WO2020062840A1 - 一种检测骨龄的方法及装置 - Google Patents

一种检测骨龄的方法及装置 Download PDF

Info

Publication number
WO2020062840A1
WO2020062840A1 PCT/CN2019/082682 CN2019082682W WO2020062840A1 WO 2020062840 A1 WO2020062840 A1 WO 2020062840A1 CN 2019082682 W CN2019082682 W CN 2019082682W WO 2020062840 A1 WO2020062840 A1 WO 2020062840A1
Authority
WO
WIPO (PCT)
Prior art keywords
bone
bone age
feature extraction
coordinates
region
Prior art date
Application number
PCT/CN2019/082682
Other languages
English (en)
French (fr)
Inventor
魏子昆
杨忠程
王�琦
Original Assignee
杭州依图医疗技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 杭州依图医疗技术有限公司 filed Critical 杭州依图医疗技术有限公司
Priority to SG11202002140VA priority Critical patent/SG11202002140VA/en
Publication of WO2020062840A1 publication Critical patent/WO2020062840A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/12—Edge-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/136—Segmentation; Edge detection involving thresholding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20024—Filtering details
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30008—Bone

Definitions

  • Embodiments of the present invention relate to the field of machine learning technologies, and in particular, to a method and a device for detecting bone age.
  • Bone age is the abbreviation of skeletal age. It is the age at which adolescents' bone development level is compared with bone development standards. It can more accurately reflect the maturity of the body than age, height, and weight, and more accurately reflect the individual. Growth level and maturity.
  • bone age films are mainly shot, and the bone age is estimated by manually viewing the bone age films.
  • This method relies on artificial estimation of the bone age based on the bone age films.
  • the accuracy is subject to human subjective factors and the image is large and the efficiency is low.
  • Embodiments of the present application provide a method and device for detecting bone age.
  • an embodiment of the present application provides a method for detecting bone age, including:
  • a classification model is used to determine the number of ossification centers in the carpal region, and the classification model is determined by training a deep residual network by using bone age slices of the carpal region labeled in advance with the number of ossification centers;
  • the bone age of the user to be detected is determined according to the number of ossification centers in the wrist bone region.
  • a classification model is used in the embodiment of the present application to automatically determine the number of ossification centers in the carpal region, and then according to the ossification centers in the carpal region
  • the number of bones determines the bone age of the user to be detected without artificially judging the bone age based on the bone age film, thereby improving the accuracy of detecting the bone age on the one hand and improving the efficiency of detecting the bone age on the other.
  • the method of determining the bone age of the user to be detected according to the number of ossification centers in the wrist region is more efficient.
  • the determining a wrist region from the bone age sheet includes:
  • a segmentation model is used to determine the coordinates of the key points corresponding to the carpal region in the bone age slice.
  • the segmentation model is determined by training multiple deep age residual networks with the coordinates of the key points marked in advance as training samples;
  • the carpal region is determined according to the coordinates of a key point corresponding to the carpal region.
  • the determining the number of ossification centers in the carpal region by using a classification model includes:
  • the bone age slice of the carpal region is processed by N consecutive convolution feature extraction blocks to obtain image features of the carpal region, where N is greater than 0, and the convolution feature extraction block includes L convolution modules, and L is greater than 0 , Any one of the convolution modules includes a convolution layer, a BN layer, and an excitation layer; for any two consecutive first convolution feature extraction blocks and second convolution feature extraction blocks of the N convolution feature extraction blocks, The second image feature output by the second convolution feature extraction block and the first image feature output by the first convolution feature extraction block are added as the input of a third convolution feature extraction block or N consecutive convolutions.
  • the third convolution feature extraction block is a convolution feature extraction block that is located after the second convolution feature extraction block and is continuous with the second convolution feature extraction block;
  • the number of ossification centers in the carpal region is determined according to the type of the carpal region.
  • the classification model is used to determine the number of ossification centers in the carpal bone area
  • the segmentation model is used to segment the carpal bone area from the bone age film, narrow the detection range, and then determine the number of ossification centers from the carpal bone area.
  • the number of ossification centers determines the bone age, thereby improving the accuracy of detecting the bone age.
  • the adjusting each bone in the bone age piece to a reference position includes:
  • An adjustment model is used to determine the coordinates of the key points in the bone age piece.
  • the adjustment model is based on multiple bone age pieces in which the coordinates of the key points are marked in advance as training samples.
