Estimating motion artifacts of HR-pQCT scans using automatically derived standard parameters of bone structure- and density-quantification

Motion grading in high-resolution peripheral quantitative computed tomography (HR-pQCT) is a critical task for data validity, dependent on the operator. In routine clinical practice, validating the grading performed by technical staff is challenging. This is of critical importance since motion also affects the derived results of HR-pQCT. To address these issues, we focused on using established clinical HR-pQCT-derived structure-density parameters to predict whether imaging samples are motion corrupted. Using a dataset of 1000 scans from the tibia and radius, we employed a logistic regression model to classify whether scans are corrupted or not, condensing the task into a binary problem. Additionally, we evaluated the performance specifically for patients with altered bone microstructure.We achieved a high sensitivity with an average of 0.7. Our results suggest that motion-corrupted scans affect derived parameters differently than disease-related changes in bone microstructure. Thresholds for model trust were established to estimate the reliability of our predictions. Furthermore, feature importance was assessed, aligning with previous work on motion artifacts.Thus, we provide a low (computational) cost, efficient, and robust internal quality-control approach for grading derived data validity in clinical routine. Additionally, we enable cleaning large cohort datasets, enabling a retrospective evaluation without the need of image data retrieval or handling. By adding conformity thresholds, users can optimize their workflow to allow for high data quality.