Ultrasonic carotid artery distal recognition device and method based on convolutional neural network

Abstract

The invention relates to medical image processing equipment, and provides an ultrasonic carotid artery far end recognizing device and method based on a convolutional neural network. According to the technical scheme for full-automatically, quickly and accurately extracting ROI, the device comprises an ultrasonic scanner and a computer, the computer is provided with the following modules: an image clipping module (1), an image equal-dividing module (2), a CNN processing module (3), and an ROI extracting module, wherein the CNN processing module (3) is used for inputting image blocks as convolutional neural networks CNN which have been trained and carrying out prediction classification, the class number is 2, namely, the image blocks contain images with 'dark-bright-dark-bright' structures and images without the structures, and the maximum-prediction-value image blocks of which the attribution in the same sub-image contains 'dark-bright-dark-bright' structures. The ultrasonic carotid artery far end recognizing device and method based on the convolutional neural network are mainly used for medical image processing.

Description

technical field [0001] The present invention relates to medical image processing, in particular, to a device and method for identifying the distal end of an ultrasonic carotid artery based on a convolutional neural network. Background technique [0002] Cardiovascular and cerebrovascular diseases have become the primary diseases that endanger human health. According to a survey by the World Health Organization, 17.5 million people died of cardiovascular and cerebrovascular diseases in the world in 2012, accounting for 30% of the total global death population. Studies have shown that carotid artery intima-media thickness (IMT) can effectively reflect the degree of atherosclerosis, and is an important indicator and main standard for predicting cardiovascular and cerebrovascular diseases. 1986, Literature [1] The IMT segmentation algorithm was proposed first, and then a large number of algorithms and schemes were proposed in succession in the next 30 years. These measurement...

AI Technical Summary

Problems solved by technology

Semi-automatic segmentation requires manual selection of points of interest [2] or ROI [3] , it is impossible to realize IMT automatic and fast measurement, which brings great challenges to the timely processing of massive medical image data

The fully automatic algorithm does not require human intervention. Literature [4] uses the active contour model to obtain the initial contour line and ROI, but the algorithm is prone to boundary leakage, and the velocity field parameters are difficult to select. Moreover, the algorithm has a high time complexity and is difficult to obtain. Meet real-time requirements

Literature [5] uses the watershed algorithm to select ROI, but this method is sensitive to noise, and noise is common in ultrasound images

Literature [6] uses the template matching algorithm to obtain ROI, but the applicability of the template is low, so it is necessary to continuously change the template to meet the image processing requirements from different ultrasound imaging equipment

Images