A bone age assessment method based on image enhancement

Abstract

A bone age assessment method based on image enhancement, comprising: X-ray image enhancement, firstly enhancing the bone edge information of the X-ray by combining the Sobel operator and mean filtering, and then enhancing the contrast of the X-ray using the power-law transformation method. Multi-region of interest extraction, extracting the metacarpal region, carpal region and thumb region in the wrist bone X-ray. Bone age assessment by combining heterogeneous information, using Inception-ResNet-V2 as the backbone, only using the feature extraction part of the Inception-ResNet-V2 network and removing the final fully connected layer of the network. The X-ray image and the region of interest image are respectively passed through the Inception-ResNet-V2 network to extract image features, and the gender information is passed through a fully connected layer with 32 neurons to calculate the corresponding feature vector. Then, the image feature vector and the gender feature vector are feature concatenated to form a new fused feature. Finally, through the calculation of two fully connected layers, the final bone age assessment result is obtained.

Description

Technical Field

[0001] The present invention relates to a bone age assessment method based on image enhancement. Background Art

[0002] Traditional bone age assessment methods all rely to a certain extent on the clinical experience of experts and require a relatively long assessment time to accurately assess bone age, which is inconvenient for clinical use. With the continuous development of deep learning, it has also become possible to use a deep learning method - convolutional neural network - to solve the problem of bone age assessment. The convolutional neural network is designed by imitating the way the human brain processes visual information and can reduce the dimensionality of a large amount of image information and effectively retain the image features. Therefore, it is applicable to various image processing tasks and has achieved satisfactory results in the fields of image classification, object detection, image segmentation, face recognition, etc. Therefore, if a model based on a convolutional neural network can be designed for automatic bone age assessment, it will provide great convenience for doctors' bone age assessment work. For this topic selection, based on the convolutional neural network model, the image information of wrist bone X-ray films is used to extract three important regions of interest in the wrist bone - the phalanx region, the metacarpal region, and the thumb region, and combined with gender information for automatic bone age assessment. Summary of the Invention

[0003] The present invention aims to overcome the above-mentioned drawbacks of the prior art and provides a bone age assessment method based on image enhancement.

[0004] The present invention adopts the following technical solutions to solve the technical problems:

[0005] A bone age assessment method based on image enhancement is designed. This method is divided into three steps: X-ray image enhancement, extraction of multiple regions of interest, and bone age assessment method combining heterogeneous information. During image enhancement, first, the Sobel operator and mean filtering are combined to enhance the bone edge information of the X-ray, and then the power-law transformation method is used to enhance the contrast of the X-ray. The extraction of regions of interest mainly extracts the metacarpal region, carpal region, and thumb region in the wrist bone X-ray, and is extracted based on 10 marked key points. When extracting, 4 key parameters of each region need to be obtained respectively, including the region center, reference direction, region width, and region length edge. Finally, a bone age assessment network combining heterogeneous information is designed for bone age assessment. The bone age assessment network uses Inception-ResNet-V2 as the backbone, only uses the feature extraction part of the Inception-ResNet-V2 network, and removes the last fully connected layer of the network. The X-ray image and the region of interest image are respectively passed through the Inception-ResNet-V2 network to extract image features, and the gender information is calculated through a fully connected layer with 32 neurons to obtain the corresponding feature vector. Then, the image feature vector and the gender feature vector are feature concatenated to form a new fused feature. Finally, through the calculation of two fully connected layers, the final bone age assessment result is obtained.

[0006] The present invention first performs image enhancement on the wrist bone X-ray to enhance the imaging quality of the X-ray, facilitating the subsequent training of the neural network. Then, the key regions of interest in the X-ray are accurately extracted, and these regions of interest are used to assist bone age assessment. Finally, a bone age assessment network combining heterogeneous information is designed, which can comprehensively utilize two different types of feature information, namely images and text, for bone age assessment together. The invention can accurately extract regions of interest and perform bone age assessment on the left hand wrist bone X-ray of teenagers, and has great application value.

[0007] The present invention has the following beneficial effects:

[0008] (1) Accurately assess the bone age of the left hand wrist bone X-ray of teenagers.

[0009] (2) Perform image enhancement on X-rays with poor image quality, enhancing the edge information and image contrast of the X-ray.

[0010] (3) Fuse the text feature information of gender and the image features, and use them together for bone age assessment, enhancing the accuracy of bone age assessment by combining heterogeneous information. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is the overall flowchart of the present invention.

[0012] Figure 2 It is the position annotation of 10 key points in the X-ray film of the present invention.

[0013] Figure 3 It is the structural diagram of the bone age assessment network for combining heterogeneous information in the present invention. Detailed implementation manners

[0014] The technical solution of the present invention will be further described below in conjunction with the accompanying drawings.

[0015] A bone age assessment method based on image enhancement according to the present invention includes the following steps:

[0016] Step 1 specifically includes:

[0017] 11) Obtain the X-ray film image of the left hand wrist bone.

