Introduction

The [Shutter-Speed-Model-Dynamic-Contrast-Enhanced] SSM-DCE-MRI pharmacokinetic analysis adds a metabolic dimension to DCE-MRI^1,2. Changes in the on‑going plasma membrane Na⁺,K⁺-ATPase [NKA] metabolic rate can be detected^1,3,4. This is of particular interest in cancers, since NKA activity seems to increase significantly. However, SSM must be applied thoughtfully, especially in Glioblastoma multiforme [GBM], because of strong tissue vascular heterogeneity across the brain field-of-view. Contrast agent (CA) extravasation is essentially negligible in normal-appearing [NA] cerebral regions with an intact blood-brain-barrier, but significant in the tumors. Here, we present a method to apply the appropriate SSM-DCE-MRI version to analyze such heterogeneous tissue on a pixel-by-pixel basis.

Methods

Recurrent GBM (subject A) and pre-diagnosis GBM (subject B) patients were studied. DCE-MRI was performed within a 3T MRI (Magnetom Skyra, Siemens Healthcare, Erlangen, Germany). 80 frames with 6.81 s per frame, (0.8 × 0.8 × 1.5) mm3 voxel size, TR/TE/flip angle 6.2ms/1.3ms/10^o, were acquired for each patient. The corrected Akaike information criterion automatically separated data for proper SSM versions – pixels with sufficient CA extravasation were analyzed with the full 3-site-2-exchange (3S2X) SSM (SSM_full) model illustrated in Figure 1. SSM_full has potentially five independent physiological-related parameters: p_b and p_o, the vascular and interstitial water mole fractions, k_bo, and k_io, the steady-state water molecule extravasation and cellular efflux rate constants, and K^trans, – the CA extravasation rate constant (k_pe)·plasma volume fraction (v_p) product [i.e., K^trans = k_pev_p = k_epv_o, where k_ep is the CA intravasation rate constant].⁵ For pixels with minimal CA extravasation, a restricted SSM version (SSM_vas) is implemented, in which trans-cytolemmal water exchange, in the vanished SS (VSS) condition, has negligible effects. When accessible, the water exchange process parameters k_io and k_bo, were processed with error analyses to eliminate unreliable estimations.

Results and Discussions

Figure 2A panels show single axial image slices pre-, 1.5 min post-, and 9 min post-CA (left-to-right) of subject A. The zoomed ΔAIC_c [ΔAIC_c ≡ AIC_c(SSM_full) – AIC_c(SSM_vas) ] map in Fig. 2B shows that most pixels in tumor region showed sufficient contrast enhancement and were better matched by SSM_full: the ΔAIC_c values were generally more negative than -10. Most normal-appearing brain (NAB) pixels were better matched by SSMvas. In the three representative tumor single pixels (Figure 3A–3C), systematic SSM_vas fitting bias was observed for Pixels A–C (Fig. 3B). For NAB pixel D, both SSM_full and SSM_vas showed good fittings but SSMfull clearly overfitted the data with positive . In the three tumor region pixels (A–C), the SSM_full AIC_c value is more negative than for the extended Tofts model (eTofts). For NAB pixels, e.g., pixel D, SSM_vas also shows better fitting than eTofts with smaller more negative AIC_c.


For the k_io error analysis, tumor core pixel A shows well-defined error bounds: 2.0 (–0.25/+0.5) s^-1 (Figure 4A). Tumor rim pixel B shows a larger upper error bound: 5.0 (–1.75/+5.25) s^-1. Tumor rim pixel C shows a poorly determined upper bound, but a lower error bound (9.0 s^-1) can still be set. Subject B NAB gray matter (GM) and white matter (WM) pixels and the subject B tumor rim pixel C were selected as representatives for kbo error analysis (Fig. 4B). Most pixels still show well-defined error bounds for 95% confidence level or a well-defined lower bound.

Figure 5 shows zoomed SSM parametric maps of subject A Fig. 2 slice and one slice of subject B. The K^trans values are much larger (and rather homogenous) in extensive tumor than in NA regions in both subjects. Subject B (Fig. 5B), with GBM confirmed by biopsy diagnosis [Dx] only just after the DCE-MRI acquisition, shows smaller K^trans elevation (0.8×10^-2 min^-1 on average) than that of subject A (3.8x10^-2 min^-1 on average, P < 10^-10). Importantly, elevated k_io is seen only in much smaller foci than K^trans, even though it could be accurately estimated in such regions. It exhibits quite different distribution patterns than K^trans, and between the two subjects: large kio in an annular shape [“ring-enhancement”] for A, smaller k_io in a focal “hotspot” for B. These different patterns continued in several contiguous slices in each subject (not shown). Interestingly, the lesion-averaged k_io in the pre-DX GBM (subject B, 0.5 s^-1) is six times smaller than in the recurrent GBM (subject A, 3.1 s^-1, P < 10^-10).

The vascular water efflux rate constant k_bo shows focal enhancements throughout the brain (Fig. 5), but suddenly decrease to near zero, in both subjects, due to neurogliovascular unit NKA activity compromise.⁶ In strong contrast, the mean k_pe* values are two orders smaller than k_bo in NAB, and rise dramatically in going from NAB to tumor (P < 10^-10). These seemingly contradictory results are strong evidence for CA and H₂O extravasating via different molecular pathways and for a kbo NKA contribution.

Conclusion

Within our analysis framework, the three supra-intensive rate constants k_bo, k_io, and k_pe can be reliably determined. Pilot data for one recurrent and one pre-diagnosis GBM patient are presented to demonstrate this. For the pre-diagnostic subject, the combination of k_io, k_bo, and k_pe maps is consistent with a new tumor, but a quite progressed malignancy for the recurrent GBM subject.

Acknowledgements

We gratefully acknowledge the financial support of the National Natural Science Foundation of China (NSFC) Grant (No. 81873894), the Fundamental Research Funds for the Central Universities, the Taishan Scholars Program (No. tsqn20161070). CSS was supported by the Brenden-Colson Center for Pancreatic Care and the OHSU Advanced Imaging Research Center.