A hybrid algorithm for dental artifact detection in large computed tomography datasets

Computed tomography (CT) is one of the most common medical imaging modalities and the main technology used in radiomics research, the computational voxel-level analysis of medical images. Analysis of CT images is vulnerable to the effects of dental artifacts (DA) caused by metal implants or fillings. Running automated analysis pipelines with uncurated datasets can reduce performance and hamper future reproducibility on new datasets. This work introduces a new tool to detect the location and magnitude of DAs in CT images based on a combination of deep learning and conventional image processing algorithms. We show the utility of this new DA detector through an analysis of the correlations between radiomic features and the location of DAs in 2,319 CT axial volumes. We were able to predict the correct DA magnitude (no, weak or strong artifacts) yielding a Matthews correlation coefficient of 0.73 (p-value=0.0002), achieving the same level of agreement as human labellers. The algorithm was able to identify the location of the DAs in the CT volumes with performance on par with human labellers. Finally, our analysis of radiomic features showed that only when strong DAs were present, the proximity of the tumour to the mouth was highly correlated with specific radiomic features. Our results suggest that removing these features, or removing CT slices containing the DAs, could reduce these unwanted correlations.