Comparison of sequential multi-detector CT and cone-beam CT perfusion maps in 39 subjects with anterior circulation acute ischemic stroke due to a large vessel occlusion

Abstract.

Purpose

The critical time between stroke onset and treatment was targeted for reduction by integrating physiological imaging into the angiography suite, potentially improving clinical outcomes. The evaluation was conducted to compare C-Arm cone beam CT perfusion (CBCTP) with multi-detector CT perfusion (MDCTP) in patients with acute ischemic stroke (AIS).

Approach

Thirty-nine patients with anterior circulation AIS underwent both MDCTP and CBCTP. Imaging results were compared using an in-house algorithm for CBCTP map generation and RAPID for post-processing. Blinded neuroradiologists assessed images for quality, diagnostic utility, and treatment decision support, with non-inferiority analysis (two one-sided tests for equivalence) and inter-reviewer consistency (Cohen’s kappa).

Results

The mean time from MDCTP to angiography suite arrival was $50\pm 34\min$, and that from arrival to the first CBCTP image was $21\pm 8\min$. Stroke diagnosis accuracies were 96% [93%, 97%] with MDCTP and 91% [90%, 93%] with CBCTP. Cohen’s kappa between observers was 0.86 for MDCTP and 0.90 for CBCTP, showing excellent inter-reader consistency. CBCTP’s scores for diagnostic utility, mismatch pattern detection, and treatment decisions were noninferior to MDCTP scores (alpha = 0.05) within 20% of the range. MDCTP was slightly superior for image quality and artifact score (1.8 versus 2.3, $p<0.01$).

Conclusions

In this small paper, CBCTP was noninferior to MDCTP, potentially saving nearly an hour per patient if they went directly to the angiography suite upon hospital arrival.

1. Introduction

As the publication of the initial five studies showing a benefit of endovascular thrombectomy over medical treatment in selected patients with an anterior circulation large vessel occlusion (LVO), multiple publications have provided further evidence of the importance of saving time in optimizing clinical outcomes.^1–7 In a recent sub-group analysis of 103 subjects selected for thrombectomy, Ribo et al.⁸ evaluated the impact of the time from stroke onset to reperfusion, time from onset to CT, and time from CT to reperfusion on clinical outcome. Overall, the odds of achieving a good outcome were reduced by 26% for every 30-min delay in reperfusion; only the interval between CT imaging to reperfusion showed a significant negative association with clinical outcome.⁸

There is broad agreement that the information provided by multi-detector CT perfusion (MDCTP) is helpful in selecting patients for thrombectomy; there is, however, also general recognition that acquiring this information can come with a significant time delay. Although the data acquisition time necessary for an MDCTP study can be short, the delay is attributed to other factors in clinical workflow, e.g., inherent delays associated with transfer to the MDCT facility, movement from the transfer gurney to the CT table, and then, back to the gurney for transfer to the ED or angiography suite. This picture-to-puncture interval (P2P) may account for a very significant amount of time delay; it also is one where the duration is highly correlated with clinical outcome.^8–10 As a time delay substantially reduces the likelihood of optimal clinical outcomes, the significance of shortening the P2P interval cannot be overstated. Eliminating the associated time cost also addresses the reluctance of many clinicians to obtain perfusion studies.¹

Small case series comparing C-Arm cone beam CT perfusion (CBCTP) with MDCTP imaging in patients with acute anterior circulation ischemic strokes have reported good correlations for identifying regions of ischemic core and penumbra with the two modalities.^11,12 More recent studies have shown both the feasibility and the potential for meaningful reductions in the time from onset to reperfusion of employing a one-stop-shop workflow where conventional imaging is bypassed and selected patients are brought directly from the ED to the angiography suite for imaging and treatment.^13,14 Improvements in C-Arm CBCT hardware, software, and artifact reduction algorithms have considerably improved the capability of a CBCTP acquisition to more closely follow a contrast bolus through the cerebrovascular circulation.^15–17 The purpose of our study was to compare the image quality and information content of MDCTP images with CBCTP images in subjects with an acute ischemic stroke (AIS) due to an LVO. We also aimed to determine the potential time saved had subjects been brought directly to the angiography suite upon hospital arrival.

