSource folder) main script. This will:
- Download the data hosted on OSF (MATLAB version > R2014b)
- Parse them into a MATLAB struct for easier use
- Perform statistical analysis and export MATLAB generated (static) figures
- Export .mat files, which are provided as input to the Jupyter Notebook (visit repo).
Explore data with interactive figures.
1) How quantitative metrics vary over weeks ?
- Robust correlation toolbox uses an interquartile range (IQR) based bivariate outlier removal. An intuitive explanation for this outlier labeling can be found here.
- Data points detected as outliers (you can repeat this analysis on MATLAB, please see above) are discarded from the boxplot to visualize the effect of outlier removal on the respective univariate distributions.
You can use buttons on the left to navigate through the data.
For example, see how diferent cardiac T1 mapping methods (MOLLI, SHMOLLI and SASHA) consistently over- or underestimate T1 values at all weeks with respect to the reference Inversion Recovery (IR) technique.
2) Voxelwise distribution of myocardial T1 values
We selected a representative sample (Heart #2) to display how voxelwise distribution of myocardial T1 values differs from method to method and to observe their evolution over weeks.
Masked myocardium can be visualized in the Jupyter Notebook.
Observe box and whiskers plots by selecting a week using the buttons on the left.
3) Comparing T1 mapping methods in one figure
You can use the legend on the right of the figure to show/hide traces. Double click on the legend to isolate one trace, double click to reset.
4) Comparison of T1 Bias between MOLLI, SHMOLLI and SASHA before and after fixation.
T1 Bias (ms) values are obtained by subtracting reference IR maps from those acquired by MOLLI, SHMOLLI and SASHA.
5) Association between T1 Bias and MTR and T2 based on phantom acquisitions.
T1 bias (ms) values are obtained by subtracting reference IR maps from those acquired by MOLLI, SHMOLLI and SASHA.
6) Association between T1 Bias and MTR and T2 based on phantom acquisitions.
7) Association between T1 Bias and T2 & T1 Bias and MTR
T1, T2 and magnetization transfer effects are not decoupled from each other. This intertwined interaction is region variant.
Here, we explore the strength of association between the imperfections of cardiac T1 mapping techniques MOLLI, SHMOLLI and SASHA and i) T2 (Fig. 4.1) & ii) MTR (Fig. 4.2)in ex-vivo pig myocardium, by taking IR T1 mapping as a reference.
How to read these figures?
Vertical axis represents the difference between cardiac T1 mapping techniques of [MOLLI, SHMOLLI and SASHA] and reference IR. Horizontal axes traverse T2 (Fig. 4.1.) and MTR (Fig 4.2.) values.
Outliers detected by skipped-correlation are described by red markers. You can hover your cursor over data points to see which week/heart they belong to.
Double click on the respective data marker in the (dark blue) legend bar below (e.g. pre MOLLI) see its distribution. Double click again to reset axes.
The inlier points are encompassed by ellipses. Red ellipses indicate insignificant associations. Significant associations are shown in the color of the respective cardiac T1 mapping method.
Intercept, slope, Pearson’s correlation coefficient (r), p-value and confidence intervals (CI) are displayed at the top of the interactive panels. Note that confidence intervals are taken as a basis for the statistical significance:
If confidence interval includes zero, then the association is statistically insignificant.
8) Association between Native T1 and T2 & Native T1 and MTR