Image segmentation method based on MA-Unet

Abstract

The invention belongs to the technical field of image segmentation, and particularly relates to an MA-Unet-based image segmentation method, which comprises the following steps of data collection, data preprocessing, model construction and model training, wherein the data collection is used for collecting a lung data set from a LUNA match and the lung data set comprises 534 two-dimensional samples and respective label images. The data preprocessing comprises the steps of denoising, normalization, data division, image zooming and the like; according to the model construction, features generated by a plurality of intermediate layers are aggregated together for prediction by constructing an MA-Unet lung image segmentation model and utilizing global information of different scales, association between the features and an attention mechanism is established, global context information is mined, a noise region is removed at the same time to help the network to emphasize areas more related to semantic classes; when the loss function of the model is not reduced any more, the model is stored in the model training.

Description

technical field [0001] The invention belongs to the technical field of image segmentation, and in particular relates to an image segmentation method based on MA-Unet. Background technique [0002] Semantic segmentation of medical images is a key step in the diagnosis, treatment and follow-up of many diseases. In clinical practice, manual or semi-manual segmentation techniques are usually used for medical image segmentation. The disadvantage of these methods is that they use handcrafted features to obtain segmentation results. On the one hand, it is difficult to design representative features for different applications. On the other hand, functions designed for one type of image often fail in another. Therefore, traditional manual or semi-manual segmentation techniques lack general feature extraction methods. Since manual intensive labeling of a large number of medical images is a tedious and error-prone task, people put forward higher requirements for accurate and reliable...

AI Technical Summary

Problems solved by technology

[0003] Reasons for problems or deficiencies: Although convolutional neural networks are driving the development of semantic segmentation of medical images, there are still some deficiencies in the standard model. In the skip connection operation, the feature maps of the encoder and decoder sub-networks have large semantic Differences, long-range feature dependencies are not effectively modeled, ignoring global contextual information at different scales

Images