Previous studies already demonstrated that a non-invasive collateral flow assessment is able to predict MCA-M1 recanalization after IV thrombolysis [7–9].
The current study demonstrates that a similar prediction is possible in patients undergoing endovascular thrombectomy. Indeed, in this series, the criterion combining baseline DWI lesion volume and volume of tissue with arterial-tissue delay over 6 s is significantly correlated with the angiographic collateral grading, and significantly discriminates recanalizers from non recanalizers after thrombectomy with the MERCI retriever.
This is the first step toward a non-invasive prediction of recanalization after thrombectomy. It provides the potential opportunity to improve the penumbral imaging-based selection of patients for thrombectomy.
Indeed, the negative results of all the recent therapeutic trials for thrombectomy not only suggest that the development of more effective endovascular devices is imperative but also that better selection of the patients could be a new strategy for future trials [23–26]. The IMS III and SYNTHESIS Expansion studies, that did not use mismatch-based imaging selection of patients, show that intravenous thrombolysis is still the first-line treatment within 4.5 h after ischemic stroke onset, even if imaging shows an occluded major intracranial artery [23, 24]. Beyond 4.5 h, the MR RESCUE trial, whose sample size is relatively small, does not provide data supporting the use of endovascular treatment in patients with an ischemic penumbra as defined in this study . However, all these trials used older retrieval devices with a significantly lower recanalization rate compared with recent stent retrievers [20, 21]. Thus, the ischemic penumbra hypothesis tested in MR RESCUE should be retested in a larger randomized trial with more efficient recent retrievers . Moreover, a recent study also finds another explanation for negative results in the MR RESCUE study. Indeed, authors of this study demonstrated that, in patients treated with IV thrombolysis and/or endovascular thrombectomy, prediction of tissue fate using CT-based penumbral imaging cannot predict clinical outcome without combining prediction of recanalization . Similarly, patients with a MRI-derived target mismatch profile, hallmarking penumbra, have a more favorable clinical outcome if reperfusion is achieved after thrombectomy . However, if recanalization is not achieved, the penumbral information becomes somewhat futile.
Thus, the prediction of recanalization and ultimately reperfusion is becoming a new challenge in the field of non-invasive multimodal imaging aimed to improve the selection of patients for recanalization therapies .
Choice of the angiographic revascularization grading
In the SWIFT randomized controlled trial, a successful recanalization was defined as TICI 2 or greater reperfusion (Thrombolysis In Cerebral Infarction) . However, in the current study, the AOL scoring was preferred to evaluate the arterial patency at the end of the procedure. AOL and TICI scores are strongly correlated . Moreover, measures performed after the end of the thrombectomy procedure in the SWIFT trial, including any rescue therapy, demonstrated that AOL 2–3, TIMI 2–3 and TICI 2b–3 scores did not differ significantly in predicting good neurological outcome (mRS 0–2 or equal to baseline mRS if baseline mRS was greater than 2, or improvement of at least 10 points in NIHSS score) (57, 62.9, 61.8 %, respectively; AOL versus TIMI p = 0.31, AOL versus TICI p = 0.095) .
In addition, in the SWIFT study, ratios of successful recanalization rate over good neurological outcome rate are very close in patients treated with the SOLITAIRE device versus those treated with the MERCI device (1.19 (69/58) and 0.91 (30/33), respectively) . Thus, the prediction of a good clinical outcome using the recanalization criterion seems independent of the thrombectomy device used which highlights the interest of this prediction in the selection of patients for thrombectomy. The aim of the current study is to perform a non invasive prediction of recanalization through the assessment of the collateral flow in patients mostly treated with the MERCI retrieval device.