2. Methods

2.1. Patient Selection and Workflow

All studies were performed under an Institutional Review Board—approved protocol. On hospital arrival, patients with a suspected LVO were evaluated by a stroke team consisting of a stroke neurologist, a neuroradiologist, and a neurointerventionalist. If an LVO was suspected, an imaging protocol consisting of a non-contrast CT, CTA, and CTP was used in making a decision to proceed with endovascular therapy. Treatment decisions were made by the treating physician. Subjects who had been enrolled in the study underwent an additional CBCTP study immediately on arrival in the angiography suite.

Of 939 patients admitted with symptoms of AIS during the interval between June 2017 and April 2019, 226 (24%) were selected for endovascular therapy. Of these, 54 were consented and enrolled in the study; data from 39 of these are the basis for this report. The inability to obtain consent from family members without delaying treatment was the major cause of non-enrollment. Patients who were pregnant, had severe comorbidities, had severe renal disease (e.g., stage 4 to 5 with $\mathrm{eGFR}<50$) or renal transplant, or were less than 18 years of age were excluded from the study. Figures 1 and 2 show the workflow for patient enrollment and the demographics of enrolled subjects.

Fig. 1.

Flow chart of patient recruitment.

Fig. 2.

Summary of enrolled patient demographics and stroke laterality.

2.2. Data Acquisition and Imaging Post-Processing

All patients were managed using an institutional stroke protocol; this included non-contrast CT (NCCT), head and neck CTA, and CT perfusion (MDCTP). The standard perfusion protocol used at our institution acquires 22 time points using the scanners’ fastest rotation times (0.35 or 0.5 s) and a 2.5 s interscan delay. The contrast injection protocol (for patients $<90\mathrm{kg}$) is 40 mL of Iopamidol $370\mathrm{mgI}/\mathrm{mL}$ (Isovue) followed by a 30-mL saline flush. This yields a total scan time of $\sim 65\mathrm{s}$. These data served as the subject’s baseline for treatment decisions. Enrolled subjects also had a CBCTP study immediately upon arrival in the angiographic suite. The CBCTP study was done using a biplane C-Arm angiographic system (Axiom Artis Zee; Siemens Healthineers, Forchheim, Germany). Contrast was injected into a peripheral vein with the use of a power injector (60 mL, $5\mathrm{mL}/\mathrm{s}$ followed by 60-mL saline flush) at the start of the rotational acquisition. The acquisition protocol consists of 10 sequential back-and-forth gantry rotations (sweeps) with five sweeps acquired in each direction. The time for each sweep was 4.2 s with a 1 s pause due to mechanical limitations of the gantry. The first and second acquisitions are used as masks for the data in the back-and-forth directions leaving eight sinogram datasets; this provided $\sim 41\mathrm{s}$ of temporal coverage.

The CBCTP data were reconstructed with the eSMART-RECON reconstruction algorithm^15,16 to enhance temporal resolution for perfusion map calculation. For each CBCTP acquisition, 40 time-resolved volumes were reconstructed. For each patient, the CBCT reconstructed volumes were registered to the MDCTP images, and pixel size and slice thickness were matched. A custom DICOM header was attached to the CBCT volumes to enable RAPID processing. The perfusion datasets for both MDCTP and CBCTP were then processed with RAPID Perfusion software (iSchemaView, Mountain View, California, United States), and parametric perfusion maps (CBF, CBV, MTT, and Tmax) were calculated. The total current processing time for the CBCTP images included $\sim 7\min$ on average for reconstruction and 3 min for perfusion map calculation.

