Image segmentation method and model training method, device, medium, and electronic equipment

Abstract

The invention discloses an image segmentation method, a training method for an image segmentation model, a training device, a computer-readable storage medium, and an electronic device. The first model is trained by training a sample image and an image of a region of interest corresponding to the training sample image, and then the calculation The region of interest boundary probability map of the training sample image, that is, the probability that each pixel in the training sample image is the region of interest boundary, and the second model is trained with the training sample and the corresponding region of interest probability map, and finally the first model is Weighted fusion with the second model to obtain the final image segmentation model, and train the first model again with the training sample image and the ROI image corresponding to the training sample image; train the second model with the ROI boundary probability map, and use the second The model guides the retraining of the first model, which can reduce the influence of blurred boundaries on model training, and improve the accuracy and robustness of the image segmentation model.

Description

technical field [0001] The present application relates to the field of image processing, and specifically relates to an image segmentation method, a training method for an image segmentation model, a training device, a computer-readable storage medium, and electronic equipment. Background technique [0002] Deep learning is one of the most popular research directions at present, and it is widely used in many fields, such as advertising recommendation, automatic driving, and healthcare. One of the main reasons why deep learning, and computer vision in particular, has seen a resurgence in recent years is that data acquisition and preservation have become easier. In the medical field, the number of medical images is increasing at an alarming rate every year, but the training of radiologists requires time for learning and accumulation of experience. Therefore, the analysis of medical images combined with deep learning is a very meaningful research direction . [0003] In medic...

AI Technical Summary

Problems solved by technology

However, some unique properties of medical images affect the performance of supervised learning to a certain extent, mainly because: the internal structure of the human body is very complex, and many tissues are in close contact with each other, which will lead to blurred structural boundaries; many medical images are It is obtained by reconstructing the radiation signal, so the quality of the reconstruction algorithm will also affect the clarity of the tissue structure; moreover, the performance of supervised learning depends on the quality of the annotation, while the annotation of medical images can only be done by related fields. of experts completed

The technical level and subjectivity of experts may affect the labeling results; at the same time, this labeling method with a threshold of professional knowledge also means that it is impossible to improve the quality of labeling by labeling the same data multiple times

Images