The Design Of Analysis System Of Automated Cephalometry

With the development and popularization of artificial intelligence technology,more and more machine learning algorithms have been applied in medical image processing,and have helped im-prove the efficiency of manual reading.In the digital oral field,X-ray film cephalometry is a key part of clinical applications such as dental implants and orthodontic.Automatically marking the landmarks of the X-ray film can greatly save the doctor’s measurement time and improve the diag-nostic accuracy and efficiency.In this thesis,we research and analyze the related machine learning algorithms,and on the basis of the improved key algorithms,develope and implement a prototype analysis system of automated cephalometry.The main work of this thesis includes:Firstly,we proposed a landmark detection algorithm based on random forest and shape regression.we create a random forest using local features of each landmark to generate local binary features,and then calculate a global linear regression matrix and uses a cascade method to predict the position of the landmarks which saves the time for manually marking work.Secondly,we proposed a landmark detection algorithm based on the regions with convolutional neural network,we use transfer learning fine-tuning the deep convolutional neural network model pre-trained on the large public dataset,making it possible to detect target in images which is more accurate than the former model.Then,we mix the models,add the latter model to the former one to improve the overall accuracy of the results.Finally,we developed a prototype analysis system of automated cephalometry and verified it with specific examples.