Iterative algorithm for the temporal decomposition of the cerebrovascular reactivity dynamic response in neurovascular patients.
Marco Piccirelli1, Christiaan Hendrik Bas van Niftrik2, Oliver Bozinov2, Athina Pangalu1, Antonio Valavanis1, Luca Regli2, and Jorn Fierstra2

1Department of Neuroradiology, University Hospital Zurich, Zurich, Switzerland, 2Department of Neurosurgery, University Hospital Zurich, Zurich, Switzerland

Synopsis

Our iterative algorithm evaluates transient phases of BOLD fMRI signal dynamics during a CO2 pseudo-square wave challenge and optimizes CO2 arrival time determination to increase sensitivity and reliability of cerebrovascular reactivity analysis. On 25 healthy controls and unilateral internal carotid occlusion patients, all BOLD-derived parameters maps are normalized in MNI space and combined for reference atlases and assess alterations within single patient. The transient phase durations, temporal delay maps and dynamic and static CVR maps were calculated. We determined the optimal CO2 arrival time and found that excluding the transient phases resulted in the best fit to the physiological data.

PURPOSE – “Why was this study/research performed?"

The BOLD cerebrovascular reactivity response to a CO2 challenge of neuro-vascular Patients is clinically relevant and has a complex temporal dynamic. This neurovascular response-function differs greatly from the commonly used canonical HRF. (Poublanc et al., 2015) Therefore, we present here a novel model-free iterative analysis for evaluating transient phase duration parameters describing BOLD fMRI signal dynamics during a monitored CO2 pseudo- square wave challenge. We aim to optimize the CO2 arrival time description to increase sensitivity and reliability of cerebrovascular reactivity analysis, test our algorithm for its clinical relevance on patient data and thereby gain further pathophysiological insight.

METHODS – “How has this problem been studied?”

The algorithm explained in Fig1 is illustrated using datasets of Patients with unilateral internal carotid occlusion and healthy Controls. Secondly, BOLD-fMRI derived normalized maps of 25 healthy subjects in MNI standard space are combined for reference atlases of all BOLD-derived parameters to better assess alterations within a single subject. All subjects underwent a standardized CO2 pseudo square wave increase of ± 8.3±1.6mmHg of partial pressure of end-tidal CO2 (PetCO2) from a calibrated baseline of 40.1±1.2mmHg PetCO2, executed by a custom built computer controlled gas blender (RepirActTM, Thornhill Research Institute, Toronto, Canada) using the prospective gas targeting algorithms. (Slessarev et al., 2007) The iterative algorithm was then applied to the data to calculate both the transient phase durations, temporal delay maps and dynamic and static CVR maps. We compared our algorithm to the state-of-the-art methods used in the literature: the maximum correlation method.

RESULTS – “Principal data and statistical analysis”

BOLD time series vs CO2 time series are presented with different CO2 arrival times applied to the data. We determined the most optimal CO2 arrival time and found that the parametric decomposition resulted in the best description of the physiological data for white and grey matter (Fig2). The linear fit of the BOLD vs. CO2 scatter plot shows clearly that only our algorithm removes the transition phases between the two static states correctly. Our algorithm corrected the too long CVR response delay obtained with the maximum correlation method: WM: 40 s, GM: 10 s; to the more realistic delay of WM 18 s, GM: 2 s. Fig2.4 shows the respective duration of the DTP: Delay to Plateau and of the DTB: Delay to Baseline for GM and WM with the dynamic (CVRcorr) and static (CVRstat) CVR maps. The CVR maps obtained with our algorithm differed significantly from previous CVR maps obtained with not correct delays. The new parametric maps obtained: DTP, DTB have a good SNR and clinically plausible.

DISCUSSION – “What is the interpretation of the data?”

Improved pathophysiological insight could be gained with the decomposition of the dynamic response: the separation of the arterial arrival time of CO2 from the CO2-BOLD relationship improved the sensitivity and reliability of CVR measurements. Further, knowing the transient phase durations allows calculation of a static CVR which can be used to compare the MRI data to static CVR data obtained from the clinical gold-standard imaging modalities like Positron-Emission-Tomography or Xenon enhanced-Computed-Tomography, .(Vagal, et al 2009)

CONCLUSION – “What is the relevance to clinical practice or future research?”

The improved quantification of monitored CVR fMRI data increased sensitivity of DYNAMIC BOLD-CVR measurements by excluding the CO2 arrival time fMRI and might allow the replacement of clinical static CVR measurements.

Acknowledgements

MP and CHBVN: equal first Author contribution

References

Poublanc, J., Crawley, A., Sobzcyk, O., Sam, K., Mandell, D., Venkatravaghan, L., Fisher, J. (2015). Measuring cerebrovascular reactivity: the dynamic response to a step hypercapnic stimulus.

Slessarev, M., Han, J., Mardimae, A., Prisman, E., Preiss, D., Volgyesi, G., Fisher, J. A. (2007). Prospective targeting and control of end-tidal CO2 and O2 concentrations. J Physiol, 581(Pt 3), 1207-1219. doi: 10.1113/jphysiol.2007.129395

Vagal, A. S., Leach, J. L., Fernandez-Ulloa, M., & Zuccarello, M. (2009). The acetazolamide challenge: techniques and applications in the evaluation of chronic cerebral ischemia. AJNR Am J Neuroradiol, 30(5), 876-884. doi: 10.3174/ajnr.A1538

Figures

Flow charts for DTP construction on a voxel-wise basis. The parameter construction uses an iterative analysis

For each processing: shifted CO2 evolution (green) against the BOLD signal of two random voxels (WM: red, GM: blue). Linear fit of BOLD vs. CO2. CVR maps. Panel4 fitting shows clearly removal of the transition phases between the two static states. DTP: Delay to Plateau, DTB: Delay to Baseline

Eight BOLD-derived CVR maps of two healthy subjects (A-B) and of two right ICA occlusion Patients are compared (3C-D). All time constant maps (first and third row) are color-coded from 0 to 80 seconds, while all the CVR maps are presented from -0.6 to 0.6 [%Signal change/mmHg CO2].

Reference Atlases from our healthy control group (n=25). A+B) Mean DTP and DTB values between 0 and 60 seconds. D+E) Mean Delaymaxcorr and Delaycorr values. similar color-coding. C+F) Mean CVRcorr and CVRstat values, color-coded between -0.6 and 0.6 percentage BOLD signal change per mmHg CO2 change.

Parameter maps z-score of two illustrative Patients with right ICA steno-occlusion, MNI normalized. Only z-scores lower than 2 standard deviations (blue color) or higher than 2 standard deviations (brown color) are shown to show abnormality in these maps.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
4379