An organ automatic sketching method based on deep integrated learning

Abstract

The invention discloses an organ automatic sketching method based on deep integration learning, and the method specifically comprises the following steps: S1, carrying out the preparation work of a CTimage and a corresponding Mask image before the image segmentation, and enabling the Mask image to be a sketched organ image; S2, after the preparation work in the step S1 is completed, automaticallysketching the organs in the single direction to obtain an automatic sketching result of each organ in the single direction; And S3, completing multi-azimuth automatic sketching integration accordingto a result obtained in the step S2 by utilizing an integration method. According to the method, integration of multi-azimuth automatic sketching is completed by utilizing an integration method, a traditional single-azimuth automatic segmentation method is broken through, and the automatic sketching precision of organs is further improved.

Description

technical field [0001] The invention belongs to the field of medical imaging and deep learning, and mainly relates to a method for automatically drawing organs based on deep integrated learning. Background technique [0002] Organ delineation is an important preparation for radiation therapy. However, the task of organ delineation is large and repetitive, and manual delineation takes a long time and the delineation accuracy mainly depends on the doctor's experience. In order to automatically outline the required organs quickly and uniformly, researchers have developed or applied many methods. At present, the best effect is the convolutional neural network (CNN) of deep learning. We input the existing CT images and the existing organ outlines, and train the network to predict the outline of each organ based on the CT images. The training process does not require manual intervention. The network can gradually adjust internal parameters according to the input and output until...

AI Technical Summary

Problems solved by technology

The 3D features of the image can be extracted using a 3D neural network, but judging from the current computer memory capacity, only 3 to 5 images with a size of 512*512 can be used in some more complex networks.

To extract more interlayer information of CT images within the allowable range of computer memory, it is necessary to reduce the size of the image or reduce the complexity of the model, which will reduce the accuracy of image segmentation

[0004] To sum up, in the automatic segmentation method of three-dimensional body, the extraction of three-dimensional features is difficult to realize, and the automatic delineation accuracy of organs is limited

Images