Performance of statistical models of shape and appearance for semiautomatic segmentations of spinal vertebrae T4-L4 on digitized vertebral fracture assessment images.

BACKGROUND CONTEXT: Despite its clinical importance, accurate identification of vertebral fractures is problematic and time-consuming. There is a recognized need to improve the detection of vertebral fractures so that appropriate high-risk patients can be selected to initiate clinically beneficial therapeutic interventions. PURPOSE: To develop and evaluate semiautomatic algorithms for detailed annotation of vertebral bodies from T4 to L4 in digitized lateral spinal dual-energy X-ray absorptiometry (DXA) vertebral fracture assessment (VFA) images. STUDY DESIGN: Using lateral spinal DXA VFA images from subjects imaged at University Hospital fracture liaison service, image algorithms were developed for semiautomatic detailed annotation of vertebral bodies from T4 to L4. PATIENT SAMPLE: Two hundred one women aged 50 years or older with nonvertebral fractures. OUTCOME MEASURES: Algorithm accuracy and precision. METHODS: Statistical models of vertebral shape and appearance from T4 to L4 were constructed using VFA images from 130 subjects. The resulting models form a part of an algorithm for performing semiautomatic detailed annotation of vertebral bodies from T4 to L4. Algorithm accuracy and precision were evaluated on a test-set of 71 independent images. RESULTS: Overall accuracy was 0.72 mm (3.00% of vertebral height) and overall precision was 0.26 mm (1.11%) for point-to-line distance. Accuracy and precision were best on normal vertebrae (0.65 mm [2.67%] and 0.21 mm [0.90%], respectively) and mild fractures (0.78 mm [3.18%] and 0.32 mm [1.39%], respectively), but accuracy and precision errors were higher for moderate (1.07 mm [4.66%] and 0.48 mm [2.15%], respectively) and severe fractures (2.07 mm [9.65%] and 1.10 mm [5.09%], respectively). Accuracy and precision results for the algorithm were comparable with other reported results in the literature. CONCLUSIONS: This semiautomatic image analysis had high overall accuracy and precision on normal vertebrae and mild fractures, but performed less well in moderate and severe fractures. It is, therefore, a useful tool to identify normality of vertebral shape and to identify mild fractures.

Methods and reliability of radiographic vertebral fracture detection in older men: the osteoporotic fractures in men study.

We describe the methods and reliability of radiographic vertebral fracture assessment in MrOS, a cohort of community dwelling men aged ≥65yrs. Lateral spine radiographs were obtained at Visit 1 (2000-2) and 4.6years later (Visit 2). Using a workflow tool (SpineAnalyzer™, Optasia Medical), a physician reader completed semi-quantitative (SQ) scoring. Prior to SQ scoring, technicians performed "triage" to reduce physician reader workload, whereby clearly normal spine images were eliminated from SQ scoring with all levels assumed to be SQ=0 (no fracture, "triage negative"); spine images with any possible fracture or abnormality were passed to the physician reader as "triage positive" images. Using a quality assurance sample of images (n=20 participants; 8 with baseline only and 12 with baseline and follow-up images) read multiple times, we calculated intra-reader kappa statistics and percent agreement for SQ scores. A subset of 494 participants' images was read regardless of triage classification to calculate the specificity and sensitivity of triage. Technically adequate images were available for 5958 of 5994 participants at Visit 1, and 4399 of 4423 participants at Visit 2. Triage identified 3215 (53.9%) participants with radiographs that required further evaluation by the physician reader. For prevalent fractures at Visit 1 (SQ≥1), intra-reader kappa statistics ranged from 0.79 to 0.92; percent agreement ranged from 96.9% to 98.9%; sensitivity of the triage was 96.8% and specificity of triage was 46.3%. In conclusion, SQ scoring had excellent intra-rater reliability in our study. The triage process reduces expert reader workload without hindering the ability to identify vertebral fractures.

Comprehensive vertebral deformity and vertebral fracture assessment in clinical practice: intra- and inter-reader agreement of a clinical workflow tool.

STUDY DESIGN: Study design randomized intra- and inter-reader reproducibility study. OBJECTIVE: To evaluate reproducibility of quantitative morphometry (QM) and agreement of dichotomous fracture/no-fracture status on lateral spinal radiographs acquired during routine clinical practice using a clinical workflow tool. SUMMARY OF BACKGROUND DATA: Several recent guidelines have underlined the importance of Genant semi-quantitative scoring and selective QM to confirm and grade suspected vertebral fractures in clinical practice. METHODS: Thoracic and lumbar spine radiographs were acquired from 98 consecutive subjects (mean age, 60.1 ± 11.7 yr) attending the clinic for osteoporosis evaluation. For each subject, QM and Genant semi-quantitative scoring were performed on all evaluable vertebrae from L4 to T4 using a software workflow tool. A radiologist and an experienced radiographical technician performed 2 repeat reading sessions of the radiographs 12 months apart, blinded to each other's results; for the second read, the cases were anonymized and the order was randomized. RESULTS: Inter-reader reproducibility results were 3.1% and 3.2% coefficient of variation (%) for heights, 0.030 and 0.031 root mean square standard deviation for height ratios. For intrareader reproducibility, these values were 2.2% and 3.5% coefficient of variation %; 0.023 and 0.034 root mean square standard deviation. Kappa score results for agreement of dichotomous fracture/no-fracture status were 0.67 and 0.72 (inter-rater) and 0.50 and 0.67 (intrarater). CONCLUSION: The software assessed in this study is a reliable clinical tool that facilitates QM and Genant semi-quantitative scoring of the spine in routine clinical practice. LEVEL OF EVIDENCE: 3.