Automated estimation of salvageable tissue: Comparison with expert readers

J Magn Reson Imaging. 2016 Jan;43(1):220-8. doi: 10.1002/jmri.24963. Epub 2015 Jun 2.

Abstract

Purpose: To assess the performance of an automatic perfusion-diffusion mismatch outlining algorithm, in a cohort of acute ischemic stroke patients imaged as part of a multicenter study.

Materials and methods: Magnetic resonance imaging (MRI) from 167 patients with anterior circulation strokes scanned at either 3T or 1.5T systems were analyzed retrospectively through an automatic perfusion-diffusion mismatch detection algorithm. In addition, four expert raters manually outlined perfusion lesions on time-to-peak (TTP) maps and diffusion lesions on diffusion-weighted images (DWI), and reference perfusion-diffusion mismatch masks were obtained as the areas where at least three experts were in agreement that tissue was part of the perfusion-weighted imaging (PWI) lesion, but not the diffusion lesion. Per-subject analyses of mismatch volumes and mismatch overlap were subsequently performed.

Results: The use of the automatic perfusion-diffusion mismatch detection algorithm resulted in a 4.0 ml mean (standard deviation 28.7 ml) difference in mismatch volume compared to the reference expert consensus (Pearson correlation, r = 0.91, P < 0.0001). The median spatial agreement was 0.71, with an interquartile range of 0.28.

Conclusion: We demonstrated excellent agreement between the perfusion-diffusion mismatch masks estimated by our proposed automatic algorithm and those achieved by expert consensus.

Keywords: PWI-DWI mismatch segmentation; acute stroke; and decision support; diffusion lesion; perfusion lesion.

Publication types

  • Comparative Study
  • Evaluation Study

MeSH terms

  • Adult
  • Aged
  • Aged, 80 and over
  • Algorithms*
  • Brain / pathology*
  • Clinical Competence
  • Female
  • Humans
  • Image Enhancement / methods
  • Image Interpretation, Computer-Assisted / methods*
  • Magnetic Resonance Imaging / methods*
  • Male
  • Middle Aged
  • Observer Variation
  • Pattern Recognition, Automated / methods*
  • Reproducibility of Results
  • Sensitivity and Specificity
  • Stroke / pathology*