An image segmentation method, device and computer-readable storage medium based on pyramid fusion learning

Abstract

The invention discloses an image segmentation method based on feature pyramid fusion, which is used for analyzing magnetic resonance images and segmenting edema, necrotic tissue and normal tissue. This method preprocesses the data to remove the magnetic resonance layer that does not contain edema and necrotic tissue information, and then performs data enhancement in order to prevent overfitting, and then sends the data into a deep learning model with feature pyramid fusion, first using downsampling The process extracts the features of the data, and then gradually restores the resolution of the input data during the upsampling process, and finally uses the pyramid fusion method to fuse the information of various scales extracted during the upsampling process to obtain the segmentation result. The experiment was completed using two data sets of BraTS2015 and BraTS2017, and 5 times of cross-validation were done. Compared with the deep learning model that does not use pyramid fusion, the segmentation accuracy of the method proposed in the present invention is significantly improved, indicating that the method proposed in the present invention is effective.

Description

technical field [0001] The invention relates to an image segmentation method, device and computer-readable storage medium based on pyramid fusion learning, belonging to the field of image processing. Background technique [0002] The semantic segmentation of images is one of the important basic problems in computer vision. Its goal is to classify each pixel of the image and divide the image into several visually meaningful or interesting regions for the benefit of subsequent images. Analysis and visual understanding. It can be used for automatic driving, beautifying pictures, face modeling, 3D map reconstruction, etc. [0003] Traditional image automatic segmentation algorithms can be generally divided into threshold method, edge detection method, region growth method, watershed algorithm, model-based method (level set) or a combination of multiple methods. The threshold method needs to manually select the threshold, while ignoring the spatial information of the image. Th...

AI Technical Summary

Problems solved by technology

However, if the deep learning method is used for segmentation, the input data undergoes multiple convolutions and downsampling, and the lesion tissue information loses the spatial position information after multiple downsampling, so that in the subsequent process of upsampling to restore the image resolution, It is difficult to accurately segment the complete diseased tissue, and in medical imaging, there are multiple scales of diseased tissue information. Therefore, how to effectively use the location information and multi-scale information of the diseased tissue to complete the precise segmentation of the diseased tissue is a important questions to address now

Images