Article Text

P-002 Automated and quantitative assessment of pial collateral recruitment on digital subtraction angiography during acute ischemic stroke
  1. G Christoforidis,
  2. C Haddad,
  3. T Carroll,
  4. M Giger
  1. Radiology, University of Chicago, Chicago, IL


Introduction This work developed and evaluated an automated and quantitative assessment of pial arterial supply to an ischemic territory based on digital subtraction angiography during MCA or ICA occlusion.

Methods Pretreatment anteroposterior arteriograms from patients who underwent conventional arteriography for endovascular treatment of m1 MCA or carotid terminus occlusions acquired at 6 frames a second were analyzed using a custom made MATLAB program. Five semiautomatically placed regions of interest (ROIs) fanning across the ischemic territory were placed within the affected hemisphere on each arteriogram (figure 1). Ten angiographic parameters were calculated which effectively represent indirect surrogates for transit time, cerebral blood flow and cerebral blood volume. Reproducibility was tested across varying frame rate and repeated arteriograms using Bland-Altman statistic. Each parameter in each ROI was then normalized to a normally perfused territory on the same angiogram to derive a value relative to normal tissue. The derived values across the ROIs were then used to derive an overall index of change in each parameter. using the exponential: FNorm(r) = (r−1) where r is the ROI index, β is the parameter of exponential response for the particular feature F, and FNorm is the normalized parametric value. The exponential response parameter β is used to automatically determine a patient’s pial index. The index values were then compared to manually derived pial collateral scores. ANOVA analysis was used to assess performance of the features for the task of distinguishing patients with favorable pial collaterals.

Results A total of 24 arteriograms in 14 patients were included. Features corresponding to CBV and CBF demonstrated favorable interclass correlation (for reproducibility) compared to features corresponding to transit time. There were statistically significant differences (p=0:00625) in all features between patients with poor and good collateral recruitment, however, the parameters demonstrating the greatest separation were those for arterial phase blood flow and arterial arrival time. In part this is related to artifacts intrinsic to projection two dimensional angiography.

Conclusion We have devised a computerized quantitative and automated feature-based method which can reproducibly assess pial collateral recruitment in acute ischemic stroke and found that it could distinguish patients with good pial collaterals. We believe that such a methodology will help reduce any potential subjectivity associated with current grading scales for pial collateral recruitment.

Abstract P-002 Figure 1

Graph demonstrates time density curves derived from ROI3 and ICA from the adjacent composite arteriogram. Parameters were derived from the curves.

Disclosures G. Christoforidis: 1; C; NIH R01-NS093908, AHA GRNT-20380798. C. Haddad: 1; C; AHA GRNT-20380798. T. Carroll: 1; C; NIH R01-NS093908, AHA GRNT-20380798. M. Giger: 1; C; AHA GRNT-20380798.

Statistics from

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.