Vertebrae Detection and Localization in CT with Two-Stage CNNs and Dense Annotations
This work addresses a domain-specific medical imaging problem for spine analysis, offering incremental improvements in accuracy.
The paper tackles vertebrae centroid detection and localization in CT scans by introducing a two-stage CNN approach with dense labeling and large anisotropic kernels, achieving a 0.87mm improvement in mean localization accuracy on a public benchmark.
We propose a new, two-stage approach to the vertebrae centroid detection and localization problem. The first stage detects where the vertebrae appear in the scan using 3D samples, the second identifies the specific vertebrae within that region-of-interest using 2D slices. Our solution utilizes new techniques to improve the accuracy of the algorithm such as a revised approach to dense labelling from sparse centroid annotations and usage of large anisotropic kernels in the base level of a U-net architecture to maximize the receptive field. Our method improves the state-of-the-art's mean localization accuracy by 0.87mm on a publicly available spine CT benchmark.