Article Text
Abstract
Background The role of mechanical thrombectomy in acute ischemic stroke (AIS) has been further expanded by recent trials which relied on the results of CT perfusion (CTP) imaging. However, CTP parameters for ischemia and infarct can vary significantly across different vendors.
Methods We compared the outcomes of the Siemens CTP software against the clinically validated RAPID software in 45 consecutive patients with suspected AIS. Both perfusion softwares initially processed images using vendor defined parameters for hypoperfusion and non-viable tissue. The software thresholds on the Siemens software were decrementally altered to see if concordant results between softwares could be attained.
Results At baseline settings, the mean values for core infarct and hypoperfusion were different (mean of 30/69 mL, respectively, for RAPID and 49/77 mL for Siemens). However, reducing the threshold values for the later software showed a concordance of values at a relative cerebral blood flow <20%, with resulting core infarct and hypoperfusion volumes at 31/69 mL, respectively, for the Siemens software. A Wilcoxon paired test showed no significant difference between the calculated core infarct and hypoperfusion values, both for the entire population as well as for the subgroup of patients with large vessel occlusion.
Conclusion Equivalent CTP results between vendor softwares may be attainable by altering the thresholds for hypoperfused and non-viable tissue, despite differences in acquisition techniques, post-processing, and scanners.
- stroke
- Ct perfusion
- technique
- brain
Statistics from Altmetric.com
Introduction
Multiple recent trials have shown better patient outcomes in both early (within 6 hours) and late window acute ischemic stroke (AIS) (beyond 6 hours) undergoing mechanical thrombectomy.1–4 Most of these utilized perfusion imaging in patient triage, which was performed using RAPID software (iSchemiaView Inc, Menlo Park, California, USA).1–3 Even though some studies have suggested that patients in extended (>6 hours) time windows may benefit from mechanical thrombectomy regardless of findings on perfusion imaging, the current thrombectomy guidelines (based on randomized trials) require CT perfusion (CTP) based selection of patients presenting beyond 6 hours.1 3 5
The process of triaging patients with AIS for thrombectomy for individual sites however, is complicated by lack of consistency of results when using different vendor software. Austein et al, for example, post-processed the CTP images using software from three different vendors (RAPID, Philips, and Siemens) and noted a significant discrepancy in the software generated volumes for hypoperfusion and ischemic core.6 This discrepancy between different software is likely a result of different recommended acquisition, post-processing techniques, and thresholds for ischemia and infarct. Cereda et al, for example, showed that using a threshold of relative cerebral blood flow (rCBF) <38% led to a greater concordance between the CTP predicted infarct volume and diffusion weighted imaging compared with an rCBF threshold of <30% or 42%.7
We hypothesized that the differences in inbuilt thresholds for ischemic and non-viable tissues and post-processing techniques are likely contributing to different values across multiple software and could potentially be corrected, thus allowing for more reliable results across different vendor platforms.
Materials and methods
Design
The study was approved by the institutional review board, and patient consent was waived, given the retrospective nature. We selected 45 consecutive patients with suspected AIS who underwent CTP between March and July 2018. As the aim of the study was to evaluate concordance between different softwares, selection was made independent of the presence or absence of AIS.
Image acquisition
Images were acquired on either a 128 slice (SOMATOM definition AS; Siemens AG, Forchheim, Germany) or a 192 slice (SOMATOM Force, Siemens AG) scanner.
A total of 26 patients were scanned on the former (80 kV and 9.6 cm scan range); the remaining 19 were scanned on the latter (70 kV and 11.4 cm scan range). All acquisitions were reconstructed to 10 mm slices with 10 mm increments, resulting in 9 and 11 slices per volume, respectively. All scans were done with 40 mL of non-ionic iodinated contrast (Isovue-370, iopamidol, 370 mg iodine/mL; Bracco Diagnostics, Princeton, New Jersey, USA). Contrast injection was performed using a power injector (Stellant D; MedRad Inc, Indianola, Pennsylvania, USA), with a 2 s delay between contrast injection and scan initiation.
