Skull image segmentation method based on deep iterative fusion deep learning model

Abstract

The invention discloses a skull image segmentation method based on a deep iterative fusion deep learning model. The method comprises the following steps of 1, selecting a brain magnetic resonance image to be segmented and preprocessing the brain magnetic resonance image; 2, randomly selecting a training set and a test set for the preprocessed brain magnetic resonance image; 3, constructing a deepiterative fusion learning model, wherein the model comprises a convolution module, a residual module and an up-sampling module; 4, performing internal data enhancement on the training set, wherein theinternal data enhancement comprises random scaling and random clipping; and 5, training a model by using the data after data enhancement, and finally testing the test set by using the trained model,so that skull segmentation can be performed on the brain magnetic resonance image. According to the characteristics of strong automatic learning and feature extraction capabilities of the convolutional neural network, a deep iterative fusion model is designed, so that the model has relatively high generalization capability and relatively high segmentation precision on a plurality of data, and a very good effect is achieved.

Description

Technical field[0001]The invention relates to a skull image segmentation method, in particular to a skull image segmentation method based on a deep iterative fusion deep learning model, and belongs to the technical field of medical image processing and deep learning.Background technique[0002]In brain image analysis, skull segmentation is a very important image preprocessing step. Segmenting the skull from brain magnetic resonance images is an important basis for subsequent feature extraction, image analysis and prediction, and clinical diagnosis. For example, in brain connection analysis, brain parenchymal regions need to be extracted to realize accurate connection analysis of brain function and brain structure, laying the foundation for subsequent accurate typing, classification and staging. Traditional skull segmentation is generally performed manually by professional doctors with rich experience, which often requires a lot of time and energy, and the segmentation results are exce...

AI Technical Summary

Problems solved by technology

But the 3D network model has too many parameters, the training time is slow and the test process has to load too many parameters

[0004] Although these deep learning models have achieved good segmentation results in brain magnetic resonance image skull segmentation, there is still a common problem, that is, the generalization ability is average, that is to say, it performs well on one data set, but not on another. Poor performance on one data set, and the segmentation accuracy still needs to be improved

Images