Abstract Background In recent years, Radiomics features (RFs) have been developed to provide quantitative, standardized information about shape, density/intensity and texture patterns on radiological images. Several studies showed limitations in the reproducibility of RFs in different acquisition settings. To date, reproducibility studies using CT images mainly rely on phantoms, due to the harness of patient exposure to X-rays. In this study we analyze the effects of CT acquisition parameters on RFs of lumbar vertebrae in a cadaveric donor. Methods 112 unique CT acquisitions from cadaveric truck were performed on 3 different CT scanners varying KV, mA, field of view and reconstruction kernel settings. Lumbar vertebrae were segmented through a deep learning convolutional neural network and RFs were computed. The effects of each protocol on each RFs were assessed by univariate and multivariate Generalized Linear Model. Further, we compared the GLM model to the ComBat algorithm in the efficiency in harmonizing CT images. Findings From GLM, mA variation was not associated with alteration of RFs , whereas kV modification was associated with exponential variation of several RFs, including First Order (94.4%), GLCM (87.5%) and NGTDM (100%). Upon cross-validation, ComBat algorithm obtained a mean R2 higher than 0.90 in 1 RFs (0.90%), whereas GLM model obtained high R2 in 21 RFs (19.6%), showing that the proposed GLM could effectively harmonize acquisitions better than ComBat. Interpretation This study represents the first attempt in describing the effects of CT acquisition parameters in bone RFs in a cadaveric donor. Our analyses showed that RFs could be substantially different according to the variation of each acquisition parameter and in dataset obtained from different CT scanners. These differences can be minimized using the proposed GLM model. Publicly available dataset and GLM could foster the research of Radiomics-based studies by increasing harmonization across CT protocols and vendors.