Abdominal CT image target organ registration method based on deep learning

Abstract

The invention discloses an abdominal CT image target organ registration method based on deep learning. Firstly, constructing an abdomen CT image database; secondly, constructing a network model basedon deep learning, and introducing a coordinate convolution layer into a convolutional neural network module of the network model so as to enhance the learning ability of the network model for target position information; then, considering that the data volume of the abdominal CT image containing the target organ bounding box is small, based on the transfer learning technology, inputting the data into a natural scene database to pre-train a network model, and then inputting the data into an abdominal CT image database to perform parameter fine adjustment on the model so as to realize abdominaltarget organ detection; and finally, constructing an abdominal target organ CT image pair, constructing a similarity measurement function according to gradient and gray level distribution characteristics between pixel points of the image pair, minimizing the function based on a gradient descent method, and realizing registration of the abdominal CT image to the target organ. According to the method, the strategy of firstly extracting the target organ area of the abdominal CT image and then registering is adopted, the influence of factors such as complex background and noise of the abdominal CTimage on target organ registration is reduced, the registration precision is high, and the robustness is high.

Description

technical field [0001] The invention belongs to the field of medical image processing, and relates to the registration of multiple organs in abdominal CT images, in particular to the registration of target organs in abdominal CT images based on deep learning. Background technique [0002] Image registration is an important technology in modern computer vision and medical image processing, and its purpose is to compare or fusion. Registration can be used to assist in target organ segmentation, 3D reconstruction, tissue parameter estimation, and respiratory motion tracking in the abdomen. The current abdominal CT image registration methods have problems such as large amount of calculation, long time consumption and poor robustness in clinical application. Therefore, it is of great significance to study an efficient and accurate registration method of abdominal CT images for disease diagnosis and radiation therapy of abdominal organs. [0003] Existing abdominal image regist...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the problems of complex background and large noise in the abdominal CT image, and the large amount of calculation and long time consumption in the construction of the similarity measure in the registration process, using a two-step method of first extracting the target organ area of ​​​​the abdominal CT image and then registering strategy, an efficient, accurate and robust method for target organ registration in abdominal CT images based on deep learning is proposed

Images