Correlations between MRI-derived parameters and the angiographic collateral flow grading
All tested parameters are significantly correlated with the ACG. However, VolATD6 and baseline DWI lesion volume perform best (Fig. 1). Moreover, their corresponding optimal thresholds tend to more accurately detect patients with excellent collateral flow compared with optimal threshold of VolATD + 1.5SD (Table 3). The Youden’s index calculation confirms higher discriminative performances of VolATD6 and VolDWI thresholds by comparison with VolATD + 1.5 threshold (Youden’s index value = 0.67, 0.66, 0.48 respectively). Trends to discriminate recanalizers versus non recanalizers are also better for VolATD6 and baseline DWI lesion volume (Table 4). Therefore, only the combination of these two parameters was tested in an attempt to improve the MRI prediction of MCA recanalization after thrombectomy.
Predictive value of baseline DWI lesion volume on MCA recanalization
Raychev et al. recently reported that the baseline volume of DWI abnormalities was an independent predictor of recanalization after thrombectomy in a cohort of 105 patients with ICA and MCA occlusion treated with multimodal mechanical device strategies (MERCI ± Penumbra ± angioplasty and stenting) (logistic regression analysis, OR = 0.238, p = 0.046) .
In the current series including a smaller cohort of 57 patients with MCA ± ICA occlusion mostly treated with the MERCI retriever ± IVT, the baseline DWI lesion volume threshold >15 ml, accurately discriminating angio grades 0–2 versus 3–4, only shows a trend to discriminate recanalizers versus non recanalizers (Fisher’s exact test, p = 0.056 in the whole cohort) (Table 4).
Similarly, Olivot et al.  reported that patients with full recanalization after endovascular therapy for MCA and/or ICA occlusion (SNARE (eV3) or SOLITAIRE (Covidien) devices, ±combined IV or IA rtPA) tend to have a smaller baseline median DWI lesion volume versus patients with partial or no recanalization (10, 21, and 19 ml respectively; p = 0.07). In this series, the largest DWI lesion volume associated with a favorable outcome in non recanalizers was relatively small and equal to 20 ml . Although Olivot et al. did not report the collateral flow status in these patients , their observations are in agreement with the significant relationship between the baseline DWI lesion volume and the ACG demonstrated in the current study. Indeed, one could expect that patients with MCA occlusion associated with an excellent collateral flow, as suggested by a small baseline DWI lesion volume, might have a favorable outcome in spite of a persisting MCA occlusion.
Interestingly, in patients with similar arterial occlusions treated with IVT, Nicoli et al.  and Nighogossian et al.  reported a significantly smaller initial DWI lesion volume in recanalizers (mean value = 13 ml and median value = 13 ml, respectively) versus non recanalizers (mean value = 23.4 ml and median value = 48 ml, respectively).
Thus, whether MCA recanalization is obtained after IVT [7, 30] or thrombectomy  recanalizers have a relatively small baseline DWI lesion volume (# <20 ml). In addition, this range of baseline DWI lesion volume is significantly correlated with very good collateral perfusion.
Predictive value of Bayesian MR-PWI-derived collateral flow index on MCA recanalization
The positive impact of the degree of collateral flow on MCA recanalization after thrombectomy has already been demonstrated [1, 2]. However, in the current series, despite a significant correlation with ACG and its very good diagnostic accuracy for discriminating patients with or without very good collateral flow (AUC = 0.84), the baseline volume of tissue with ATD > 6 s. only shows a nonsignificant trend to predict MCA M1 recanalization after thrombectomy (Fisher’s exact test, p = 0.111 in the whole cohort, p = 0.093 in the MERCI cohort) (Table 4); whereas the collateral grading alone, performed later during angiography, is able to accurately discriminate recanalizers from nonrecanalizers. Further investigations are required to determine if the discrepancy between the prediction of MCA recanalization by VolATD6 versus ACG is related to an insufficient accuracy of the PWI-based collateral flow assessment, and/or to fluctuations in hemodynamic conditions and collateral supply during the MRI-to-angiography time that would decrease the predictive value of the VolATD6 measurement on later ACG and recanalization.
