ABSTRACT
INTRODUCTION: Metastases increase the risk of fracture when affecting the femur. Consequently, clinicians need to know if the patient's femur can withstand the stress of daily activities. The current tools used in clinics are not sufficiently precise. A new method, the CT-scan-based finite element analysis, gives good predictive results. However, none of the existing models were tested for reproducibility. This is a critical issue to address in order to apply the technique on a large cohort around the world to help evaluate bone metastatic fracture risk in patients. The aim of this study is then to evaluate 1) the reproducibility 2) the transposition of the reproduced model to another dataset and 3) the global sensitivity of one of the most promising models of the literature (original model). METHODS: The model was reproduced based on the paper describing it and discussion with authors to avoid reproduction errors. The reproducibility was evaluated by comparing the results given in the original model by the original first team (Leuven, Belgium) and the reproduced model made by another team (Lyon, France) on the same dataset of CT-scans of ex vivo femurs. The transposition of the model was evaluated by comparing the results of the reproduced model on two different datasets. The global sensitivity analysis was done by using the Morris method and evaluates the influence of the density calibration coefficient, the segmentation, the orientations and the length of the femur. RESULTS: The original and reproduced models are highly correlated (r2 = 0.95), even though the reproduced model gives systematically higher failure loads. When using the reproduced model on another dataset, predictions are less accurate (r2 with the experimental failure load decreases, errors increase). The global sensitivity analysis showed high influence of the density calibration coefficient (mean variation of failure load of 84 %) and non-negligible influence of the segmentation, orientation and length of the femur (mean variation of failure load between 7 and 10 %). CONCLUSION: This study showed that, although being validated, the reproduced model underperformed when using another dataset. The difference in performance depending on the dataset is commonly the cause of overfitting when creating the model. However, the dataset used in the original paper (Sas et al., 2020a) and the Leuven's dataset gave similar performance, which indicates a lesser probability for the overfitting cause. Also, the model is highly sensitive to density parameters and automation of measurement may minimize the uncertainty on failure load. An uncertainty propagation analysis would give the actual precision of such model and improve our understanding of its behavior and is part of future work.
ABSTRACT
Bone segmentation is an important step to perform biomechanical failure load simulations on in-vivo CT data of patients with bone metastasis, as it is a mandatory operation to obtain meshes needed for numerical simulations. Segmentation can be a tedious and time consuming task when done manually, and expert segmentations are subject to intra- and inter-operator variability. Deep learning methods are increasingly employed to automatically carry out image segmentation tasks. These networks usually need to be trained on a large image dataset along with the manual segmentations to maximize generalization to new images, but it is not always possible to have access to a multitude of CT-scans with the associated ground truth. It then becomes necessary to use training techniques to make the best use of the limited available data. In this paper, we propose a dedicated pipeline of preprocessing, deep learning based segmentation method and post-processing for in-vivo human femurs and vertebrae segmentation from CT-scans volumes. We experimented with three U-Net architectures and showed that out-of-the-box models enable automatic and high-quality volume segmentation if carefully trained. We compared the failure load simulation results obtained on femurs and vertebrae using either automatic or manual segmentations and studied the sensitivity of the simulations on small variations of the automatic segmentation. The failure loads obtained using automatic segmentations were comparable to those obtained using manual expert segmentations for all the femurs and vertebrae tested, demonstrating the effectiveness of the automated segmentation approach for failure load simulations.