Inter- and intraobserver reproducibility of cerebral atrophy assessment on MRI scans with hemispheric infarcts

Eur Neurol. 1996;36(5):268-72. doi: 10.1159/000117270.


Cerebral atrophy (CA) in stroke patients is associated with poststroke dementia and may reflect underlying neurodegenerative pathology. Therefore, regional CA may be valuable to study in patients who develop poststroke dementia. The aim of this study was to test the reproducibility of a qualitative rating scale of CA on MRI. MRI scans were performed in 50 consecutive patients (age range 19-81) admitted for an acute hemispheric ischemic stroke. CA was assessed on 2 occasions 24 h apart, on axial T2-weighted sequences by 4 independent observers. We evaluated CA in 13 regions on a 0-3 scale. The sum of the subscores was called the CA score (range: 0-39). The level of agreement was expressed by kappa statistics as well as by analysis of variance for interexaminer reproducibility studies. The mean CA scores ranged from 2.8 to 11.0, indicating the low prevalence of CA in this sample. Complete agreement was reached in 41.7% during the first assessment and in 44.1% in the second assessment. The interobserver agreement was moderate in the first session (mean overall kappa: 0.48) and substantial in the second (mean overall kappa: 0.67). The intraobserver agreement was good for all raters (mean kappa: 0.65). Standardized to the range of the scale, standard deviations of the differences between CA scores of the 4 raters in the 2 sessions were 11.1 and 11.2%; within raters it was 4.4%. We conclude that the assessment of CA using this rating scale is possible in stroke patients. It provides regional atrophy measurements and is reproducible when performed by 1 rater.

Publication types

  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adult
  • Aged
  • Aged, 80 and over
  • Atrophy
  • Brain / pathology*
  • Cerebral Infarction / diagnosis*
  • Evaluation Studies as Topic
  • Female
  • Humans
  • Magnetic Resonance Imaging*
  • Male
  • Middle Aged
  • Observer Variation
  • Reproducibility of Results