Inter- and intra-rater reliability of computer-assisted planimetry in experimental stroke research.

Highlights • Computer-assisted planimetry is widely used in experimental stroke research. • An acceptable reliability can be achieved after sufficient training of the rater. • Still, even an experienced rater can generate data with serious variation. • Although automated approaches exist, planimetry skills are important in stroke research. • Inter- and intra-rater dependent bias must be considered during study design. Abstract Background Computer-assisted planimetry is widely used in experimental stroke research to assess the size of the ischemic lesion or hemispheric volume. New method Only insufficient data exist on the training required to achieve sufficient reliability in planimetry. Therefore, planimetry was performed over 15 months by two blinded raters who were initially inexperienced in the method. For inter-rater reliability, the hemispheric and lesional volume of 227 male Wistar Unilever rats subjected to middle cerebral artery occlusion were determined in diffusion- and T2-weighted sequences. For the intra-rater agreement, one investigator assessed the hemispheric and lesional volume in 87 T2-weighted sequences twice within a six-week interval. The correlation was calculated using Krippendorff's alpha and Bland-Altman plots illustrated the agreement. Results Inter-rater agreement increased during the first seven weeks and remained at high values (Krippendorff's alpha > 0.88). For intra-rater agreement, Krippendorff's alpha was 0.84 for hemispheric and 0.85 for lesional volume. The Bland-Altman plot indicated solid agreement between raters in the absence of systematic errors. Comparison with existing methods Simplified geometrical models or automated methods for planimetry can be used to determine lesional volume, but both approaches are inappropriate to assess hemispheric volume. Conclusion Computer-assisted planimetry can be an appropriate method to determine hemispheric or ischemic lesion volume in rodents but requires a sufficiently long learning period of approximately two months. Even an experienced investigator can generate data with serious variation. Inter- and intra-rater-dependent bias should be considered during the design and performance of respective studies. [ABSTRACT FROM AUTHOR]