Automated Segmentation of CBCT Image with Prior-Guided Sequential Random Forest

A major limitation of CBCT scans is the widespread image artifacts such as noise, beam hardening and inhomogeneity, causing great difficulty for accurate segmentation of bony structures from soft tissues, as well as separation of mandible from maxilla. In this paper, we present a novel fully automated method for CBCT image segmentation. Specifically, we first employ majority voting to estimate the initial probability maps of mandible and maxilla. We then extract both the appearance features from CBCTs and the context features from the initial probability maps to train the first-layer of classifier. Based on the first-layer of trained classifier, the probability maps are updated, which will be employed to further train the next layer of classifier. By iteratively training the subsequent classifier and the updated segmentation probability maps, we can derive a sequence of classifiers. Experimental results on 30 CBCTs show that the proposed method achieves the state-of-the-art performance.