ECE2022 Poster Presentations Calcium and Bone (68 abstracts)
Artificial intelligence based on radiomic analysis of lumbar spine computed tomography (ct) scan may improve accuracy in detecting osteoporosis
Emilia Biamonte 1 , Federico Garoli 2 , Riccardo Levi 3 , Walter Vena 1 , Flaminia Carrone 1 , Simona Jafaar 1 , Maurizio Fornari 4 , Marco Grimaldi 2 , Letterio Politi 2 , Andrea Lania 1 & Gherardo Mazziotti 1
1Humanitas Research Hospital, Endocrinology, Diabetology and Andrology Unit, Rozzano, Italy; 2Humanitas Research Hospital, Radiology, Rozzano, Italy; 3Humanitas Research Hospital, Rozzano, Italy; 4Humanitas Research Hospital, Neurosurgery, Rozzano, Italy
Background: Osteoporosis is characterized by reduced bone mass and a compromised bone microstructure, leading to increased bone fragility and fracture risk. Currently, the gold standard for diagnosis is the bone mineral density (BMD) measurement by DXA. However, approximately half of fragility fractures occurs in the context of normal or slightly decreased BMD values.
Protocol: In this cross-sectional study, we performed an artificial intelligence (AI)-based analysis on radiomic of images of opportunistic computed tomography (CT) of lumbar spine in 240 consecutive subjects (mean age 61±14,5, 130 males). Exclusion criteria were: 1) bone-active drugs; 2) neoplastic diseases; 3) spine surgical intervention; 4) spine trauma. Fifty-eight subjects had vertebral fractures (VFs) as assessed by a morphometric approach on CT or XR-ray spine (D4-L4) images. On CT images, the ROI was acquired as a 3D-spherical region of 9 mm in the middle of non-fractured lumbar vertebral bodies. A total of 93 RF were extracted: 19 first-order and 74 textural features. The most discriminative ones were selected by applying bootstrap recursive feature elimination procedures with random sampling for train/test split (100 iterations). The Linear Support Vector (LSV) model was adopted, and the Tree-Parzen Estimator Bayesian approach was employed. Results were evaluated on a stratified test set (25% of the total population), not included in the training phase. The final model was evaluated on the test set, using accuracy, sensitivity, specificity, and area under the ROC curve.
Results: Univariate analysis showed 20 significative RF (P<0.05), used to develop the LSV model. The model reached 0.83 of ROC, and the 71,7%, 78.0%, and 69,6 % of accuracy, sensitivity, and specificity respectively. Patients with VFs had significatively lower first-order features compared with those without VFs and were associated with textural features denoting a bone microarchitecture more rarefied and with higher inter-trabeculae distance. Furthermore, patients with a more compromised spine (SDI ≥ 2) had significatively lower first-order features compared with those without or with a mild VFs (SDI 0 and SDI 1) and, conversely, were significantly associated with textural features denoting a bone structure more rarefied and less coarse.
Conclusions: Artificial intelligence-based radiomic of lumbar CT scans identifies patients with skeletal fragility. If confirmed, these results may suggest that radiomics could be an important diagnostic tool in osteoporosis detection and in fragility fracture prediction in the next future.
Article tools
My recent searches
My recently viewed abstracts
Authors
Endocrine Abstracts
ISSN 1470-3947 (print) | ISSN 1479-6848 (online)
© Bioscientifica 2024 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more