Anatomical Landmark Localization For Cephalometric X-ray Image By Using Context- Aware Two-layer Regression Forests

As an indispensable tool for modern orthodontics,orthognathic treatment and maxillofacial surgery,cephalometric analysis provides valuable information of patients’ bony,dental and soft tissue structures.When conducting a routine analysis in a 2D cephalometric X-ray images,i.e.lateral cephalograms,orthodontists need to manually identify anatomical landmarks from which some cephalometric tracing clinical measurements are calculated.Then,anatomical abnormalities can be diagnosed by comparing these measurements of patients with those of the normal.However,manual marking usually suffers from inter-and intra-observer variability,which affects the accuracy of cephalometric analysis.Besides,another disadvantage of manual marking is time-consuming.It will take an experienced orthodontist more than 20 minutes to identify anatomical landmarks on a 2D cephalometric X-ray images.Thus,there is a clinical need to develop automated identification of anatomical landmarks on lateral cephalograms.To automatically detect cephalometric landmarks in dental X-Ray images,a context-aware landmark detection method using two-layer regression forest models is proposed.First,it extracts the appearance features from the XRay images to train first-layer regression forest model which can be used to generate a displacement map for each landmark per training image.From the displacement maps,the context features are computed and combined with appearance features to train second-layer regression forest.Exerting the trained two-layer model on the new dental X-Ray images to be processed will produce the displacement map for each target landmark.Finally,the proposed method uses regression voting to acquire the landmark position in the testing image.Experimental results show that the proposed method has good performance in detection of cephalometric landmarks in dental X-Ray images.Then,in order to facilitate the use of non-developers,a set of automatic positioning system for anatomical landmarks of X-ray cephalometric images was developed based on Python language.The system integrates the training and testing functions of the model while modifying various parameters in the training and testing.In order to make the test results more friendly,a visualization module has been added to directly mark the test results on the test image.In order to improve the processing speed of the software,the Python GIL lock is removed,and the resources of the multi-core CPU are fully utilized.