Cone beam computed tomography image feature extraction and corresponding method

Abstract

The invention discloses a cone beam computed tomography image feature extraction and corresponding method. The method comprises the following steps: calculating to obtain a super voxel decomposition and spectral space orthogonal basis function of a cone beam CT image; 2) establishing a task-oriented deep neural network corresponding to the volume image and performing pre-training to obtain task-oriented deep neural network parameters corresponding to the volume image; training a task-oriented deep neural network corresponding to the volume image; and utilizing the trained task-oriented deep neural network corresponding to the volume image to realize feature extraction and correspondence of the to-be-processed cone beam CT image. By adopting the technical scheme provided by the invention, three-dimensional cone beam computed tomography image correspondence and registration can be quickly established, and further, computer-assisted intraoperative intervention and online feature point positioning and segmentation label migration can be carried out.

Description

technical field [0001] The invention relates to the technical fields of oral clinical medical image processing and computer vision, in particular to a method for extracting and corresponding features of cone-beam computed tomography images. Background technique [0002] Cone beam computed tomography (cone beam CT) images are used in many clinical fields, especially in orthodontic surgery. The dense correspondence of cone-beam CT images is a key technology for statistical shape analysis and measurement of tissue structural deformation due to growth and development or clinical treatment. Cone-beam CT image correspondence algorithms are the basis for applications such as attribute transfer and label transfer. In recent years, many hand-designed features have been applied to volume images, including gray distribution histograms, directional gradient features, features based on self-similarity, scale-invariant features, and contextual features with modality invariance. However,...

AI Technical Summary

Problems solved by technology

However, the above methods all rely on hand-designed image features, which are usually not specific to the task of solving image correspondences.

At the same time, the above methods are all supervised methods, which require data labeling in advance, while manual data labeling is not only time-consuming and laborious, but also has subjective errors

It can be seen that the existing technology lacks the use of unsupervised methods, and can effectively extract the cone beam CT image features specific to the corresponding task, and at the same time calculate the image feature extraction and corresponding method corresponding to the cone beam CT image

Images