2.3. Data Analysis and Image Evaluation

2.3.1. Quantitative image comparison

The DEFUSE 3 criteria provide an objective way to make a treatment decision based on quantitative measures in parametric perfusion maps.¹⁸ The DEFUSE 3 criteria are as follows: (1) the core infarct volume should be less than 70 mL, (2) the penumbra-to-core mismatch ratio should be greater than 1.8, and (3) the penumbra volume (salvageable tissue) should be at least 15 mL. The volumes of rCBF deficit (defined as $\mathrm{rCBF}<30\%$ relative maximum; core infarct size surrogate defined by DEFUSE 3), tMax delay (defined as $\mathrm{tMax}>6.0\mathrm{s}$), penumbra volume (volume difference of $\mathrm{tMax}>6\mathrm{s}$ and core infarct), and penumbra to core mismatch ratio indicated by RAPID were recorded for all cases to align with these criteria. For each of these four measurements (CBF volume, tMax volume, penumbra volume, and mismatch ratio) the means from MDCTP and CBCTP were compared using a paired two-sided $t$-test without assuming equal variance between the two samples. Bland–Altman analysis was also used to understand the bias or value dependence of the measurements, and Box-and-whisker plots were used to evaluate outliers.

In addition, the measurements were compared against the DEFUSE 3 criteria to determine if treatment was recommended. Using the MDCTP-derived measurements as the ground truth, the treatment decisions based on CBCTP measurements were evaluated. A confusion matrix and Fisher’s Exact Test were used to assess whether the treatment decisions from CBCTP would align with those based on the DEFUSE 3 criteria and MDCTP measurements.

2.3.2. Qualitative reader evaluation

The MDCTP and CBCTP maps were imported into a research platform (Flywheel.io Minneapolis, Minnesota, United States) for evaluation. With guidance from an experienced neuroradiologist, the windows and levels were adjusted separately to optimize maps for viewing and analysis. The 74 data sets (37 MDCTP and 37 CBCTP) were then randomized into a PowerPoint presentation with a scrollable AVI file containing all image slices for the volume, with each of the 74 slides showing the CBF, CBV, MTT, and $T_{\max }$ maps of one modality for one patient. Three experienced neuroradiologists independently evaluated the overall image quality, capability of making stroke diagnosis, capability of making a treatment decision (using penumbra volumes and ratios as well as infarct volume as well as parametric maps), and capability of detecting mismatch patterns (if applicable) using the five-point Likert scale (1: best, 5: worst). Each radiologist was also asked to choose one of the following stroke diagnosis decisions (no stroke; left hemisphere stroke; right hemisphere stroke; bilateral stroke). The exact questions seen by the radiologists for review are provided in the Appendix. The ground truth of stroke diagnosis was determined by the mechanical thrombectomy treatment record and other clinical medical records. For both MDCTP and CBCTP, stroke diagnostic accuracy was given by calculating the percentage of the number of responses that are consistent with the ground truth. For other metrics, whether CBCTP is noninferior to MDCTP was analyzed using two one-sided tests for equivalence (TOST) with an equivalence bound of 20% of the score range. The inter-reviewer consistency was quantified using Cohen’s kappa.

Three experienced neuroradiologists blinded to all clinical information independently assessed each case by completing a series of questions shown along with each slide. These were designed to provide data that would allow a comparison of image quality and information content of each of the two modalities.

3. Results

The mean time from MDCTP imaging to angiography suite arrival was $50\pm 34\min$, and the average time from arrival in the angiography suite to the first arterial DSA image was $21\pm 8\min$ (Fig. 3). The length of this interval is integral to the endovascular treatment and is dependent on many factors primarily related to the patients’ clinical condition and the time needed to prepare for groin puncture. The time required to obtain a CBCTP acquisition is $\sim 1\min$. From these data, it appears that eliminating MDCTP imaging could result in a very meaningful reduction in the time from hospital arrival to the start of treatment, i.e., on average $\sim 1\mathrm{h}$. Obtaining imaging in the angiography suite with CBCTP did not result in either an increase in clinically relevant radiation exposure or the contrast medium dose.

Fig. 3.

Distribution of time elapsed from MDCTP imaging to arrival in an angiography suite (a) distribution of time elapsed from arrival in the angiography suite and first arterial image (b) and distribution of time elapsed between MDCTP imaging and CBCTP imaging in the angiography suite (c).

