Vertebral morphometry derived from digital images.

Abstract

We describe a method for capturing measurement data directly from digitized images using specialized software and high-resolution workstations. We have evaluated the reliability, accuracy, and reproducibility of this method in an international clinical trial involving vertebral morphometry. Accuracy was determined using clinical radiographs measured with vernier calipers and a film phantom. Intra- and interobserver variabilities were assessed, and longitudinal reproducibility was evaluated. As part of the trial, spinal radiographs were collected from more than 200 international health care facilities and digitized at four screening centers. Digitized images were stored and sent to our central facility for morphometry and archiving. Timeliness and variability of the process were tracked. Relative accuracy was nearly 100%. Correlation with clinical measurements was high (r = .96; p < .05). The mean coefficient of variation for interobserver variability was 2%. Intraobserver variation was 3-5%. The coefficient of variation for longitudinal reproducibility ranged from 4% to 6%. After 9 months of operation, our trial included 9494 patients. Of approximately 36,000 radiographs, 98% passed quality review. Only 1% of vertebral levels were not measurable. Hardware and software problems were minimal. The use of digitized images for morphometry is accurate, reproducible, and convenient. When applied to a large-scale clinical trial, it offers unique advantages that may justify the cost and complexity that exceed those of conventional radiographs.