  • the key points are determined after training the deep residual network.
  • the coordinates of the preset reference point are coordinates in the same coordinate system;
  • an embodiment of the present application provides a device for detecting bone age, including:
  • An acquisition module configured to acquire a bone age slice of a user to be detected, and adjust each bone in the bone age slice to a reference position
  • a segmentation module configured to determine a carpal bone region from the bone age slice
  • a classification module is used to determine the number of ossification centers in the carpal region using a classification model.
  • the classification model is based on bone age slices of the carpal region in which the number of ossification centers is marked in advance as training samples to train a deep residual network. Determined later
  • a detection module configured to determine the bone age of the user to be detected according to the number of ossification centers in the wrist region.
  • a segmentation model is used to determine the coordinates of the key points corresponding to the carpal region in the bone age slice.
  • the segmentation model is determined by training multiple deep age residual networks with the coordinates of the key points marked in advance as training samples;
  • the carpal region is determined according to the coordinates of a key point corresponding to the carpal region.
  • the classification module is specifically configured to:
  • the bone age slice of the carpal region is processed by N consecutive convolution feature extraction blocks to obtain image features of the carpal region, where N is greater than 0, and the convolution feature extraction block includes L convolution modules, and L is greater than 0 , Any one of the convolution modules includes a convolution layer, a BN layer, and an excitation layer; for any two consecutive first convolution feature extraction blocks and second convolution feature extraction blocks of the N convolution feature extraction blocks, The second image feature output by the second convolution feature extraction block and the first image feature output by the first convolution feature extraction block are added as the input of a third convolution feature extraction block or N consecutive convolutions.
  • the third convolution feature extraction block is a convolution feature extraction block that is located after the second convolution feature extraction block and is continuous with the second convolution feature extraction block;
  • the number of ossification centers in the carpal region is determined according to the type of the carpal region.
  • the obtaining module is specifically configured to:
  • An adjustment model is used to determine the coordinates of the key points in the bone age piece.
  • the adjustment model is based on multiple bone age pieces in which the coordinates of the key points are marked in advance as training samples.
  • the key points are determined after training the deep residual network.
  • the coordinates of the preset reference point are coordinates in the same coordinate system;
  • an embodiment of the present application provides a device for detecting bone age, including at least one processor and at least one memory, wherein the storage unit stores a computer program, and when the program is executed by the processor, So that the processor executes the steps of the method according to the first aspect.
  • an embodiment of the present application provides a computer-readable medium that stores a computer program executable by a device that detects bone age, and when the program is run on the device that detects bone age, the device that makes the device detect bone age Perform the steps of the method described in the first aspect.
  • FIG. 1 is a schematic flowchart of a method for detecting bone age according to an embodiment of the present application
  • FIG. 2 is a schematic diagram of a bone age slice provided by an embodiment of the present application.
  • FIG. 3 is a schematic diagram of a wrist region provided by an embodiment of the present application.
  • FIG. 4 is a schematic flowchart of a method for adjusting a bone age piece according to an embodiment of the present application
  • FIG. 5 is a schematic structural diagram of a deep residual network according to an embodiment of the present application.
  • FIG. 7 is a schematic flowchart of a method for detecting the number of ossification centers according to an embodiment of the present application
  • FIG. 8 is a schematic structural diagram of a device for detecting bone age according to an embodiment of the present application.
  • Bone age short for bone age, which needs to be determined with the help of specific images of bones in X-ray photography. Usually, an X-ray film of the wrist of the left hand of a person is taken. The doctor observes the development of the ossification center of the left palmar phalanx, wrist and radius ulna to determine the bone age.
  • the technical solution for detecting the bone age in the embodiments of the present application is suitable for detecting the bone age of adolescents in a hospital, providing a reference for judging adolescents' growth and development, and helping doctors diagnose diseases such as chromosomal abnormalities, hereditary diseases, and endocrine diseases.
  • FIG. 1 exemplarily illustrates a schematic flowchart of a method for detecting bone age provided by an embodiment of the present application.
  • the process may be performed by a device for detecting bone age, and specifically includes the following steps:
  • step S101 a bone age piece of the user to be detected is obtained, and each bone in the bone age piece is adjusted to a reference position.
  • the bone age film refers to a specific image taken by X-rays, and an X-ray film of a person's left hand is usually taken as a bone age film. For example, as shown in FIG. 2.