Both subjective and quantitative qualities of CBCTP and MDCTP perfusion maps were comparable (Figs. 4 and 5 provide representative cases). The mean, median, and interquartile ranges (IQRs) for the CBF and tMax deficit volumes along with the mismatch volumes and ratios for the MDCTP and CBCTP images are summarized in Table 1. The difference in mean CBF and tMax deficit volumes were significantly different between MDCTP and CBCTP; however, significance was not noted in the mismatch volumes or ratios. Bland–Altman plots showing these measurements are shown in Fig. 6, and Box-and-whisker plots are shown in Fig. 7. A correlation plot showing the difference in ischemic core volume versus the time interval between MDCT and CBCT acquisitions is presented in Fig. 8. The figure demonstrates that ischemic core size measured by CBCT was larger for almost all patients; however, the difference in measured ischemic core volume did not correlate with the time interval. The treatment decisions predicted using DEFUSE 3 criteria are shown in a confusion matrix in Figs. 8 and 9. The Fisher’s exact test produced an odds ratio of 6.52, indicating that treatment decisions were significantly more likely to be made based on MDCT than CBCT. This difference was statistically significant, with a $p$-value of 0.019.

Fig. 4.

79-Year-old male with L ICA occlusion. The subject arrived in the angiography suite 37 min after MDCTP imaging. CBCTP imaging done in an angiography suite 42 min after MDCTP imaging. The top panel shows the summary maps, and the lower panel shows representative perfusion parameter maps from the slice indicated in the white dashed box. Over the interval between MDCTP and CBCTP imaging, the size of the core infarct and volume of potentially salvageable parenchyma showed little change. In the MDCTP images, the reported $\mathrm{rCBF}<30\%$ volume was 26 mL, and the reported $\mathrm{tMax}>6.0\mathrm{s}$ was 181 mL, yielding a mismatch volume of 155 mL and a mismatch ratio of 7.0. In the CBCTP images, the reported $\mathrm{rCBF}<30\%$ volume was 17 mL, and the reported $\mathrm{tMax}>6.0\mathrm{s}$ was 259 mL, yielding a mismatch volume of 242 mL and a mismatch ratio of 15.2. The similarity between scans is expected due to the short interval between MDCTP and CBCTP imaging acquisition times.

Fig. 5.

79-Year-old female with L M1 occlusion. The subject arrived in the angiography suite 135 min after MDCTP imaging was performed, and CBCTP imaging was performed 148 min after MDCTP imaging. The top panel shows the summary maps, whereas the lower panel shows representative perfusion parameter maps from the slice indicated in the white dashed box. Over the interval between MDCTP and CBCTP, the imaging indicates the core infarct grew, whereas the salvageable penumbra region demonstrated little change. In the MDCTP images, the reported $\mathrm{rCBF}<30\%$ volume was 0 mL, and the reported $\mathrm{tMax}>6.0\mathrm{s}$ was 78 mL yielding a mismatch volume of 78 mL and an infinite mismatch ratio. In the CBCTP images, the reported $\mathrm{rCBF}<30\%$ volume was 49 mL, and the reported $\mathrm{tMax}>6.0\mathrm{s}$ was 81 mL yielding a mismatch volume of 32 mL and a mismatch ratio of 1.7. This finding demonstrates that CBCTP immediately prior to endovascular therapy has the potential to change treatment-making decisions with the most accurate information provided at the time of invention instead of a delay that can occur when transferring from MDCTP acquisition to the angiography suite.

Table 1.

Summary statistics, including mean, median, and interquartile range (IQR), for RAPID MDCTP and CBCTP measurements of CBF volume, tMax volume, mismatch volume, and mismatch ratio. Corresponding $p$-values from t-tests are provided below the means being compared for each measurement.⁶

	Mean	Median	IQR	Mean	Median	IQR
	CBF volume (mL)			tMax volume (mL)
MDCTP	51.75	36.5	66.5	138.86	141.5	86.5
CBCTP	105.31	97.55	119.47	223.71	233.01	195.79
$p$-Value	0.0018	—	—	0.0017	—	—
	Mismatch volume (mL)			Mismatch ratio
MDCTP	87.11	87.5	79.25	5.47	3.15	3
CBCTP	118.4	133.67	119.65	3.78	2.38	2.34
$p$-Value	0.227	—	—	0.2694	—	—

Fig. 6.

