Bone age evaluating method

A skeletal age and skeleton technology, applied in the field of skeletal age assessment, can solve the problems of limiting the generalization ability of the model, the model cannot be trained end-to-end, and affecting the accuracy of the evaluation results, and achieves good generalization, fast evaluation speed, and accuracy. high effect

Active Publication Date: 2019-08-13
UNIV OF SCI & TECH OF CHINA
4 Cites 2 Cited by

AI-Extracted Technical Summary

Problems solved by technology

However, most of these artificial intelligence technologies introduce specific standards, focus on specific skeletal parts (such as wrist bones, proximal phalanges, etc.) through ...
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

Method used

Compared with existing methods, this method does not need any detection, segmentation, preprocessing links, so the model can be trained end-to-end; the method can automatically mine the key bone positions of interest, without relying on specific human p...
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

Abstract

The invention discloses a bone age evaluating method which comprises the steps of based on an attention mechanism, mining a plurality of interested bone areas in a characteristic graph of an originalmetacarpal bone image X; based on an identification mechanism, identifying a plurality of interested bone areas which satisfy a requirement in the plurality of interested bone areas, and splicing theidentified characteristic graph of the plurality of interested bone areas with the characteristic graph of the original metacarpal bone image X for obtaining a characteristic vector C; and finally, after integrating the characteristic graph of each identified interested bone area, the characteristic graph of the original metacarpal bone image X and the characteristic vector C through a full connecting layer, performing bone age predicting, and integrating all predicting results for obtaining a bone age evaluating result. The bone age evaluating method realizes relatively high accuracy and relatively high evaluation speed.

Application Domain

Technology Topic

Bone ageDelayed bone maturation +3

Image

  • Bone age evaluating method
  • Bone age evaluating method
  • Bone age evaluating method

Examples

  • Experimental program(1)

