Research And Application On Three Dimensional Reconstruction And Semantic Retrieval Of Low-dose CT Image

With the rapid development of medical imaging technology, medical imaging is playing a more and more important role in the diagnos and treatment of disease. CT image reconstruction is a computer-aided diagnostic method which can improve efficiency and diagnostic accuracy of the physicians. But high-dose CT imaging are proved to have great impact on the health of patients, so the low-dose CT imaging is widely used in many inspection items. However, the low-dose CT imaging also means that the interval of image sequence would be increased, leading to declining quality of three-dimensional reconstruction. In addition, the current CT images retrieval ways are very backward, most of which are based on definite keywords. CT image retrieval based on semantic is closer to the people’s understanding of images, which can greatly improve the CT images retrieval efficiency, and the CT image retrieval has gradually become the focus of current researches.Aiming at the problem of three dimensional reconstruction with low-dose CT images, this paper studied the main technology in three dimensional reconstruction based on liver CT images, and compared the algorithms of image preprocess, image segmentation and three dimensional rendering through simulation. Aiming at the problem of big image sequence interval in low-dose CT images, this paper studied the interpolation algorithms in-depth, and improved the image interpolation algorithm based shape and profile by image registration, optimizing the point pairs matching process and improving the linear interpolation. The modified interpolation algorithm has higher efficiency and accuracy, which is more suitable for three dimensional reconstruction of low-dose CT images.Aiming at problem of CT image retrieval,this paper studied the related technologies of semantic extraction, and extracted the semantic information of liver CT images by the method of combining extracting semantic information from DICOM files’metadata and mapping image low-level features to semantic information. Firstly, this paper studied the information model of DICOM standard and the method of DICOM files parsing, and extracted semantic information from metadata of DICOM files directly. Secondly, this paper studied the method of extracting low-level features of medical images and SVM classifier, and compared the semantic mapping effects of different low-level features and feature combinations,then obtained the semantic mapping model by the best feature combination and the SVM classifier. This model obtained superior semantic mapping accuracy, which can solve the semantic gap problem.