Data were acquired using the protocol recommended by RAPID, with 4 scans 3 s apart followed by 15 scans 1.5 s apart, and another 9 scans 3 s apart, totaling 28 scans over approximately 60 s.
Post-processing
Both softwares performed the initial post-processing using the vendor specified parameters to include initial motion correction, bone removal and brain segmentation, 4D noise reduction, and reference vessel and arterial input function detection. Hypoperfusion volume calculation on one of the patients with intracranial hemorrhage failed on RAPID. The patient was therefore excluded from the final analysis, leaving a total of 44 patients.
The RAPID software (V.4.6.1), used in application service mode, provided infarct core/non-viable tissue (NVT) with vendor recommended thresholds as rCBF <30%, and hypoperfused volume as Tmax >6 s, and the mismatch ratio (MR)=hypoperfused/ NVT.
The Syngo software (Syngo.via CT Neuro Perfusion VB30; Siemens Healthineers, Erlangen, Germany) was used in a slightly modified form to allow repetitive runs with different settings in batch mode and to dump numerical results directly in data files. The data were post processed using different rCBF thresholds between 20% and 30%, with increments of 2%. Other than the modification to allow for repetitive runs using different thresholds, the actual code for each individual run was identical to the commercial product. In Syngo, NVT is defined as rCBF <threshold. Voxels not treated as NVT are classified as ‘tissue at risk’ (TAR) if Tmax >6 s. MR was calculated as MR=(TAR+NVT)/NVT.
Results
Of the 44 patients, 23 were men and 21 were women. Mean patient age was 71.3 years (range 38–95 years). All patients presented with suspected AIS. National Institutes of Health Stroke Scale (NIHSS) score was calculated in all except one patient who was comatose at presentation. Mean NIHSS score was 7 (range 1–24). Thirteen patients underwent mechanical thrombectomy. A follow-up study within 24 hours was available for 39 patients (33 with MRI and six with CT) and confirmed the presence (27) or absence of infarction. The 12 patients with no infarction at 24 hours were considered stroke mimics by clinical assessment. Three patients chose palliative care, one went back to the local hospital after deemed to be not a candidate for thrombectomy, and one could not be imaged in the 24 hour window due to multiple comorbidities. These patients were not subsequently imaged.
As per the TOAST (Trial of Org 10172 in Acute Stroke Treatment) criteria,8 13 patients had a cardioembolic etiology, 4 had large artery atherosclerosis while the etiology remained undetermined in 15 cases. A total of 18 patients received intravenous tissue plasminogen activator. Of the 13 patients who underwent mechanical thrombectomy, the Thrombolysis in Cerebral Infarction status was 2b in seven patients and 3 in six patients. At follow-up, three patients died while four were discharged to a hospice. A 90 day modified Rankin Scale score was available only in 14 patients and was 0–4 (mean 1.6).
Using RAPID, the mean core infarct volume was 30 mL with a mean hypoperfusion volume of 69 mL. Using Syngo.via at the predefined setting (defining core as rCBF <30%), the corresponding values were 49 mL and 77 mL, respectively. The reference values on Syngo.via were then decrementally altered by 2% to achieve values similar to the RAPID output. This was achieved with an rCBF threshold of 20%, giving a mean core of 31 mL and hypoperfusion volume of 69 mL. A Wilcoxon paired test did not show any statistically significant difference between the calculated values for the core and hypoperfusion volumes (p values of 0.19 and 0.3, respectively). Additionally, a Bland–Altman analysis for the 44 patients who were analyzed showed no statistically significant difference in 40 cases (91%) in terms of NVT and in 42 cases (95%) in terms of hypoperfusion (figure 1).
