Introduction

With further development of imaging techniques and instrumentation, real-time phase contrast sequences (RT-PC) have become possible¹. Compared to conventional phase-contrast sequences, its most significant feature is the quantification of cerebral blood flow or cerebrospinal fluid (CSF) in "real-time". Although not yet applied in clinical practice, recent studies have shown that RT-PC has a great potential for non-invasive quantification of the effects of respiration on cerebral fluids circulation^2-4. However, the fast imaging speed that impairs RT-PC image quality, and a large number of images make it more difficult to post-process RT-PC data. In general, the post-processing of RT-PC data often requires a lot of manual operations and several software to perform the different steps such as region of interest (ROI) segmentation, background field correction, anti-aliasing correction, extraction of flow signals, and finally quantify the effect of respiration on blood flow by comparing respiration and flow signals. To our knowledge, there is no post-processing software specifically dedicated to RT-PC data. Based on the original version⁵, we have developed Flow 2.0 software integrating all the above-mentioned features and allowing a high level of automation for end-to-end quantification of the effect of respiration on cerebral blood flow.

Methods

Flow 2.0 was developed using IDL (Interactive Data Language) and was compiled into a 150Mb executable installation-free program which is available on Windows, Linux and Mac systems. Figure 1 shows the post-processing workflow diagram of the software, it contains two major post-processing steps, the flow quantification (Figure 2) and the quantification of respiratory effects (Figure 3).

• Flow quantification

Data extraction, can import the RT-PC images series and the parametric data from the DICOMDIR file according to the conformance statements of the different manufacturers. ROI segmentation, is consisted of two semi-automatic algorithms. One is based on region growth and active contour algorithms to segment vessels with large deformations or displacements during acquisition such as arteries. A second is based on a fully automatic segmentation algorithm in the frequency domain that can easily identify pixels in which dynamic signals are synchronized with the cardiac frequency. This one is used for ROIs with insignificant displacements such as CSF and venous sinuses. Background field correction, is a necessary step to correct eddy current-induced velocity quantization errors⁶. Typically, a ROI must be drawn manually in the stationary tissue. Our software includes a background field correction algorithm that selects automatically the stationary tissues around the selected ROI. It used the characteristics of the phase and amplitude in the pixel to recognize static tissue. Curves flows, are calculated and displayed in the viewing interface. The main key-parameters of the flow curve (flow rate, stroke volume, areas of the ROI, maximum velocities, …) can be automatically calculated and presented in the viewing interface as well as with ECG and respiratory signals if recorded. In addition, the software also includes anti-velocity aliasing⁵ and image denoising functions.
A frequency domain signal processing tool interface that is useful for other processing functions was also developed. An Export function can save all the results in a text file, easily importable by other software.

• Quantification of respiratory effects

The first step is to separate all the individual Cardiac Cycle Flow Curve (CCFC) of the total flow calculated in the ROI and across the entire images of series. This Signal segmentation, automatically finds the maximum flows systolic points, uses them to cut the continuous flow rate signal into multiple independent CCFCs. Signal reconstruction, the CCFCs are either labelled as expiratory interval or inspiratory interval based on the recorded respiratory signal. Two average expiratory and inspiratory CCFCs are calculated and used to compare the two respiratory periods. Diff_Ex-In (parameter, delay) signal, allows to compare each parameter of CCFC (amplitude, mean flow, stroke volume, cardiac period) between the expiratory and the inspiratory period. We hypothesis that breathing impacts blood and CSF flows with a possible delay between the flow curves and the signal recorded by the respiratory belt sensor. This delay was calculated by applying a phase shift on the respiratory signal interval that produce the maximum Diff_Ex-In for the considered parameter of interest. Extraction results, uses the maximum value of Diff_Ex-In (parameter, delay) to indicate the intensity and delay of the effect of respiration on this parameter of blood or CSF flow.

Results

Figure 4 shows an example of an in vivo application of the software. The automatic algorithm accurately segmented the CSF Roi in the spinal canal. We can see the flow rate signal and the respiratory quantification results.

Discussion & Conclusion

We have developed a post-processing software for RT-PC that integrates the required functions, with a high degree of automation, with low requirements for a priori knowledge of anatomy, high robustness and supporting functional extensions. The software is compatible with the DICOM files generated by the MRI vendors. Our development makes the post-processing of RT-PC substantially easier, while also providing a multivariate approach to analyze the effects of respiration on cerebral fluids circulation.

Acknowledgements

This research was supported by equipEX FIGURES (Facing Faces Institute Guilding Research), European Union Interreg REVERT Project, Hanuman ANR-18-CE45-0014 and Region Haut de France.

Thanks to the staff members at the Facing Faces Institute (Amiens, France) for technical assistance.

Thanks to David Chechin from Phillips industry for his scientific support.