engleski

Reliability of image analysis in experimental neurosurgical research: Possible problems with dispersed patters

Image analysis programs are often used in the basic neurosciences, when there is a need for exact calculation of various interconnected graphical patterns. We decided to investigate possible difficulties in processing and interpretation at different levels of the image analysing process, with special emphasis on individual differences in interpretation of pictorial material. We postulated that such differences might influence the final result of image analysis. Two independent investigators analyzed 36 pathohistological specimens of partially regenerated rat sciatic nerve transections, divided into control and treated groups, and stained with osmium tetroxide according to the Swang-Dewenport method. The software program calculated the percentage of transverse section through the nerve that comprises the demyelinated area. The data were statistically analyzed. Comparing the results of image analysis performed by two independent persons, some differences were noted. First, precision of nerve delineation was not performed in the same fashion, which influenced, although not significantly, the final result. Secondly, the "gray level analysis" showed differences in selecting appropriate pictorial patterns among two investigators, more in the control than in the treated groups. These differences were significant among some control groups, perhaps due to a larger amount of degenerated myelin which was labeled in "grey level analysis". However, the overall data from both investigators were not significantly different. We conclude that different investigators may have slightly different approaches in measuring the same parameters of image analysis, which sometimes may significantly influence the final results. That may lead to false-positive results. Therefore, we prefer the individual, rather than a team approach, in image analysis systems.

nije evidentirano

nije evidentirano