Convolution neural network training method, ultrasonic image recognition and location method and system
A convolutional neural network and ultrasound image technology, applied in the field of image analysis, can solve problems such as differences in inspection results, unbalanced distribution of medical resources, and long cycles, and achieve the effect of improving accuracy
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Publication Date
- 2019-03-08
Smart Images

Figure 1 
Figure 2 
Figure 3
Abstract
Description
technical field
[0001] The invention relates to the technical field of image analysis, in particular to a convolutional neural network training method, an ultrasonic image recognition and positioning method and system. Background technique
[0002] Statistics from 2003 to 2011 in my country show that the incidence and death of thyroid cancer increased at an annual rate of 20.1% and 1.6%, respectively, and the incidence of thyroid cancer in all tumor registration points in the country showed an upward trend. Worldwide, the incidence of thyroid cancer has continued to rise over the past few decades, especially among women, showing explosive growth. Thyroid nodules have become a common clinical problem. Ultrasound is the most commonly used method for differential diagnosis of thyroid cancer. It is non-invasive, simple, and economical, and has been widely used. The only disadvantage is that it requires relatively high requirements for examiners and requires experienced Only sen...
Examples
Embodiment Construction
[0027] In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.
[0028] The convolutional neural network used in the present invention is mainly composed of a convolutional layer and a pooling layer, and its architecture diagram is, for example, figure 1 As shown, it is possible to have 2 fully connected layers in the last two layers. Among them, the present invention alternately uses 1×1 convolutional layers, which can reduce the complexity of the feature space. The convolutional neural network of the present invention does not need to search for the target through the region candidate frame, but directly completes the object position and category judgment through regression analysis, and converts the object detection problem into a regression problem to solve. Specifically, it divid...