Added value of DWI and PWI to predict recanalization after thrombectomy
In the current series, the combination of VolATD6 and DWI lesion volume measurements significantly discriminates recanalizers from nonrecanalizers, whether patients are treated with the MERCI retriever only or not (Table 4). These findings suggest that these two parameters, significantly correlated with the ACG, provide complementary and non redundant information about the degree of collateral flow and its ability to increase the full MCA recanalization rate. The highest sensitivity of VolATD6 optimal threshold (100 %) combined with the highest specificity of VolDWI optimal threshold (85.7 %) for diagnosing very good collateral flow may contribute to the better prediction of full MCA-M1 recanalization. Indeed, the post hoc analysis of the correlation between these two MR-derived collateral flow indices and the ACG (0–1 vs 2 vs 3–4) demonstrates that this correlation with the ACG is higher for the combined criterion (VolATD6 > 27 ml: Pearson’s chi square = 12.75, Cramer’s V = 0.47, p = 0.0017; DWI > 15 ml: Pearson’s chi square = 13.92, Cramer’s V = 0.49, p = 0.0009; combined criterion: Pearson’s chi square = 18.42, Cramer’s V = 0.57, p < 0.0001). In addition, at the time of the MRI examination, the VolATD6 measurement provides an instantaneous estimation of the degree of collateral flow while the DWI lesion volume corresponds to a tissue marker for the efficiency of collateral perfusion to preserve tissue from ischemia during the time to MRI. Thus, the combined criterion increases the MRI information content on the degree of collateral flow and facilitates discrimination of recanalizers and nonrecanalizers.
Moreover, like the angiographic collateral grading, this combined criterion is an independent predictor of MCA recanalization after thrombectomy with the MERCI retriever.
Impact of the combined therapy on post-thrombectomy recanalization
In patients with MCA-M1 occlusion treated with IVT within the 3-h time window, the vigor of the collateral flow is predictive of a full MCA recanalization . Beyond 3 h, collateral circulation does not significantly potentiate the action of IV tPA on the clot, probably because the longer a clot persists over time, the more resistant to fibrinolysis it becomes . However, even when failing to recanalize the occluded artery, IVT performed before thrombectomy could be able to (1) limit the thrombus extension by delivering rtPA to proximal and distal parts of the clot via efficient collaterals  and (2) make the mechanical thrombectomy easier thanks to a relatively lower clot size compared with thrombectomy alone.
Thus, the relationship between collateral flow and the success of the thrombectomy procedure could only be driven by IV rtPA. Interestingly, this bridging therapy was applied in 33 to 58 % of the cases treated with a stent retriever in SWIFT and TREVO II trials .
However, in the current series, only 16 patients were treated with IVT and the MERCI retriever making difficult to definitely conclude about the rational for the influence of the vigor of the collateral flow on MCA recanalization in patients treated with bridging therapy versus thrombectomy alone. The frequency of this combined therapy was not statistically different in recanalizers versus nonrecanalizers (31.8 % (7/22) and 32.1 % (9/28) respectively). In addition, the multivariate analysis including the variable “combined therapy” demonstrates that only the ACG and the combined MR criteria are independent predictors of full MCA recanalization after thrombectomy with the MERCI retriever.
How may the vigor of the collateral flow influence MCA recanalization in patients treated with thrombectomy alone?
The answer to this question raised by the results of the aforementioned multivariate analysis is more speculative and needs more investigations. Virchow’s triad is traditionally invoked to explain pathophysiologic mechanisms leading to thrombosis, alleging concerted roles for abnormalities in blood composition (e.g., platelet activation), endothelium injury, and reduced blood flow in the development of arterial and venous thrombosis . Given this pathophysiology, one could suggest that in patients treated with thrombectomy alone, a better collateral flow decreases the relative importance of the reduced blood flow component of the Virchow’s triad within the MCA vascular bed, which could limit the thrombus extension and make the thrombectomy procedure easier.
