Automated perfusion lesion outlining accurately speeds stroke assessment

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
 - brain, stroke

An algorithm for automated identification of hypoperfused tissue in acute stroke patients has been shown to agree well with manually defined lesions and is superior to standard threshold approaches of labeling hypoperfused tissue, according to a study published online May 17 in Radiology.

“Our simulations demonstrate that the approach permits accurate outlining of hypoperfused tissue even in the presence of considerable image noise, whereas the typical threshold approach was shown to lead to either over- or underestimation of the correct lesion volume,” wrote Kim Mouridsen, PhD, of Århus University Hospital in Denmark, and colleagues.

The algorithm was based on the assumption that average tissue mean transit time (MTT) in a hypoperfused region is higher than average MTT in normal regions, explained the authors. The border of the hypoperfused lesion was approximated by a smooth curve minimizing variation inside and outside of the boundary.

The technique was tested in 14 acute stroke patients and compared with manual outlines determined by a consensus of four neuroradiologists and with a standard threshold approach. “The threshold approach of perfusion values is a common technique for the identification of tissue lesions and has been suggested as a means for automated identification of hypoperfusion,” wrote the authors. “However, perfusion image values are inherently sensitive to the way in which perfusion raw images are acquired and postprocessed. Therefore, global thresholds may not be comparable across or within patient studies.”

Consensus of four expert readers was used as the standard reference since manual outlining is affected by high interobserver variability. Spatial agreement was quantified using the Dice coefficient, which measures lesion overlap relative to mean lesion volume.

Mouridsen and colleagues reported that mean difference in lesion volume between automated outlines and manual outlines was -9.0 mL, while the lowest mean volume difference for the threshold approach was -25.8 mL. The Dice coefficient observed with the algorithm was significantly higher than the threshold approach at 0.71 and 0.50, respectively.

“We speculate that this fast, user-independent algorithm may help standardize the assessment of perfusion images during the treatment of patients who are suspected of having an acute stroke and facilitate the safe administration of thrombolytic therapy,” concluded Mouridsen and colleagues.