Predictive modeling of vertebral compression fracture healing using radiomics and PMMA–nanocomposite properties

Vertebral compression fractures (VCFs) cause a significant clinical problem especially in osteoporotic groups of patients, whose delayed or incomplete recovery may result in chronic pain and diminished quality of life. After vertebral augmentation, precise estimation of fracture healing is not an easy task as bone morphology and biomaterial properties interact in a complex way. This paper suggests a predictive modeling system to combine radiomics aspects of vertebral imaging with physicomechanical attributes of PMMA-nanocomposite bone cement to predict the outcome of fracture healing. A carefully constructed dataset of 450 samples containing both quantitative radiomics descriptors such as texture, shape, and intensity, and PMMA characteristics such as elastic modulus, porosity, and nanoparticle concentration were created. The main outcome variable was a continuous fracture healing score. Exploratory data analysis showed that the variability was under control, feature distributions were normalized and correlation between predictors and the healing outcome was moderate, thus, the data is appropriate to multivariate and machine-learn methods. Correlation analysis indicated that both radiomics and PMMA related variables have a significant contribution to prediction of healing with the cement elastic modulus being the highest associated. The format of the dataset facilitates the exploration of the synergistic interaction between imaging biomarkers and biomaterial properties, which makes it possible to develop and verify the models. The offered data framework indicates the opportunity to integrate radiomics and nanocomposite material features to enhance the prognostic accuracy in the treatment of VCF. This will offer a basis of individual treatment planning and optimization of biomaterial design in vertebral augmentation procedures.