Lung lobe segmentation method and device based on few-sample learning

Abstract

The invention provides a lung lobe segmentation method and device based on few sample learning, and the method comprises the steps: extracting a support set left lung mask, a support set right lung mask and a support set background mask from a support set lung image, and determining feature prototypes of all lung lobes of a left lung, all lung lobes of a right lung and a background region; interference caused by similar left and right lung lobe structures is effectively avoided, and the feature prototype extraction precision is improved; extracting a to-be-segmented left lung mask and a to-be-segmented right lung mask in the to-be-segmented lung image, and performing feature extraction so as to obtain a to-be-segmented lung image based on image features of each pixel in the to-be-segmented left lung mask and the to-be-segmented right lung mask and feature prototypes of each lung lobe of the left lung, each lung lobe of the right lung and the background region; according to the method, each lung lobe in the to-be-segmented lung image is segmented, and a multi-type classification problem is reduced into a few-type classification problem, so that the pixel classification difficulty in a few-sample lung lobe segmentation scene can be reduced, and the segmentation precision in the scene is improved on the premise of greatly reducing the marking cost.

Description

technical field [0001] The present invention relates to the technical field of image segmentation, and in particular, to a method and device for lung lobe segmentation based on few-sample learning. Background technique [0002] Lung lobe segmentation is an important task for computer-aided diagnosis systems related to lung diseases. Accurate lung lobe segmentation can provide the specific location of lung diseases and help determine treatment plans. [0003] Traditional lung lobe segmentation methods mainly use supervised convolutional neural networks to complete the segmentation task. However, training a supervised convolutional neural network requires a large amount of labeled data. For the lung lobe segmentation task, the labeling is difficult and the labeling process is time-consuming. Therefore, there is a need for a lung lobe segmentation method that can accurately segment lung lobes based on a small number of labeled samples. When lung lobe segmentation is performed...

AI Technical Summary

Problems solved by technology

[0004] The present invention provides a lung lobe segmentation method and device based on few-sample learning, which is used to solve the defects in the prior art that a large amount of labeled data is required and the labeling cost is high, and the existing lung lobe segmentation method based on a small number of samples has insufficient accuracy in lung lobe edge segmentation. good defect

Images