Three-dimensional image segmentation method based on convolutional neural network supervision

Abstract

The invention discloses a three-dimensional image segmentation method based on convolutional neural network supervision, and the method comprises the steps of combining a 3D UNet based on convolutional neural network supervision with a bidirectional convolutional recurrent neural network, and constructing a brand-new ABVUS three-dimensional image segmentation network model architecture; segmentinga three-dimensional tumor, a gland layer, a fat layer and a subcutaneous tissue layer from the three-dimensional mammary gland ultrasonic image; the invention aims at solving the problems that an encoder of a convolutional neural network may have gradient disappearance and performance degradation in the training process. Therefore, the deep boundary supervision is added to the encoder part of theconvolutional neural network to serve as the convolutional neural network, and the shallow network is trained more fully by means of the boundary prompt of the deep boundary supervision convolutionalneural network, so that the learning ability of the convolutional neural network can be improved while gradient disappearance is avoided, and therefore, the noise of the ultrasonic image can be better resisted.

Description

technical field [0001] The invention relates to the field of ultrasonic imaging methods, in particular to a three-dimensional image segmentation method based on convolutional neural network supervision. Background technique [0002] Automated Breast Volume Ultrasound System (ABVUS) is an emerging ultrasound technology. Currently, only three companies, Siemens in Germany, GM in the United States and Shantou Institute of Ultrasound Instruments Co., Ltd. (SIUI) in China, have launched it. Its own breast volume ultrasound imaging product, the present invention is an artificial intelligence segmentation scheme attached to the breast volume ultrasound imaging product of Shantou Ultrasound Instrument Research Institute Co., Ltd. (SIUI), China. [0003] Due to the high-quality segmentation of breast ultrasound medical images, it is of great significance in AI-assisted diagnosis. Segmenting three-dimensional breast ultrasound images into three-dimensional tumor shapes helps doctors ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to propose a three-dimensional image segmentation method based on convolutional neural network supervision, to solve one or more technical problems in the prior art, at least provide a beneficial choice or create conditions

Images