Interestingly, when the results were analyzed using the DEFUSE III (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke III) criteria as a ‘go versus no go’ from the point of view of selection for mechanical thrombectomy, a concordant result was obtained in 40/44 cases (91%), with two false negatives for both RAPID and Syngo.via, respectively. Of the four patients with discordant results, two patients did not have large vessel occlusion (NIHSS scores of 11 and 1, respectively). In the remaining two patients, one had an M2 middle cerebral artery (MCA) occlusion with an NIHSS score of 23 while the other had an M2 MCA occlusion with an NIHSS of 3 and a predominant penumbra-like pattern. It should be noted that neither case would have been misclassified when anatomic location was taken into consideration, as these cases either had no demonstrable large vessel occlusion or involved M2 occlusions that would not be eligible for DEFUSE III or DAWN (Diffusion weighted imaging or computerized tomograph perfusion Assessment with clinical mismatch in the triage of Wake-up and late presenting strokes undergoing Neurointervention with Trevo).
Additionally, when analyzed as a subgroup, in patients with large vessel occlusions (intracranial artery and M1 MCA) (n=17 in the current study), the mean core infarct and hypoperfusion volumes for RAPID and Syngo.via were 75/151 and 74.4/147 mL, respectively. Again, a Wilcoxon paired test did not show any statistically significant difference between the calculated values for the core and hypoperfusion volumes (p values of 0.78 and 0.54, respectively). Similarly, a Bland–Altman analysis did not show any significant difference between the core volumes, and agreement in 16/17 cases for the hypoperfusion volumes (figure 2). In the one case with disagreement, the patient had a large core and would not have qualified for mechanical thrombectomy regardless. Table 1 lists the patients with large vessel occlusion, along with the site of occlusion and results of the perfusion analysis using both processing softwares. Similarly, in the subgroup of patients who underwent thrombectomy (n=13), the mean volumes for the two softwares were 26/105 and 29/101 mL, respectively. In both of these groups, both softwares showed 100% concordance in terms of ‘go versus no go’.
Discussion
Our preliminary findings suggest that a high concordance, both in terms of NVT and TAR volumes as well as therapeutic decision making in AIS, is feasible between different CTP post-processing softwares, despite differences in acquisitions and post-processing, and may not be a time or resource intensive task. Of the five different thresholds used to define the NVT based on different values of rCBF, similar results were obtained with an rCBF of 20%, with a mean difference in NVT of <1 mL and a mean difference in hypoperfusion volume of <1 mL. In fact, with the modified threshold of rCBF of 20%, there was no statistically significant difference between the mean values for core and hypoperfusion volumes, both for the entire group as well as for patients with demonstrated large vessel occlusion.
Additionally, Bland–Altman analysis of all patients presenting with suspected AIS showed no significant differences in NVT values in 91% (and 100% in the LVO subgroup) when using a SD of 15 mL to define the limits of agreement. Even more reassuringly, in patients where there was a difference, none of these would change therapy decisions based on DEFUSE III criteria as two patients had a large core and would be excluded, while the remaining two still met the criteria for intervention based on both softwares. In addition, analysis of the results from the perspective of selection for mechanical thrombectomy also showed high concordance (91% based on perfusion parameters alone and 100% when anatomical site of occlusion was also considered).
The baseline discrepancy in calculated NVT and TAR between RAPID and Syngo.via likely existed due to differences in acquisition techniques, predefined thresholds for NVT and TAR used by the software, and differences in receiver operating characteristic analysis. Even though both RAPID and Syngo.via define NVT as rCBF <30%, the latter uses different reference values for grey and white matter, which might explain the discrepancy in NVT volumes. Lowering the threshold by about 10% results in a mean difference of NVT volumes of <1 mL.
Some authors who have previously compared CTP parameters across different vendors noted significant discrepancies in the values for NVT and TAR.6 This was however based on thresholds that were different from the current study. The current study used Tmax to define hypoperfusion instead of CBF based values. Application of Tmax as a reliable parameter to define hypoperfusion has been previously shown.9 Importantly, the values for TAR in the current study were not significantly different between the two softwares. This may be explained by the similar values (Tmax >6 s) used to define hypoperfusion and the fact that both softwares use a delay insensitive deconvolution algorithm which corrects for the delay in contrast arrival, thereby resulting in better concordance. Given the lack of any meaningful differences in the core and hypoperfusion volumes, both for all incomers and the subgroup of patients with large vessel occlusion, and 100% concordance for go versus no go in patients with large vessel occlusion, our preliminary analysis suggests that the two softwares could potentially be used interchangeably.