Example Embodiment

[0012] The following describes the technical solutions in the embodiments of the present invention clearly and completely with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.
[0013] The embodiment of the present invention provides a bone age assessment method, which directly performs automatic detection and recognition on metacarpal images to obtain an assessment result. This method has fast processing speed, high efficiency, and high precision. It can be applied to the imaging department of hospitals, or scientific research institutions, or schools. It can be installed on work computers in the form of software to provide real-time detection. It can also be installed in hospitals or some The back-end server of scientific research institutions provides large-scale back-end testing.
[0014] Such as figure 1 As shown, the method is mainly divided into two parts: (1) attention mechanism for mining specific bone parts of interest; (2) recognition mechanism based on bone age assessment and integration of multiple classifiers; mainly as follows:
[0015] (1) Attention mechanism.
[0016] Based on the attention mechanism, mine multiple bone regions of interest from the feature map of the original metacarpal image X, specifically: extract the feature map of the original metacarpal image X through a feature extractor (which can be achieved through the ResNet50 network); The mechanism's region suggestion network mines the bone region of interest according to the feature map information, and selects N bone regions of interest after non-maximum value suppression. In addition, the region suggestion network for each bone region of interest R i Assign the corresponding suggested value Exemplarily, N=6 can be set. These bone regions of interest and the original metacarpal images X will be input into the recognition mechanism for recognition.
[0017] In order to allow the attention mechanism to better select distinguishable bone parts, an embodiment of the present invention proposes a new optimization algorithm.
[0018] If a bone region of interest has a higher recommended value Then the final evaluation result of the corresponding bone region of interest has a smaller error which is: among them Is the estimated value of bone age;
[0019] Then design the sort loss to optimize the attention mechanism:
[0020]
[0021] Where φ is the hinge loss function:
[0022] φ(x)=max{1-x,0}.
[0023] (2) Identification mechanism.
[0024] Based on the recognition mechanism, several skeletal regions of interest that meet the requirements are identified from multiple skeletal regions of interest, and the feature maps of the identified skeletal regions of interest are spliced ​​with the feature maps of the original metacarpal image X to obtain feature vector C; Finally, the identified feature map of each bone region of interest, the feature map of the original metacarpal bone image X, and the feature vector C are respectively passed through the fully connected layer, and then the bone age prediction is performed, and all the prediction results are integrated to obtain the bone age assessment result.
[0025] In the embodiment of the present invention, according to the suggested value of each bone region of interest given by the region suggestion network, sort from large to small, and then select the top K bone regions of interest for subsequent bone age prediction calculations.
[0026] Similarly, in the recognition mechanism, ResNet50 can also be used as a feature extractor to perform feature extraction operations on the bone region of interest and the original metacarpal image X.
[0027] The formula for integrating all prediction results is:
[0028]
[0029] In the above formula, A asb Is the bone age assessment result, A C , A X , Followed by the feature vector C, the feature map of the original metacarpal image X, and the identified bone region of interest R k The result of bone age prediction; K is the number of identified bone regions of interest.
[0030] The recognition mechanism is essentially a group of multiple regressions. The recognition mechanism is optimized by regression loss function. The N bone regions of interest selected before are used in the optimization stage, and the loss function is:
[0031]
[0032] Among them, R(·) is the loss function of the regression, the corresponding R(C) is the loss function corresponding to the splicing feature vector, R(X) is the loss function corresponding to the original metacarpal image X, R(R i ) Is the loss function corresponding to each bone region.
[0033] Such as figure 1 The entire scheme shown can be understood as a network model, in the training stage, the total loss function of the network model is the attention mechanism loss function L att And the recognition mechanism loss function L cls Sum:
[0034] L total =L cls +L att.
[0035] In the optimization process of the training phase, the attention mechanism continuously improves the accuracy of specific bone parts extraction, extracts more distinguishable bone parts, and sends them to the recognition mechanism. The recognition mechanism continuously improves recognition accuracy and reduces prediction errors. At the same time, the recognition result will also be fed back to the attention mechanism for optimization of the attention mechanism. Therefore, the two mechanisms can reinforce each other, and the network model can be trained end-to-end without relying on human priors.
[0036] Compared with existing methods, this method does not require any detection, segmentation, and preprocessing steps, so the model can be trained end-to-end; this method can automatically mine the key bone parts of interest and does not rely on specific human prior knowledge, so It has better generalization; this method is completely based on 2D convolutional neural network (ie, the feature extractor described in the article), with fast speed and high accuracy, and the average diagnosis error is within 4.4 months.
[0037] Through the description of the foregoing embodiments, those skilled in the art can clearly understand that the foregoing embodiments can be implemented by software, or can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solutions of the foregoing embodiments can be embodied in the form of software products, which can be stored in a non-volatile storage medium (which can be CD-ROM, U disk, mobile hard disk, etc.), including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute the methods described in the various embodiments of the present invention.
[0038] The above are only preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or changes within the technical scope disclosed in the present invention. All replacements shall be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the 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

PUM

no PUM

Description & Claims & Application Information

We can also present the details of the Description, Claims and Application information to help users get a comprehensive understanding of the technical details of the patent, such as background art, summary of invention, brief description of drawings, description of embodiments, and other original content. On the other hand, users can also determine the specific scope of protection of the technology through the list of claims; as well as understand the changes in the life cycle of the technology with the presentation of the patent timeline. Login to view more.
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

Similar technology patents

Water surface object recognition system and method and application based on 5G MEC

PendingCN113139497AGood generalizationThe best balance between detection accuracy and efficiencyCharacter and pattern recognitionNeural architecturesGarbage collectionObstacle avoidance
Owner:GUANGDONG UNIV OF TECH

Subject determination method and device of text specified information and computer storage medium

PendingCN110866396AReduce labor costsGood generalizationManual annotationSpeech recognition
Owner:BEIJING MININGLAMP SOFTWARE SYST CO LTD

Classification and recommendation of technical efficacy words

  • Good generalization
  • Fast evaluation

Clamp structure used for eccentricity molding machining of glass mold

InactiveCN103273353AGood generalizationGood clamping adaptabilityPositioning apparatusMetal-working holdersEngineeringMachining
Owner:ORI MOLD TECH SUZHOU

Surface roughness evaluation method based on Hoyer coefficient of processed surface image

ActiveCN114359112AFast evaluationReduce the impactImage enhancementImage analysisComputer visionScale space
Owner:LYNCWELL INNOVATION INTELLIGENT SYST ZHEJIANG CO LTD +1
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