Similarly, Jovin et al. had previously suggested that patients with M1 or ICA terminus occlusion and severely hypoperfused hemisphere have a higher proportion of ischemic core in which, because of stagnant flow, the vessels supplying the area have a higher clot burden in the vascular bed of the affected territory which decreases the probability of recanalization . Interestingly, these authors demonstrated that the higher the regional CBF in the ipsilateral hemisphere, the higher the likelihood of successful intra-arterial thrombolysis . Given the significant relationship between the ipsilateral rCBF and the collateral flow deficit index VolATD6 (linear regression, negative correlation, r
2 = 0.63; p < 0.0001) demonstrated in patients with MCA-M1 occlusion , the results from Jovin et al. suggest that a good collateral flow might also facilitate recanalization beyond 3 h during intra-arterial thrombolysis, thanks to a potential lower clot burden due to a reduced intravascular stagnant flow .
To sum up, the instantaneous estimation of the vigor of the collateral flow is predictive of MCA recanalization when performed just before the recanalization therapy, either by means of PWI before IV thrombolysis [7–9] or by means of angiography before thrombectomy [1, 2, current study].
By contrast, when the MRI-based collateral perfusion assessment is performed before the endovascular procedure, the time between this measurement and the reperfusion therapy is longer and the instantaneous estimation of the vigor of the collateral flow using a PWI-derived index is no longer effective alone to predict recanalization after a thrombectomy performed later.
However, if this index is combined with the baseline DWI lesion volume, predictions of the delayed angiographic collateral grading and of post-thrombectomy recanalization are both significantly improved.
The baseline DWI lesion volume has a strong prognostic value for dependency, death and intracerebral hemorrhage in patients with acute ischemic stroke including those who underwent endovascular treatment [29, 33, 34]. It should also be considered as an index of average reactivity of the collateral supply in acute MCA M1 occlusion with potential hemodynamic fluctuations during the onset-to-MRI time. Thus, it might be more informative than a simple instantaneous measurement of the collateral flow efficiency and an important complementary functional parameter to take into account in a MRI-based collateral supply assessment.
Limitations of the study
The first limitation is the retrospective design of the study. However, patients analyzed in this series have been included in a prospective clinical registry from one stroke center with consistent and standardized medical care.
The 9-year inclusion period could be responsible for learning-curve related biases, such as a higher recanalization rate due to increased thrombectomy experience in the stroke center. However, the multivariate analysis demonstrated that this parameter was not an independent predictor of recanalization.
Another limitation is the time elapsed between the MR examination and the angiographic exploration. Indeed, fluctuations in the degree of collateral flow may occur and impact the correlations between early MR-PWI collateral flow assessment and the later angiographic collateral grading [35, 36]. Although the relatively short mean MRI-to-angiography time (Table 1) minimizes such bias, further prospective study is warranted, also with a particular caution about the systematic registration of hemodynamic fluctuations occurring during the MRI-to-angiography time.
In addition, the baseline DWI lesion volume threshold that accurately identifies patients with excellent collateral flow in the current study must not be extrapolated to patients with MCA-M1 occlusion treated within a shorter onset-to-treatment time. Indeed, within 3 h after stroke onset, the baseline DWI lesion volume might be small in spite of a weak collateral flow because the MRI was performed early during the infarct growth. Conversely, the same small baseline DWI lesion volume diagnosed beyond 3 h, a usual onset-to-endovascular thrombectomy time, argues for a slow infarct growth, more probably in relation with a good collateral flow.
Finally, the almost exclusive use of the MERCI retriever in the current study impedes us to extrapolate the predictive value of the combined criterion to predict MCA recanalization after endovascular thrombectomy using other devices. Indeed, it has been recently demonstrated that recanalization rate depends on the retriever used for thrombectomy [20, 21]. Therefore, these results need to be confirmed in a larger cohort of patients having undergone thrombectomy with different retrievers.