Bland–Altman plots comparing RAPID MDCTP and CBCTP measures for CBF volume (a), tMax volume (b), mismatch volume (c), and mismatch ratio (d). The plots display the mean difference (bias) between the two measurement methods and the limits of agreement (mean difference ± 1.96 standard deviations).

Fig. 7.

Box-and-whisker plots comparing RAPID MDCTP and CBCTP measures for CBF volume (a), tMax volume (b), mismatch volume (c), and mismatch ratio (d). Each plot displays the median, interquartile range, and potential outliers for both methods, highlighting the distribution and variability of the measurements. Statistically significant differences between the methods are indicated with an asterisk.

Fig. 8.

Correlation plot showing the difference in measured infarct core size between MDCT and CBCT plotted against the time gap between the two acquisitions.

Fig. 9.

Confusion matrix comparing treatment decisions between MDCT (considered ground truth) and CBCT using the DEFUSE 3 criteria. The matrix shows the distribution of true positives, true negatives, false positives, and false negatives. The color-coding highlights agreement and discrepancies between the two methods, with MDCT showing more frequent “treat” decisions compared with CBCT.

Regarding stroke diagnoses the reviewer’s diagnostic accuracies using the MDCTP maps was 96% [93%, 97%] as compared with 91% [90%, 93%] when using the CBCTP maps. Inter-reader consistency was excellent. Cohen’s kappa coefficient between observers was 0.86 for MDCTP-based diagnosis and 0.90 for CBCTP-based diagnosis. The observers’ scores for using CBCTP maps for diagnosis, to detect a mismatch between CBF and tMax, and to recommend a treatment decision, i.e., appropriate for endovascular therapy based on the CBCTP maps were found to be noninferior to the decisions made using the MDCTP images (alpha = 0.05 within 20% of the score range). The subjective scores of image quality and the presence artifacts of the MDCTP maps were slightly superior to those for the CBCTP maps (1.8 versus 2.3, $p<0.01$).

4. Discussion

In this study of patients with an anterior circulation AIS due to an LVO, we found that CBCTP studies obtained in the angiography suite were not inferior to MDCTP studies done using conventional techniques. We also established that a proposed workflow where patients were brought directly from hospital arrival to the angiography suite, bypassing conventional imaging, could result in a very meaningful decrease in the interval between stroke onset and the start of endovascular therapy. This time savings will have limitations due to requirements for patient and room preparation time, clinical consultations with neurology, and patient transport. This improvement is achievable with no increase in the contrast medium dose and with a decrease in radiation exposure as compared with MDCTP imaging. Data from a multi-sweep CBCTP acquisition can also be reconstructed as fully time-resolved cone-beam CTAs. Having this information available without a need for additional contrast or radiation dose could also enhance the accuracy of diagnosis and treatment planning.¹⁹ In addition to the potential time-saving, imaging in the angiography suite would also allow for additional perfusion imaging to be done during or at the end of endovascular therapy. The potential of this to influence both prognostication and post-treatment management is obvious.

Quantitative analysis did show some differences in the measured regions of CBF and tMax deficit; however, notably, the mismatch volumes and ratios, which are crucial for determining the extent of salvageable tissue, were not significantly different across the two methods. The infarct volume did show a change in the case presented in Fig. 5, which could change the treatment decision, although in this particular case, the infarct core remained less than the suggested 70-mL threshold in DEFUSE 3.¹⁸ Although the two imaging acquisitions are close in time (median difference: 68 min, IQR 67), the strokes may evolve in that time window with growth in the “ischemic core” possible. In this study, cohort the measured ischemic core volume on CBCT was found to be greater on average than the measured volume from MDCT [Fig. 7(a) as well as Fig. 8], with all patients except for five demonstrating an increase in ischemic core volume measurement. This may be attributable either to bias in the CBCT measurement or possibly to core growth during the time between MDCT and CBCT imaging. The evaluation of treatment decisions based on DEFUSE 3 criteria did also find that more patients would have been ineligible to be treated based on CBCTP imaging, and the most common reason for this was the infarct core measurement $>70\mathrm{mL}$ It would be interesting in the future work to evaluate whether changes in the underlying ischemic core size are actually measured with CBCT perfusion.

