Quantification of myocardial perfusion by MRI requires a wise combination of advanced and complex techniques from images acquisition, segmentation, deconvolution and/or modelling prior deriving perfusion indexes estimations. Accuracy/reproducibility of measures are crucial for medical diagnostic but rely on the optimization of many algorithms. Hence, the latter techniques at each step shall be made available to the community, challenged and improved for optimal state-of-the-art implementation. We propose a software platform that can easily integrate any methods addressing issues or limitations at each step of the process, integrating sharpened tools for precise assessment of individual step to global pipeline performances.
The platform provides a user interface composed of thumbs referred as activities dedicated for specific works listed below:
Phantom generation:
This workbench enables to generate a set of perfusion time intensity/concentration curves from a user defined modelling, a range of perfusion indexes, delay and SNR. The dataset shape is by default squared but can be realistic being full of interest for spatio-temporal based methods3. As many models are available, and many of them have already been developed and validated by Physiome, our platform is connected to jSim4 to use its simulation engine, generating dataset with a strong precision guarantee.
Patient data Preparation:
This workbench is dedicated to pre-processing of clinical datasets acquired with dual-delay saturation recovery TurboFlash sequence5 before processing of perfusion indexes extraction. It includes: coil heterogeneity correction, automated Arterial Input Function (AIF) extraction, manual definition of myocardium, conversion on time intensity curves (TIC) to time concentration curve, semi-automatic AIF fit, and finally the image series display were user can visually check time concentration curve of any voxel of the myocardium.
Myocardium segmentation:
Lesion classification enables myocardial tissue segmentation including simple methods such as bullseye representation6, but also more advanced methods like automated region growing and Spatio Temporal Mean Shift7.
Segmentation results are then displayed, and it is possible to easily check the segmentation mask region by region over image series and even to manually correct possible imperfections, due to under/over segmentation errors of automated algorithms.
Data deconvolution/modeling fit:
This thumb provides a large number of linear shift invariant (LSI) techniques for processing perfusion parameters estimation. Among proposed approaches are Fermi function, and spatio-temporal5 approaches but also modelling fit that can be combined with a previous LSI processing to save processing time and constrain modelling. Because of large amount of data this step is often time consuming but was widely reduced by parallel processing implementation. Architecture was designed to process either synthetic data, produced by phantom generation workbench for assessing techniques performances or on clinical dataset having been preprocessed with “patient data preparation” workbench. Furthermore, in the case of clinical dataset, if segmentation step was processed, the ROI average signal is also deconvolved.
Results Viewer:
Even if previous workbenches provide useful tooltips, perfusion data observation requires more tools to sharpen comprehension. These observations can be enabled by the result viewer. Results produced by previous workbenches can be analyzed at this step for various purposes:
Features display:
In addition to the results viewer, we provide an original way to display features of TIC in a dedicated features space. Available features are: Time To Peak, Peak Value, Maximum Slope, Area Under the Curve, ROI Surface, Maximum Slope Position. The features can be calculated from myocardium voxels or from user defined ROI TIC. This visualization approach is worth for patient myocardial perfusion understanding.