In European journal of radiology ; h5-index 47.0
PURPOSE : To investigate the potential of texture parameters from opportunistic MRI and CT for the detection of patients with vertebral fragility fracture, to design a decision tree and to compute a Random Forest analysis for the prediction of fracture risk.
METHODS : One hundred and eighty vertebrae of sixty patients with at least one (30) or without (30) a fragility fracture were retrospectively assessed. Patients had a DXA, an MRI and a CT scan from the three first lumbar vertebrae. Vertebrae texture analysis was performed in routine abdominal or lumbar CT and lumbar MRI using 1st and 2nd order texture parameters. Hounsfield Unit Bone density (HU BD) was also measured on CT-scan images.
RESULTS : Twelve texture parameters, Z-score and HU BD were significantly different between the two groups whereas T score and BMD were not. The inter observer reproducibility was good to excellent. Decision tree showed that age and HU BD were the most relevant factors to predict the fracture risk with a 93 % sensitivity and 56 % specificity. AUC was 0.91 in MRI and 0.92 in CT-scan using the Random Forest analysis. The corresponding sensitivity and specificity were 72 % and 93 % in MRI and 83 and 89 % in CT.
CONCLUSIONS : This study is the first to compare texture indices computed from opportunistic CT and MR images. Age and HU-BD together with selected texture parameters could be used to assess risk fracture. Machine learning algorithm can detect fracture risk in opportunistic CT and MR imaging and might be of high interest for the diagnosis of osteoporosis.
Poullain François, Champsaur Pierre, Pauly Vanessa, Knoepflin Paul, Le Corroller Thomas, Creze Maud, Pithioux Martine, Bendahan David, Guenoun Daphne
2022-Dec-06
CT, MRI, Opportunistic, Osteoporosis, Spine, Texture