Our study also provides additional evidence of the feasibility and safety of using the “one stop shop” workflow for the evaluation and treatment of patients with an anterior circulation stroke due to an LVO. In a recent single-center observational study, Psychogios et al.¹⁴ reported their results on 230 AIS patients brought directly from hospital arrival to the angiography suite for imaging. They concluded that, in their specific hospital setting, this reduced intra-hospital time delays. Their study offered no data regarding a comparison of CBCTP with MDCTP imaging.¹⁴ Bouslama and colleagues published results of a single-center prospective study where conventional imaging, in a selected cohort of 54 patients, was replaced using tri-phase CBCT imaging done in the angiography suite.²⁰ They concluded that the approach was safe and that it could result in a significant reduction in treatment times. Except for the small case series reported by Struffert¹¹ and by Niu,¹² there are, to our knowledge, no reports comparing CBCTP imaging and MDCTP imaging done in close proximity in patients with an anterior circulation stroke due to an LVO. One potential concern with the direct-to-angio workflow is the implications it has on angiography suite availability and how it will work if those suites are already overbooked. The first point is a concern if the goal of the imaging is truly a screening procedure rather than a diagnostic. However, for patients who are appropriately triaged based on external imaging or strong clinical indications, the purpose of performing CBCTP in the angiography suite is to confirm both that there is indeed a stroke and that there is salvageable tissue. Although there will be some patients who begin this workflow and then are not treated, with appropriate triage, this should be rare.

Safe implementation of a complete “one stop shop” workflow requires that a non-contrast CBCT be available and adequate for the detection of hemorrhage. A non-contrast CT can be obtained using data from the first two rotations of a multi-sweep acquisition. Based on our experience, however, this is not yet adequate to use for the detection of bleeding. Nevertheless, a non-contrast CBCT with improved image quality can also be obtained using an acquisition focused on soft-tissue delineation, i.e., DYNA CT (Siemens Healthineers, Forchheim, Germany). Although there are reports of CBCT equivalency with MDCT for hemorrhage detection using a variety of soft tissue-focused acquisition implementations, this equivalency has not, in our opinion, been adequately validated.²¹ Currently, the great majority of our AIS patients are transferred from other facilities where a non-contrast MDCT has been obtained as part of the initial evaluation. Studies aimed at determining the sensitivity and specificity of CBCT for hemorrhage detection are ongoing at our institution. Continuous improvement in detector capabilities, reconstruction, and artifact-reducing algorithms (motion, scatter, and beam hardening) give us confidence that, in the near term, a non-contrast CT derived from a multi-sweep acquisition will be adequate for hemorrhage detection.

Generalization of the results from our study is limited because of the still relatively small sample size. Potential differences between the MDCTP studies done in-house and those done at an outside facility before patient transfer to our facility cannot be excluded. An additional limitation of our study is the potential bias of only including patients who are already selected for endovascular therapy. Given the need for additional contrast and ionizing radiation in this cohort, we could only recruit patients selected for treatment but acknowledge that this may introduce some bias. Another limitation is mechanical: the c-arm gantries used in this study were only capable of providing acquisition of 41 s, 24 s shorter than our standard MDCTP protocol. This may result in more difficulty getting the bolus timing right and potentially truncating the bolus and diminishing perfusion map quality. Finally, our study does not address the issues of the sensitivity or specificity of how patient selection for a “one stop shop” protocol using only clinical criteria, i.e., exam and NIHSS might result in the exclusion of patents who might benefit from treatment.

The proposed “one stop shop” workflow offers an opportunity for a major reduction in the interval between stroke onset and the initiation of treatment. It eliminates the time cost currently associated with MDCT imaging. It also allows for additional perfusion imaging to be done when/if needed during endovascular treatment and at the end of treatment. We believe that these capabilities will combine to improve the outcomes of patients undergoing EVT for an AIS.

5. Conclusion

In this paper, CBCTP studies were found to be noninferior to conventional MDCTP studies. It was estimated that almost an hour of time would have been saved for each patient had they come directly to the angiographic suite from hospital arrival.

