Model training method, model training device, medical image registration method and medical image registration device

Abstract

The invention provides a model training method, which is used for iteratively training an initial neural network model to obtain a non-rigid registration model, and model constraint functions utilized in the iteration training process comprise a global similarity loss function, a smooth loss function and a hepatic artery loss function. The training method comprises the following steps: based on an initial neural network model, obtaining registration data of current training by using a pre-registered arterial phase liver image sequence sample and a portal vein phase liver image sequence sample; determining a loss function value of the current training based on the model constraint function by using the registration data of the current training, the registration data of the last training corresponding to the current training, the portal vein stage liver image sequence sample and the corresponding portal vein segmentation data; and adjusting parameters of the initial neural network model based on the loss function value of the current training, and carrying out loop iteration training on the initial neural network model until the determined loss function value meets a preset condition, thereby obtaining a non-rigid registration model.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to a model training method and its device, a medical image registration method and its device, electronic equipment, and a computer-readable storage medium. Background technique [0002] In the prior art, in order to facilitate doctors to accurately observe and analyze images, quickly locate the lesion area, and provide effective reference for the diagnosis of liver disease and liver surgery planning, it is necessary to register the arterial phase liver image sequence to the portal venous phase liver image sequence . [0003] However, since the internal liver arteries in the portal venous phase liver image sequence are almost invisible, and the internal liver arteries are small and easily affected by structures such as masses, therefore, in the process of registering the arterial phase liver image sequence to the portal venous phase liver image sequence Among the...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the embodiment of the present application provides a model training method, a model training device, a medical image registration method, a medical image registration device, electronic equipment, and a computer-readable storage medium to solve the problem of arterial In the process of registering the liver image sequence in the portal phase to the liver image sequence in the portal phase, the intrahepatic artery is difficult to register and the registration effect is not ideal.

Images