MEDU-Net + network-based medical image segmentation method and system

Abstract

The invention discloses a medical image segmentation method based on an MEDU-Net + network, and the method comprises the steps: a 3 * 3 convolution layer used for extracting image feature information in an original U-Net network is replaced with an inception module in GoogLeNet, and enabling the inception module to comprise a plurality of branches to form a multi-scale encoder; a decoder of the U-Net network is correspondingly optimized, and a multi-scale decoding mode is adopted to recover the obtained semantic information of different scales; wherein each branch of the encoder and each branch of the decoder are in one-to-one correspondence, a layer-to-return jump connection is introduced to directly transmit information extracted by an encoding end to a decoding end, and each part of the intermediate connection is a transposition convolution of a next adjacent layer; the generalized Dice loss function and the Focal loss function are combined, and the weight is introduced according to the characteristics of the medical image so as to generate a loss function in a combination form. Image features can be learned as much as possible through a small amount of data, and a better segmentation result is obtained.

Description

technical field [0001] The invention relates to the technical field of image segmentation, in particular to a MEDU-Net+ network-based medical image segmentation method and system. Background technique [0002] With the development of artificial intelligence, deep learning methods have received extensive attention, and many efficient, convenient and simple image segmentation methods have been gradually proposed. Most deep learning-based image segmentation methods require enough images for training and testing processing, however, medical images for training and testing need to be annotated, and due to professional limitations, the number of images that can be utilized is always limited, so how to In the case of only a small amount of data, the network structure itself can collect as much information as possible, which has become a problem to be solved in the field of medical image segmentation. The emergence of the U-Net network provides a more efficient way to utilize the e...

AI Technical Summary

Problems solved by technology

Since the U-Net network has a structure such as skip connections and a unique U-shaped structure, more detailed image information can be aggregated according to the deep and shallow features of the image, so that the existing improved U-Net network can basically extract some relevant elements from the image. , to obtain more accurate image segmentation results, but these methods still only focus on the internal information of the image, ignoring the details and edge information of the target to be segmented

Images