Digital Tomosynthesis and Fractal Analysis Predict Prevalent Vertebral Fractures in Patients With Multiple Myeloma: A Preliminary In Vivo Study.
Recommended Citation
Oravec D, Yaldo O, Bolton C, Flynn MJ, van Holsbeeck M, and Yeni YN. Digital Tomosynthesis and Fractal Analysis Predict Prevalent Vertebral Fractures in Patients With Multiple Myeloma: A Preliminary In Vivo Study. AJR Am J Roentgenol 2019; Epub ahead of print.
Document Type
Article
Publication Date
4-11-2019
Publication Title
AJR. American journal of roentgenology
Abstract
OBJECTIVE: The objective of this study was to investigate the association of fractal-derived bone microstructural parameters with vertebral fracture status using in vivo digital tomosynthesis images of the spine. MATERIALS AND METHODS: Digital tomosynthesis images of the thoracic and lumbar spine from T1 to L5 were acquired from 36 patients with newly diagnosed multiple myeloma or monoclonal gammopathy of uncertain significance (age range, 39-85 years old). Scans were performed with patients in the supine position with reconstructed planes formed in the coronal direction. Bone mineral density (BMD) was recorded for 10 patients who had recently undergone dual x-ray absorptiometry. Vertebral fracture and lytic lesion status was determined by a radiologist from digital radiographs. Radiologist interpretation was reviewed to identify levels with a minimum number of fractures or lesions. For fractal analysis, the largest possible cuboid volume of interest within the cancellous bone was cropped from T7 and T11 images. Mean and SD of fractal variables between slices of fractal dimension (FD, a measure of self-similarity in the texture), mean lacunarity (lambda, a measure of heterogeneity) and the slope of lacunarity versus box size relationship (Slambda, a measure of sensitivity of heterogeneity to size scale) were calculated using a box-counting method. A generalized estimating equation (GEE) platform was used to examine fractal variables as predictors of fracture status. RESULTS: Fracture status was not significantly associated with sex, race, age, stage of myeloma, presence of lesion in the spine, or BMD. In light of these results, no correction was made for these variables in further analyses of fractal variables. No interaction was found between vertebral level and any of the fractal variables (p = 0.12-0.77). Therefore, vertebral level was not considered further as an independent variable. Logistic regression analysis within GEE indicated that probability of fracture decreased with increasing mean FD (p = 0.02). In contrast, probability of fracture increased with increasing mean lambda (p = 0.03). Although not to a statistically significant degree, probability of fracture increased with increasing mean Slambda (p = 0.08), SD of FD (p = 0.07), SD of lambda (p = 0.07), and SD of Slambda (p = 0.06). CONCLUSION: We found FD and lacunarity calculated within the cancellous centrum of T7 and T11 vertebrae to be significantly associated with the presence of a vertebral fracture in this cohort. The decreased probability of fracture with increasing fractal dimension and increased probability of fracture with increasing lacunarity are consistent with the idea that cancellous bone with a better organized trabecular architecture is mechanically more competent. To our knowledge, this is the first in vivo evidence that fractal analysis of vertebral bone from tomosynthesis images may be useful in assessing vertebral fracture risk in patients with multiple myeloma.
PubMed ID
30973772
ePublication
ePub ahead of print