  • the pre-processing process mainly includes gland segmentation and image normalization.
  • Gland segmentation includes the following steps: First, Gaussian filtering is applied to the entire bone age slice, and the filtered result is binarized.
  • the binarization threshold is obtained by obtaining the maximum class spacing method of the image gray histogram.
  • the binarized result is inflated to obtain individual area blocks by the flood method, and the area of each area block is counted. The largest area block is retained and the segmented hand bone image. Paste the segmented hand bone image onto a pure black image that matches the length and width of the hand bone image.
  • the image normalization includes the following steps:
  • the bone age slice is an image in the dicom format. First, according to the dicom information, a window width and a window level are selected, and the bone age slice image is converted into a png format. By adding black borders on the top or both sides of the bone age film image, the bone age film image aspect ratio is adjusted to 1: 1, and finally the bone age film image is scaled to a size of 512 * 512.
  • the carpal area is located in the wrist and consists of 8 small bones arranged in two rows.
  • the proximal row is from the radial side to the ulnar side. It is the scaphoid bone, lunar bone, triangular bone, and pea bone. All but the pea bone participate in the radial wrist joint. Composition.
  • the distal row from the radial side to the ulnar side is a large horn bone, a small polygonal bone, a skull bone, and a hook bone, all of which are involved in the composition of the carpal and palm joints, as an example, as shown in FIG. 3.
  • step S103 the number of ossification centers in the carpal region is determined using a classification model.
  • the first ossification site is called the ossification center.
  • the small bones in the carpal area do not grow at the same time, but grow sequentially. Therefore, the ossification center of each small bone in the carpal area is detected, and then the bone age can be determined according to the number of ossification centers.
  • the classification model is determined by training the deep residual network based on the bone age slices of the carpal bone area labeled with the number of ossification centers in advance.
  • Step S104 Determine the bone age of the user to be detected according to the number of ossification centers in the wrist region.
  • a comparison reference table between the number of ossification centers and the bone age in the wrist region is set in advance. After the classification model outputs the number of ossification centers, the bone age of the user to be detected is determined by looking up the table.
  • adjusting each bone in the bone age slice to the reference position includes the following steps, as shown in FIG. 4:
  • Step S401 Acquire the coordinates of a preset reference point.
  • the coordinates of the key point and the coordinates of the preset reference point are coordinates in the same coordinate system.
  • multiple bone age slices are obtained, and then key points around the little finger in the bone age slices are manually labeled, and then the bone age slices marked with the key points are input to a deep residual network for training.
  • the objective function of the deep residual network meets a preset condition .
  • Determine the adjustment model When the bone age piece of the user to be detected is obtained, the bone age piece is input into the adjustment model to determine the key points around the little finger in the bone age piece.
  • Step S403 Determine the correspondence between the current position of each bone of the bone age slice and the reference position according to the coordinates of the preset reference point and the coordinates of the key points.
  • step S404 each bone in the bone age piece is adjusted to a reference position according to the corresponding relationship.
  • the current position of the little finger in the bone age film and the reference position of the little finger in the bone age film can be determined according to the coordinates of the preset reference point and the key point coordinates. Corresponding relationship can be further obtained between the current position and the reference position of other bones in the bone age film, wherein the corresponding relationship includes translation relationship and rotation relationship. Then, each bone in the bone age film is adjusted to the reference position according to the corresponding relationship. Before determining the wrist bone area, first adjust each bone in the bone age slice of the user to be tested to the reference position, thereby improving the accuracy of determining the wrist bone area and further improving the accuracy of detecting the bone age.
  • the segmentation model may be used to determine the wrist bone region from the unadjusted bone age sheet, or the segmentation model may be used to determine the wrist bone region from the adjusted bone age sheet, where adjusting the bone age sheet refers to adjusting the bone age sheet.
  • the segmentation model is determined by training the deep residual network by using multiple bone age slices in which coordinates of key points are marked in advance as training samples.
  • other neural network models can also be used for training to obtain a segmentation model for determining the carpal bone region from the adjusted bone age slice, which is not limited here.
  • multiple bone age slices are obtained as training samples. For each bone age piece, pretreat the bone age piece, and then adjust each bone in the bone age piece to the reference position.
  • the process of adjusting the position of the bone age piece and pretreating the bone age piece has been described in the foregoing. I will not repeat them here.
