Abstract
Currently, the diagnosis of a vertebral fracture in clinical trials relies on a 20% decrease in body height of a vertebra as measured on digitized lateral X-rays. Due to the categorical nature of this study parameter, clinical trials continue to demand a large number of participants and long follow-up time. The aim of the present study was to obtain initial experience with a novel computerized technique taking into account morphometric alterations over the full shape of the vertebra.
The proposed method incorporates two key features: 1) A measure of fracture is based on changes in the shape of the vertebral body 2) The changes are quantified with reference to a predicted normal shape of the respective vertebra, incorporating data on the biological variation of intact vertebrae. The present analysis was based on 91 randomly chosen radiographs of the lumbar spine (L1-L4). Radiographs were digitalized and subsequently annotated with pre-defined landmarks. The statistical models of the normal vertebral shape were derived from a dataset of unfractured vertebrae to predict the normal shape of one vertebra from the shape of adjacent vertebrae (regularized multiple linear regression/ridge regression).
The regression model accurately predicts the shape of the intact vertebra. In 95 % of the cases was the prediction error <5% of the mean vertebral height with an average prediction error of 2.7% (<0.8mm). The obtained prediction accuracy is close to the maximum that can be expected with the current model. The prediction error gives the deviation from the expected intact shape and can therefore be directly used to assess the degree of abnormality for each vertebra. Presence of fracture had a mean prediction error of 11.7%. An expample of fracture with reference to predicted normal shape is shown in Figure 1.
In conclusion, this novel computerized technique carries notable potentials for the quantitative monitoring of vertebral deformations. Accurate measures of the changes in vertebral shape might provide us with continuous measures of vertebral deformation, which in turn could decrease number of patients and/or follow-up time in clinical trials assessing the efficacy of drug candidates.
The proposed method incorporates two key features: 1) A measure of fracture is based on changes in the shape of the vertebral body 2) The changes are quantified with reference to a predicted normal shape of the respective vertebra, incorporating data on the biological variation of intact vertebrae. The present analysis was based on 91 randomly chosen radiographs of the lumbar spine (L1-L4). Radiographs were digitalized and subsequently annotated with pre-defined landmarks. The statistical models of the normal vertebral shape were derived from a dataset of unfractured vertebrae to predict the normal shape of one vertebra from the shape of adjacent vertebrae (regularized multiple linear regression/ridge regression).
The regression model accurately predicts the shape of the intact vertebra. In 95 % of the cases was the prediction error <5% of the mean vertebral height with an average prediction error of 2.7% (<0.8mm). The obtained prediction accuracy is close to the maximum that can be expected with the current model. The prediction error gives the deviation from the expected intact shape and can therefore be directly used to assess the degree of abnormality for each vertebra. Presence of fracture had a mean prediction error of 11.7%. An expample of fracture with reference to predicted normal shape is shown in Figure 1.
In conclusion, this novel computerized technique carries notable potentials for the quantitative monitoring of vertebral deformations. Accurate measures of the changes in vertebral shape might provide us with continuous measures of vertebral deformation, which in turn could decrease number of patients and/or follow-up time in clinical trials assessing the efficacy of drug candidates.
| Originalsprog | Engelsk |
|---|---|
| Titel | Abstracts of the 27th Annual Meeting of the American Society for Bone and Mineral Research |
| Forlag | ASMBR |
| Publikationsdato | 2005 |
| Status | Udgivet - 2005 |
| Udgivet eksternt | Ja |
| Begivenhed | American Society for Bone and Mineral Research (ASBM) 27th Annual Meeting - Nashville, Tennessee, USA Varighed: 29 nov. 2010 → … Konferencens nummer: 27 |
Konference
| Konference | American Society for Bone and Mineral Research (ASBM) 27th Annual Meeting |
|---|---|
| Nummer | 27 |
| Land/Område | USA |
| By | Nashville, Tennessee |
| Periode | 29/11/2010 → … |