Reliability of an image analysis system for quantifying the radiographic trabecular pattern

A reliability evaluation technique was used to examine the reliability of an image analysis system of the trabecular pattern and to determine the contribution of three possible sources of error variance. Two series of radiographs were taken of 14 lumbar vertebral slices (28 radiographs). Every radiograph was placed on a viewing box for digitization four times by a single operator (112 positions of radiographs) and from every position of a radiograph an area of 15 mm x 15 mm was digitized twice (224 samples for analysis). Ten geometrical characteristics of the trabecular pattern were studied and its orientation was analyzed in 12 directions. Reliability was determined by calculating Cronbach's alpha. This design enabled dividing the measurement error (1-alpha) into fractions associated with the X-ray procedure, the operator and the system. Using this reliability evaluation technique, it was found that the orientation variables are more reliable than the geometric variables. It was found that effort to increase the reliability should be directed toward improving the technical procedure of this image analysis system. Also, repeated measurements will increase the reliability. The number of repeated measurements based on a desired reliability can be calculated. This procedure of evaluation gives the opportunity to select a source of error variance which have to be reduced to increase reliability most effectively.