  • the key points are marked by the markers in each bone age film, and the key points are points near the carpal bone region in the bone age film. Then perform data augmentation on the training samples to increase the data amount to 10 times the original data amount.
  • the data augmentation methods include, but are not limited to:
  • the segmentation model is used to determine the coordinates of the key points corresponding to the carpal region in the bone age film, and the carpal region is determined according to the coordinates of the key points corresponding to the carpal region.
  • the segmentation model is used to segment the carpal bone area from the bone age film to reduce the detection range.
  • the number of ossification centers determines the bone age, thereby improving the accuracy of detecting the bone age.
  • the training process of the classification model is as follows:
  • a bone age piece of the carpal bone region segmented from multiple bone age pieces is obtained as a training sample. Bone age piece for each carpal area. The number of ossification centers was marked by the labeling staff on the bone age films of each wrist region. Data augmentation is then performed on the training samples. Methods for data augmentation include, but are not limited to:
  • the training samples are then input to the deep residual network for training.
  • the loss function is calculated based on the number of labeled ossification centers and the number of labeled ossification centers predicted by the network. It is trained by the back propagation method.
  • the training optimization algorithm uses the sgd algorithm with momentum and step attenuation.
  • the structure of the above-mentioned deep residual network is shown in FIG. 5 and includes N consecutive convolutional feature extraction blocks and a fully connected layer.
  • N convolutional feature extraction blocks For any two consecutive first volumes of the N convolutional feature extraction blocks, Convolution feature extraction block and second convolution feature extraction block, the second image feature output by the second convolution feature extraction block and the first image feature output by the first convolution feature extraction block are added as a third convolution feature extraction block Input or the output of N consecutive convolutional feature extraction blocks.
  • the third convolutional feature extraction block is a convolutional feature extraction block that is located after the second convolutional feature extraction block and is continuous with the second convolutional feature extraction block.
  • the convolution feature extraction block includes L convolution modules, where L is greater than 0. Any convolution module includes a convolution layer, a BN layer, and an excitation layer, as shown in FIG. 6.
  • the classification model is used to determine the number of ossification centers in the carpal region, including the following steps, as shown in FIG. 7:
  • step S701 the bone age slice of the carpal bone area is processed by N consecutive convolution feature extraction blocks to obtain image features of the carpal bone area.
  • step S702 the image features of the carpal bone region are input into the fully connected layer, and the type of the carpal bone region is output.
  • Step S703 Determine the number of ossification centers in the wrist region according to the type of the wrist region.
  • the type of the carpal bone region output from the fully connected layer is nine types of 0 to 8 types, where type 0 indicates that the number of ossification centers in the wrist region is 0, and 1 indicates that the number of ossification centers in the wrist region is 1, And so on. Then query the comparison table according to the number of ossification centers in the carpal bone area to determine the bone age of the user to be detected.
  • a classification model is used in the embodiments of the present application to automatically determine the number of ossification centers in the carpal region, and then determine the user to be detected based on the number of ossification centers in the carpal region. It is not necessary to judge the bone age based on the bone age film artificially and subjectively, thereby improving the accuracy of detecting the bone age on the one hand, and improving the efficiency of detecting the bone age on the other. Compared with the method of determining the bone age by identifying the morphology of the wrist bone, the method of determining the bone age of the user to be detected according to the number of ossification centers in the wrist region is more efficient.
  • a segmentation module 802 configured to determine a carpal bone region from the bone age slice
  • a classification module 803 is configured to determine the number of ossification centers in the carpal region by using a classification model.
  • the classification model is based on a bone age slice of the carpal region labeled with the number of ossification centers in advance as a training sample to perform a deep residual network. Determined after training;
  • a detection module 804 is configured to determine the bone age of the user to be detected according to the number of ossification centers in the wrist region.
  • the segmentation module 802 is specifically configured to:
  • a segmentation model is used to determine the coordinates of the key points corresponding to the carpal region in the bone age slice.
  • the segmentation model is determined by training multiple deep age residual networks with the coordinates of the key points marked in advance as training samples;
  • the carpal region is determined according to the coordinates of a key point corresponding to the carpal region.