[0018] 12) Use the Sobel operator to obtain the bone edge of the wrist bone.

[0019] 13) Use mean filtering to smooth the edge image.

[0020] 14) Add the smoothed edge image to the original image.

[0021] 15) Use power-law transformation to enhance the image contrast of the X-ray film.

[0022] In step 2, the metacarpal region, carpal region, and thumb region in the X-ray film of the wrist bone are extracted. When extracting, 4 key parameters of each region need to be obtained respectively, including the region center, reference direction, region width, and region length:

[0023] 21) Region center:

[0024] First, determine 10 key points in the X-ray film of the wrist bone, which are: the radial side of the distal end of the ulna (close to the radius side), the center of the capitate bone, the center of the first proximal phalanx, the center of the first distal phalanx, the second middle phalanx, the center of the third proximal phalanx, the center of the third distal phalanx, the fourth middle phalanx, the center of the fifth proximal phalanx, the center of the fifth middle phalanx, and the center of the fifth distal phalanx. Respectively use p 1 、p 2 、p 3 …p 10 to represent these 10 feature points.

[0025] Represent the center points of the three regions with the characters C Carpal 、C Thumb 、C Phanlange respectively. The coordinates of the center points are as follows:

[0026]

[0027] 22) Reference direction:

[0028] The reference directions of the three regions of interest are respectively They are respectively:

[0029]

[0030] 23) Region width:

[0031] The region widths of the three regions of interest are respectively denoted by W Carpal , W Thumb , W Phanlange They are respectively:

[0032]

[0033] 24) Region length:

[0034] The region lengths of the three regions of interest are respectively denoted by H Carpal , H Thumb , H Phanlange They are respectively:

[0035]

[0036]

[0037] Step 3: Bone age assessment method based on heterogeneous information combination

[0038] 31) Obtain the image of the wrist bone X-ray film, the images of the three extracted regions of interest, and the gender information corresponding to the X-ray film.

[0039] 32) Use the bone age assessment network for bone age assessment. The bone age assessment network designed in this patent uses Inception-ResNet-V2 as the backbone, only uses the feature extraction part of the Inception-ResNet-V2 network, and removes the last fully connected layer of the network. The X-ray film image and the images of the regions of interest are respectively passed through the Inception-ResNet-V2 network to extract image features, and the gender information is passed through a fully connected layer with 32 neurons to calculate the corresponding feature vector. Then, the image feature vector and the gender feature vector are concatenated to form a new fused feature. Finally, through the calculation of two fully connected layers, the final bone age assessment result is obtained.

[0040] The content described in the embodiments of this specification is only an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments, and the protection scope of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A bone age assessment method based on image enhancement, comprising the following steps: Step 1: X-ray image enhancement; specifically including: 11) Obtain the X-ray image of the left hand wrist bone; 12) Use the Sobel operator to obtain the bone edge of the wrist; 13) Use mean filtering to smooth the edge image; 14) Add the smoothed edge image to the original image; 15) Use power-law transformation to enhance the image contrast of the X-ray; Step 2: Extract multiple regions of interest; extract the metacarpal region, carpal region, and thumb region in the X-ray of the wrist bone. When extracting, 4 key parameters of each region need to be obtained respectively, including the region center, reference direction, region width, and region length: 21) Region center: First, determine 10 key points in the wrist bone X-ray film, namely: the radial side of the distal ulna (the side close to the radius), the center of the capitate bone, the center of the first proximal phalanx, the center of the first distal phalanx, the second middle phalanx, the center of the third proximal phalanx, the center of the third distal phalanx, the fourth middle phalanx, the center of the fifth proximal phalanx, the center of the fifth middle phalanx, and the center of the fifth distal phalanx; respectively represented by p 1 , p 2 , p 3 … p 10 represent these 10 feature points; Use the character C to represent the center points of the three regions Carpal , C Thumb , C Phanlange respectively, and the coordinates of the center points are as follows: 22) Reference direction: The reference directions of the three regions of interest are respectively They are respectively: 23) Region width: The regional widths of the three regions of interest are respectively denoted as W Carpal , W Thumb , W Phanlange , and they are respectively: 24) Region length: The regional lengths of the three regions of interest are H respectively Carpal , H Thumb , H Phanlange , and they are respectively: Step 3: Bone age assessment by combining heterogeneous information; 31) Obtain the X-ray image of the wrist bone, the three extracted regions of interest images, and the gender information corresponding to the X-ray; 32) Use the bone age assessment network to perform bone age assessment; the designed bone age assessment network takes Inception-ResNet-V2 as the backbone, only uses the feature extraction part of the Inception-ResNet-V2 network, and removes the last fully connected layer of the network; the X-ray image and the regions of interest images are respectively passed through the Inception-ResNet-V2 network to extract image features, and the gender information passes through a fully connected layer with 32 neurons to calculate the corresponding feature vector; then, the image feature vector and the gender feature vector are feature concatenated to form a new fused feature; finally, through the calculation of two fully connected layers, the final bone age assessment result is obtained.

Images