Semi-supervised renal artery segmentation method based on dense bias network and auto-encoder

Abstract

The invention discloses a semi-supervised renal artery segmentation method based on a dense bias network and an auto-encoder. The method comprises the following steps: for an existing abdominal CT angiography image, segmenting a kidney region in the image to obtain a region-of-interest image, marking the region-of-interest image to obtain a real mask of a renal artery, and forming a supervised training data set and an unsupervised training data set; inputting the unsupervised training data set into a three-dimensional convolution denoising auto-encoder for image reconstruction training to obtain a trained denoising auto-encoder model; inputting the supervised training data set into a denoising auto-encoder model to obtain prior anatomical features of each image, and inputting the prior anatomical features and the corresponding images into a constructed dense bias network for segmentation training to obtain a segmentation model; inputting the new abdominal CT angiography image to be segmented into the denoising auto-encoder model to obtain prior anatomical features of the image, and inputting the prior anatomical features into the segmentation model to obtain a segmentation result.According to the invention, a high-accuracy output result can be obtained, and renal artery segmentation can be quickly realized.

Description

technical field [0001] The invention relates to a semi-supervised renal artery segmentation method based on a dense bias network and an autoencoder, and belongs to the technical field of image processing. Background technique [0002] Renal artery segmentation on abdominal CT angiography images with the aim of obtaining a 3D renal artery tree mask up to the end of the interlobar artery. Clinicians can easily find the blood supply area corresponding to each interlobar artery by using the mask obtained by segmentation, which is very important for the diagnosis and preoperative planning of kidney diseases. As the probability of renal disease increases, 3D renal artery segmentation will play an important role in diagnosis and treatment. However, this is a very challenging task, and no one has succeeded in fine 3D renal artery segmentation so far for the following reasons: 1) The vessel scales in the kidney vary greatly. The patient's thickest renal artery can reach 7.4mm, whic...

AI Technical Summary

Problems solved by technology

This makes the network must be sensitive to features of different scales, increasing the difficulty of feature extraction

2) The anatomical shape of the renal artery varies greatly among different patients

This makes it difficult for small datasets to cover all anatomical variations and makes the network prone to overfitting

3) Small blood vessel structure

4) Small volume ratio

The renal artery only accounts for 0.27% of the region of interest in the kidney, which will cause a serious class imbalance problem and make the network difficult to train

5) Limitation on the number of labeled data

It is extremely difficult to learn feature representations of different renal artery anatomy on a small dataset, which will limit the generalization ability of the network

Images