Identification method and device for training thyroid tumor ultrasonic image on line

A technology for thyroid tumors and ultrasound images, which is applied in the field of image recognition and can solve problems such as low recognition accuracy.

Inactive Publication Date: 2018-08-03
FUNDAN UNIVERSITY SHANGHAI CANER CENTER +1
View PDF2 Cites 15 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] In view of the above-mentioned problems in the prior art, the purpose of the present invention is to provide a method and device for ultrasonic image recognition of thyroid tumors that can effectively select tumor

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Identification method and device for training thyroid tumor ultrasonic image on line
  • Identification method and device for training thyroid tumor ultrasonic image on line
  • Identification method and device for training thyroid tumor ultrasonic image on line

Examples

Experimental program
Comparison scheme
Effect test

Example Embodiment

[0073] Example 1 Obtaining ultrasound images of thyroid tumor

[0074] The original data set was collected from patients with thyroid nodules from January 2014 to December 2016 by the Cancer Hospital of Fudan University. All papillary thyroid carcinoma (PTC) injuries were confirmed pathologically by surgically resected specimens. Each image in the original data set was ranked by three experienced radiologists using the Thyroid Imaging Report and Data System (TI-RADS). TI-RADS scores are divided into 2, 3, 4a, 4b, and 4c, which respectively indicate: no suspicion, possibly benign nodules, single suspicious feature, two suspicious features, and three or more suspicious features. The three radiologists also collected the imaging characteristics, including its composition, echo, calcification, edge and shape.

[0075] The entire data set has a total of 2836 original images, of which 1484 are PTC-confirmed ultrasound images, and 1352 are collected from benign tumors, including nodular...

Example Embodiment

[0076] Example 2 Tumor region selection and edge range amplification

[0077] All images in the original data set are taken from the phenomenon reporting system. If the entire image is used as a training sample, its inherent background and text will greatly affect the extraction of tumor characteristics. Therefore, it is necessary to select the tumor area from the original image, that is, use the rectangular frame to select the tumor mass completely, such as figure 2 As shown in the inner rectangular box. The selection process is completed one by one by the doctor using a semi-automatic selection script.

[0078] figure 2 The area shown in the middle and inner rectangular frame only contains the nodule itself, and the information of the surrounding tissue of the nodule is very necessary for judging the benign and malignant tumors, so the concept of edge range is added when cutting the image. figure 2 The outer rectangular frame in the two images is the cutting area after adding ...

Example Embodiment

[0082] Example 3 Image cutting and annotation form a training set

[0083] Split the original data set at a ratio of 6:1. The original data set has a total of 2836 images. In the experiment, 1275 PTC images and 1162 benign images are randomly selected as the training set, and the remaining 209 PTC images and 190 benign images are used as the test set. In the process of data set segmentation, it should be noted that the images of the same patient can only exist in the training set or the test set at the same time, which prevents the interference of similar images.

[0084] In the experimental testing process, the large difference in the size of the tumor itself is also considered, so the data set is subdivided into less than 0.5cm, 0.5cm to 1cm, and greater than 1cm. The tumor images of each level are in the training set, test set and size The distribution on Figure 4 As shown, the same distribution of training and testing data is satisfied.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses an identification method and device for training a thyroid tumor ultrasonic image on line. The method comprises the steps that a set of ultrasonic images are acquired, a tumorarea is selected from the ultrasonic images, a certain edge range of the tumor area is amplified for cutting, benign and malignant labeling is carried out, and the cut image is stored in an image baseand forms a training set with the partial original images in the image base; the training set is used for training an initial thyroid tumor ultrasonic image identification model to form an advanced thyroid tumor ultrasonic image identification model; a to-be-identified thyroid tumor ultrasonic image is acquired, a tumor area is selected, and a certain edge range of the image is amplified for cutting, and the advanced thyroid tumor ultrasonic image identification model is used for identifying the benign and malignant. The method can reuse a case image, and learning, memorizing and accumulatingcharacteristics of the thyroid tumor image is achieved; with the case increasing, the generalization capability and predicting accuracy of the model are gradually improved, which plays an important role in accumulating clinical diagnosis experience.

Description

technical field [0001] The present invention relates to the field of image recognition, in particular to a deep learning-based online training thyroid tumor ultrasound image recognition method and its device. Background technique [0002] Epidemiological studies have shown that thyroid cancer is one of the most common malignant tumors in women, and its incidence has risen rapidly in recent years. Papillary thyroid carcinoma (PTC) is the main pathological type of thyroid cancer, accounting for a large proportion of the incidence. Ultrasound images are widely regarded as the main diagnostic tool for thyroid nodule screening and the preoperative evaluation method for PTC. Ultrasonographic features such as microcalcifications, high solid content, and irregular edges and shapes are typical for PTC considerations. The biggest limitation in the actual operation process is the dependence of the operator, and the accuracy of judgment of radiologists with different experience levels...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06T7/00A61B8/08A61B8/00
CPCA61B8/085A61B8/5215A61B8/5223G06T7/0012G06T2207/10132G06T2207/30096
Inventor 向俊卢宏涛官青王芬王蕴珺李端树杜佳俊秦宇
Owner FUNDAN UNIVERSITY SHANGHAI CANER CENTER
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap