Cascaded regression for CT slice localization

Automated computational tools are needed to estimate the position of a slice of interest within a contiguous stack of slices. Such estimation is useful to retrieve relevant slices from a volume of slices in clinical analysis or it can be used as an initialization step to other post-processing and image analysis techniques. In this paper, we present a novel method to determine the location of a slice of interest within a given volume by formulating it as a regression problem. The input variables for the regression are obtained from simple intensity features computed from a pyramid representation of the slice. We assess the performance of the proposed method by comparing the estimated positions of slices of interest in CT data with manual annotations. Our method was validated on a dataset of 45 volumes and promising results were obtained for 5 different target slices, the average error being 2 slices.