Breast ultrasound image tumor segmentation method based on fully convolutional network

Abstract

The invention belongs to the technical field of image processing, in particular to a breast ultrasound image tumor segmentation method based on a fully convolutional neural network. The method of the present invention includes: constructing a fully convolutional neural network based on atrous convolution, which is used for roughly segmenting ultrasound images to obtain breast tumors; in the constructed DFCN network, using atrous convolution, so that the network maintains the resolution of deeper feature maps rate to ensure that the tumor can be well segmented even in the case of a large number of shaded areas; in addition, batch normalization technology is also used in the DFCN network, which makes the network have a higher learning rate and accelerates the training process; using phase-based The dynamic contour PBAC model of information optimizes the rough segmentation results to obtain the final fine segmentation results; the experimental results show that the present invention can accurately segment tumors, especially for ultrasound images with blurred boundaries and many shadows.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a breast ultrasound image tumor segmentation method based on a fully convolutional network. Background technique [0002] Ultrasound imaging technology has the advantages of non-invasive, non-radiation, good real-time performance and low cost, and is widely used in breast tumor screening and diagnosis. In clinical applications, tumor contours in breast ultrasound images are usually manually delineated by sonographers, which is very time-consuming. In addition, the results of manual segmentation are highly dependent on the experience of sonographers, and the segmentation results of different observers are not the same. However, due to the characteristics of ultrasound imaging, the automatic segmentation of breast ultrasound images has the following problems: 1) severe speckle noise leads to low contrast and blurred boundaries of ultrasound images; 2) there ar...

AI Technical Summary

Problems solved by technology

However, due to the characteristics of ultrasound imaging, the automatic segmentation of breast ultrasound images has the following problems: 1) severe speckle noise leads to low contrast and blurred boundaries of ultrasound images; 2) there are a large number of shadow areas in the image, which are in gray 3) The shape, size and location of breast tumors are quite different, which puts forward higher requirements for the accuracy and robustness of the segmentation algorithm[1]

However, there are a lot of shadows and speckle noise in breast ultrasound images, making it difficult to obtain satisfactory segmentation results by block-based CNN or U-net

FCN is more suitable for breast ultrasound image segmentation, but the original FCN-8s network has too many parameters, the training process takes a long time, and the segmentation accuracy needs to be improved

Images