A method for determining an image to be labeled, a method and device for model training

Abstract

The present application discloses a method for determining an image to be labeled. The method is applied in the field of artificial intelligence, including: obtaining an original image and an autoencoder set, wherein the original image is an unlabeled image, and the autoencoder set includes N Self-encoder; obtain the coded image set corresponding to the original image through the self-encoder set, the coded image set includes N coded images, and there is a correspondence between the coded image and the self-encoder; obtain the coded image set and the original image through the image segmentation network The segmentation result set corresponding to the image, the image segmentation network includes M sub-image segmentation networks, and the segmentation result set includes [(N+1)*M] segmentation results; the uncertainty corresponding to the original image is determined according to the segmentation result set. This application discloses a method for model training. The present application can make the output result change more, thereby improving the calculation rationality of the uncertainty, and then helping to select more accurate images to be labeled.

Description

technical field [0001] The present application relates to the field of artificial intelligence, in particular to a method for determining an image to be marked, a method and a device for training a model. Background technique [0002] Medical image refers to the internal tissue image of the human body or a certain part of the human body obtained in a non-invasive way for medical treatment or medical research. It is an important means and reference factor for assisting clinical diagnosis. The inherent heterogeneity of different diseases is also reflected in its imaging phenotype. Therefore, etiological diagnosis or lesion segmentation through medical images is the most challenging task in medical image analysis. [0003] The deep convolutional neural network algorithm has been widely used in the classification and segmentation of lesions. Most of the classic segmentation and classification methods require high-precision labeling, so a lot of time is required for labeling. A...

AI Technical Summary

Problems solved by technology

[0004] However, there is such a problem in extracting uncertainty based on the dropout mechanism. Since the dropout mechanism is obtained based on a network with a similar structure and the same input image, different nodes are used each time, but there are still a large number of nodes with the same parameters. Therefore, , the output results still have a high correlation, and it is difficult to process the image into results with huge differences, which leads to the lack of rationality in the calculation of uncertainty, which is not conducive to the selection of accurate images to be labeled

Images