A medical image registration algorithm based on a convolutional neural network

Abstract

The invention discloses a medical image registration algorithm based on a convolutional neural network. The algorithm specifically comprises the following steps: data acquisition and preprocessing; Making a neural network training data set; Training a neural network; Performing coarse registration on the neural network model; And fine registration is carried out based on a gray value registrationalgorithm. According to the invention, the convolutional neural network can be skillfully combined with the traditional gray value-based registration algorithm; The invention discloses a convolutionalneural network elimination method in medical image processing. the complete labeled input sample is lacked; Training sample limitation and other problems, the spatial difference between a group of input fixed images and floating images is quickly calculated; Meanwhile, a neural network model rapidly obtains the spatial difference between the input images to be registered, the large spatial difference is reduced, the situation that a local optimal solution can be obtained in the optimization iteration process based on a gray value registration algorithm is avoided, and therefore the spatial difference between the two images to be registered can be rapidly and accurately calculated.

Description

technical field [0001] The invention belongs to the fields of fast and accurate medical image registration and multimodal medical image fusion, and relates to a medical image registration algorithm based on a convolutional neural network. Background technique [0002] Medical image registration refers to seeking a (or a series of) geometric space transformation between two images (or multiple images) of medical images at different times, different scenes, different modalities, or even different subjects. The image can be matched with another image (or other multiple) medical images. This matching enables anatomical points or interest points with the same anatomical structure to have a one-to-one correspondence in the geometric space of two (or other multiple) medical images. Medical image registration is an important topic in medical image analysis, which has important theoretical research and clinical application value. Image registration is often to find the mapping rela...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a medical image registration algorithm based on convolutional neural network, which can skillfully combine convolutional neural network with traditional registration algorithm based on gray value, and eliminate the effect of convolutional neural network on medical images. In processing, due to the lack of fully labeled input samples, limited training samples, etc.

Images