Abstract
Early diagnosis and treatment of patients at high risk of developing fragility fractures is crucial in the management of osteoporosis. The purpose of this study was to investigate whether the shape of the spine as can be observed from lateral X-rays is indicative for the risk of future development of fragility fractures in the spine.
The study included 568 elderly women of whom 455 maintained skeletal integrity during the mean observation period of 4.8 years and 113 sustained at least one vertebral fracture in the same period. At baseline, none of the women had experienced a previous osteoporotic fracture, and the two groups were not significantly different in terms of age (66.2 ± 0.2 vs. 66.1 ± 0.4), spine BMD (0.77 ± 0.004 vs. 0.76 ± 0.008), body weight (64.7 ± 0.4 vs. 64.6 ± 0.8), height (160.6 ± 0.3 vs. 161 ± 0.5), and number of years since menopause. A radiologist annotated the corner points and mid points of the vertebral end plates of each vertebra from L5 to T4 on digitized lateral radiographs taken at the baseline visit. These points together describe a combination of factors characterizing the spinal shape, including the shape and the size of individual vertebral bodies and intervertebral disks, alignment of vertebrae, and spinal curvature. The positions of the points were subsequently used as the input features to train a pattern classification system to discriminate between spines of women maintaining skeletal health and spines sustaining a fracture in the near future (regularized linear discriminant analysis). Applied to an annotated X-ray image of an unfractured spine, this classification model then provides a measure of the probability that the spine will develop a fracture.
In a leave-one-out experiment, in which the classification models were constructed from a training set excluding any images of the patient under study, fracture probability measures were significantly different between the two groups at baseline (0.26 ± 0.02 vs. 0.18 ± 0.006, p < 10-6). Incident fractures could be predicted from the baseline image with 80% accuracy; the area under the ROC curve (AUC) was 0.65, and the odds ratio (OR) for fracture 5.2 [95% CI 2.3, 11.6]. Significant predictive value remained after adjustment for age and spine BMD (p < 10-6, AUC=0.66, OR=2.0 [1.0, 3.9]).Measures of spine shape can predict vertebral fractures in postmenopausal women, independent of age and spine BMD. The herein presented computer based diagnostic tool could be a useful supplement to existing approaches to fracture risk assessment.
The study included 568 elderly women of whom 455 maintained skeletal integrity during the mean observation period of 4.8 years and 113 sustained at least one vertebral fracture in the same period. At baseline, none of the women had experienced a previous osteoporotic fracture, and the two groups were not significantly different in terms of age (66.2 ± 0.2 vs. 66.1 ± 0.4), spine BMD (0.77 ± 0.004 vs. 0.76 ± 0.008), body weight (64.7 ± 0.4 vs. 64.6 ± 0.8), height (160.6 ± 0.3 vs. 161 ± 0.5), and number of years since menopause. A radiologist annotated the corner points and mid points of the vertebral end plates of each vertebra from L5 to T4 on digitized lateral radiographs taken at the baseline visit. These points together describe a combination of factors characterizing the spinal shape, including the shape and the size of individual vertebral bodies and intervertebral disks, alignment of vertebrae, and spinal curvature. The positions of the points were subsequently used as the input features to train a pattern classification system to discriminate between spines of women maintaining skeletal health and spines sustaining a fracture in the near future (regularized linear discriminant analysis). Applied to an annotated X-ray image of an unfractured spine, this classification model then provides a measure of the probability that the spine will develop a fracture.
In a leave-one-out experiment, in which the classification models were constructed from a training set excluding any images of the patient under study, fracture probability measures were significantly different between the two groups at baseline (0.26 ± 0.02 vs. 0.18 ± 0.006, p < 10-6). Incident fractures could be predicted from the baseline image with 80% accuracy; the area under the ROC curve (AUC) was 0.65, and the odds ratio (OR) for fracture 5.2 [95% CI 2.3, 11.6]. Significant predictive value remained after adjustment for age and spine BMD (p < 10-6, AUC=0.66, OR=2.0 [1.0, 3.9]).Measures of spine shape can predict vertebral fractures in postmenopausal women, independent of age and spine BMD. The herein presented computer based diagnostic tool could be a useful supplement to existing approaches to fracture risk assessment.
| Original language | English |
|---|---|
| Journal | Journal of Bone and Mineral Research |
| Issue number | Supl. S |
| Pages (from-to) | S37 |
| Number of pages | 1 |
| ISSN | 0884-0431 |
| Publication status | Published - 2008 |
| Event | 30th Annual Meeting of the American Society for Bone and Mineral Research (ASBMR) - Montreal, Canada Duration: 12 Sept 2008 → 16 Sept 2008 Conference number: 30 |
Conference
| Conference | 30th Annual Meeting of the American Society for Bone and Mineral Research (ASBMR) |
|---|---|
| Number | 30 |
| Country/Territory | Canada |
| City | Montreal |
| Period | 12/09/2008 → 16/09/2008 |