Our preliminary findings are important for multiple reasons. First, despite the differences in acquisition guidelines for RAPID and Syngo.via and scanner platform, the results were consistent. For example, the vendor proposed data acquisition guidelines for RAPID (detailed above) and Syngo.via (30 scans with 1.5 s sampling for 45 s) are different. This is relevant as previous studies have defined thresholds for absolute and relative CBF for various softwares using follow-up MRI/CT as the ground truth, but have not specifically evaluated if a ‘cross talk’ between outcome values across different platforms is possible.7 10 Additionally, the scans in the current study were acquired on different scanners. Our results therefore suggest that equivalent results can be obtained despite differences in scanning parameters and scanners. Moreover, as the role of CTP based decision algorithms emerges and gets more refined, especially in late window AIS, so does the need for comparability between different post-processing softwares, allowing for communication and decision uniformity across centers which may not be using the same softwares and scanning parameters. Our study is therefore a logical next step in unifying stroke care independent of imaging techniques and post-processing.
Limitations of our study include a small sample size and inclusion of both normal and abnormal cases. However, the current study was aimed at comparing the results of post-processing across a spectrum of normal and abnormal cases and consecutive cases were therefore selected. Another potential limitation is the absence of a correlation between the study results and the actual infarcted volume on diffusion weighted imaging. Some authors have also noted that CTP may overestimate penumbra.11 However, as the study was aimed at evaluating if similar output CTP parameters were achievable between different softwares, the comparison with MR imaging, or differentiation between ‘at risk’ penumbra and benign oligemia, was not addressed in the current study. Additionally, we only compared the rCBF and Tmax parameters. Comparison of rCBV may be more meaningful if true outcome in terms of actual infarcted volume is available, which was not possible in the current analysis, given the patients with no stroke and infinite mismatch ratios. Another limitation of the current preliminary analysis is the inclusion of all incomers, regardless of time of onset since stroke. As discussed above, CTP based patient selection for mechanical thrombectomy is currently only recommended in patients presenting between 6 and 24 hours, based on the findings of the randomized controlled trials.12 Previous studies have shown that CTP may not reliably identify patients with early AIS who may benefit from intra-arterial therapy.13 Moreover, using a combination of non-contrast CT and CT angiogram findings for patient selection (instead of CTP), as explored in more recent retrospective studies, also needs to be prospectively compared with CTP results alone.5 Moving forward, we propose a larger, more indepth comparison of these softwares in patients with late window AIS, comparing the results not only in terms of NVT and TAR, but also in terms of clinical decision algorithms, which may be more relevant from a clinical perspective.
In conclusion, vendor perfusion software can be used to accurately triage patients for thrombectomy despite differences in acquisition techniques, and may allow for comparable results across different platforms.
References
Footnotes
Contributors Guarantor of integrity of the entire study: GB, CD, and EK. Study concepts and design: GB, CD, and BP. Literature research: GB, EK, and KL. Experimental studies/data analysis: GB, EK, MJ, and KL. Statistical analysis: EK, MJ, and GB. Manuscript preparation: GB, BP, and CD. Manuscript editing: GB, CD, EK, and KL.
Funding The department has a research agreement with Siemens AG, Forchheim, Germany. The study was funded through a research grant under the agreement.
Competing interests EK and MJ are full time employees of Siemens AG, Forchheim, Germany.
Ethics approval The study was approved by the local institutional review board.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Data may be made available from the corresponding authors upon reasonable request.
Presented at Accepted as an abstract for ASNR 2019.
Patient consent for publication Not required.