Multi-scale feature fusion ultrasonic image semantic segmentation method based on adversarial learning

Abstract

The invention provides a multi-scale feature fusion ultrasonic image semantic segmentation method based on adversarial learning, and the method comprises the following steps: building a multi-scale feature fusion semantic segmentation network model, building an adversarial discrimination network model, carrying out the adversarial training and model parameter learning, and carrying out the automatic segmentation of a breast lesion. The method provided by the invention achieves the prediction of a pixel class through the multi-scale features of input images with different resolutions, improvesthe pixel class label prediction accuracy, employs expanding convolution for replacing partial pooling so as to improve the resolution of a segmented image, enables the segmented image generated by asegmentation network guided by an adversarial discrimination network not to be distinguished from a segmentation label, guarantees the good appearance and spatial continuity of the segmented image, and obtains a more precise high-resolution ultrasonic breast lesion segmented image.

Description

Technical field [0001] The invention relates to the technical field of medical image understanding, in particular to a method for semantic segmentation of multi-scale feature fusion ultrasound images based on adversarial learning. Background technique [0002] Breast cancer is a malignant tumor that occurs in breast epithelial tissue, which seriously threatens women's health and quality of life. Ultrasound examination is simple, convenient, economical and radiation-free, and has become an important tool for clinical breast cancer diagnosis. Ultrasound B-mode image combined with ultrasound elastography is an important method for clinical diagnosis of breast diseases. The B mode images the breast tissue structure, and the elastography detects the elasticity of the breast tissue. The two modal images can be compared with each other to more accurately detect and locate breast lesions. [0003] Accurately identifying and segmenting breast lesions from ultrasound images can provide imp...

AI Technical Summary

Problems solved by technology

However, the current semantic segmentation technology of breast ultrasound images based on deep learning has the following shortcomings: (1) predicting the category of pixels through single-scale features does not make full use of local features and global context features, which is prone to misclassification points; (2) The network passes through multi-level pooling (a typical VGG network has five pooling, and the feature map is reduced to 1 / 32 of the original image). The resolution of the feature map is significantly reduced, making the final segmentation map smaller; (3) Each pixel label The prediction does not consider the similarity between the pixel features, resulting in the lack of high-order spatial continuity in the segmentation map output by the network

Images