6. Appendix

Questions posed during reader study:

• 1.What is the image quality?
  • 1- excellent image quality: no limitations for clinical use
  • 2- good image quality: minimal limitations for clinical use
  • 3- sufficient image quality: moderate limitations for clinical use but no substantial loss of information
  • 4- restricted image quality: relevant limitations for clinical use, clear loss of information
  • 5- poor image quality: image not usable, loss of information, image must be repeated
• 2.Based on these maps, can you make a stroke diagnosis?
  • 1- definitely can
  • 2- possibly can
  • 3- unsure
  • 4- possibly can not
  • 5- definitely can not
• 3.If you are forced to make a decision, what would be the stroke diagnosis?

  • N, L, R or B (no stroke, left, right, or bilateral stroke)

• 4.Based on these maps, can you make the decision whether there is a mismatch?
  • 1- definitely can
  • 2- possibly can
  • 3- unsure
  • 4- possibly can not
  • 5- definitely can not
• 5.If you are forced to make a decision, will you say there is a mismatch?

  • Y or N (yes or no)

• 6.Based only on these maps, can you make treatment decisions:
  • 1- definitely can
  • 2- possibly can
  • 3- unsure
  • 4- possibly can not
  • 5- definitely can not
• 7.Artifacts: Are there noticeable artifacts in the perfusion maps?
  • 1- No, no noticeable artifacts
  • 2- Some limited artifacts with no impact on diagnosis
  • 3- Some artifacts with minor impact on diagnosis
  • 4- Some artifacts with major impact on diagnosis
  • 5- Artifacts rendering images non-diagnostic

Biographies

John W. Garrett is an associate professor of radiology, medical physics, and biostatistics and medical informatics at the University of Wisconsin. He received his PhD in medical physics in 2017. His research focuses on translating deep learning techniques into medical imaging, particularly in multi-site settings through federated learning. He is also interested in AI robustness in imaging and works on methods for large-scale secure computing, with expertise spanning x-ray/CT physics and high-performance computing.

Kelly Capel is a clinical assistant professor of radiology in neuroradiology with expertise in head and neck imaging. She enjoys teaching trainees with professional interests including vascular and flow imaging and stroke. She graduated from the biomedical sciences program at Marquette University, attended Rush University for medical school and completed a transitional year at Aurora St. Luke’s in Milwaukee, Wisconsin before completing radiology residency and neuroradiology fellowship at the University of Wisconsin.

Azam Ahmed specializes in surgery for complex skull base tumors (meningiomas, pituitary adenomas, and acoustic neuromas), cerebrovascular disorders and neuroendovascular surgery. He also works closely with ear, nose and throat physicians for endoscopic skull base surgery. Some of his research was directed at improving visualization for neuroendovascular procedures. The main goal of this research was to streamline stroke care and assist with clinical decision making while developing a user-friendly interface.

David Niemann is an associate professor of neurological surgery at the University of Wisconsin School of Medicine and Public Health. He completed his medical degree and residency at the Medical College of Wisconsin and his fellowship at Radcliffe Infirmary, Oxford University. His special interests include cerebrovascular and endovascular neurosurgery. He is one of the nation’s most experienced physicians in the use of a liquid embolic system, Onyx®, that treats arteriovenous malformations at risk for rupture.

Beverly Aagaard-Kienitz is professor of radiology and neurosurgery, program co-director for the neuroendovascular fellowship and administers the neuroendovascular practice at the University of Wisconsin. She practices neuroendovascular surgery exclusively. She has actively engaged in research as site PI, co-PI or investigator in a number of trials: HEAL, MAPS, COCOA, RACER, PUFFs (adjudicator), VISSIT, SENTIS, Cerecyte Coil Trial, SAMMPRIS, Cerecyte Coil Registry, the Wingspan Registry, ASPIRe, PREMIER, RAGE and others.

Biographies of the other authors are not available.

Disclosures

The paper is partially supported by the National Institutes of Health (Grant No. U01EB021183). No authors have relevant conflicts of interest. The angiographic system used had undergone non-commercial modifications.

Code and Data Availability

The datasets used in this project are not publicly available because they include the head and face of patients which may constitute ePHI. The eSMART-RECON reconstruction code will be shared upon reasonable request to the corresponding author.