  • the classification module 803 is specifically configured to:
  • the bone age slice of the carpal region is processed by N consecutive convolution feature extraction blocks to obtain image features of the carpal region, where N is greater than 0, and the convolution feature extraction block includes L convolution modules, and L is greater than 0 , Any one of the convolution modules includes a convolution layer, a BN layer, and an excitation layer; for any two consecutive first convolution feature extraction blocks and second convolution feature extraction blocks of the N convolution feature extraction blocks, The second image feature output by the second convolution feature extraction block and the first image feature output by the first convolution feature extraction block are added as the input of a third convolution feature extraction block or N consecutive convolutions.
  • the third convolution feature extraction block is a convolution feature extraction block that is located after the second convolution feature extraction block and is continuous with the second convolution feature extraction block;
  • the number of ossification centers in the carpal region is determined according to the type of the carpal region.
  • the obtaining module 801 is specifically configured to:
  • An adjustment model is used to determine the coordinates of the key points in the bone age piece.
  • the adjustment model is based on multiple bone age pieces in which the coordinates of the key points are marked in advance as training samples.
  • the key points are determined after training the deep residual network.
  • the coordinates of the preset reference point are coordinates in the same coordinate system;
  • the embodiment of the present application provides a device for detecting bone age.
  • the device includes at least one processor 901 and a memory 902 connected to the at least one processor.
  • the connection between the processor 901 and the memory 902 through a bus in FIG. 9 is taken as an example.
  • the bus can be divided into an address bus, a data bus, a control bus, and the like.
  • the memory 902 stores instructions that can be executed by at least one processor 901.
  • the at least one processor 901 can execute the steps included in the foregoing method for detecting bone age by executing the instructions stored in the memory 902.
  • the processor 901 is a control center of a device for detecting bone age, and can use various interfaces and lines to connect various parts of the device for detecting bone age, by running or executing instructions stored in the memory 902 and calling data stored in the memory 902 To detect bone age.
  • the processor 901 may include one or more processing units, and the processor 901 may integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, and an application program, etc.
  • the tuning processor mainly handles wireless communication. It can be understood that the foregoing modem processor may not be integrated into the processor 901.
  • the processor 901 and the memory 902 may be implemented on the same chip, and in some embodiments, they may also be implemented on separate chips.
  • the processor 901 may be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application specific integrated circuit (ASIC), a field programmable gate array or other programmable logic device, a discrete gate, or a transistor
  • the logic device and discrete hardware components can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application.
  • a general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in combination with the embodiments of the present application may be directly implemented by a hardware processor, or may be performed by a combination of hardware and software modules in the processor.
  • the memory 902 is a non-volatile computer-readable storage medium, and may be used to store non-volatile software programs, non-volatile computer executable programs, and modules.
  • the memory 902 may include at least one type of storage medium, for example, it may include a flash memory, a hard disk, a multimedia card, a card-type memory, a Random Access Memory (RAM), a Static Random Access Memory (SRAM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Magnetic Memory, Disk , CDs and more.
  • the memory 902 is any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.
  • the memory 902 in the embodiment of the present application may also be a circuit or any other device capable of implementing a storage function, and is configured to store program instructions and / or data.
  • an embodiment of the present application further provides a computer-readable medium that stores a computer program executable by a device that detects bone age, and when the program is run on the device that detects bone age, the detected bone age is made The device performs the steps of the method of detecting bone age.
  • the embodiments of the present invention may be provided as a method or a computer program product. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to work in a particular manner such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, the instructions
  • the device implements the functions specified in one or more flowcharts and / or one or more blocks of the block diagram.
  • These computer program instructions can also be loaded on a computer or other programmable data processing device, so that a series of steps can be performed on the computer or other programmable device to produce a computer-implemented process, which can be executed on the computer or other programmable device.
  • the instructions provide steps for implementing the functions specified in one or more flowcharts and / or one or more blocks of the block diagrams.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

一种检测骨龄的方法及装置，涉及机器学习技术领域，该方法包括：获取待检测用户的骨龄片，并调整骨龄片中的各骨头至基准位置（101），从骨龄片中确定腕骨区域（102），然后采用分类模型确定腕骨区域中骨化中心的数量（103），最后根据腕骨区域中骨化中心的数量确定待检测用户的骨龄（104）。由于腕骨区域中骨化中心的数量代表着不同的骨龄阶段，因此该方法采用分类模型自动确定腕骨区域中骨化中心的数量，然后根据腕骨区域中骨化中心的数量确定待检测用户的骨龄，而不需要人工主观根据骨龄片判断骨龄，从而一方面提高了检测骨龄的精度，另一方面提高了检测骨龄的效率。

Description

一种检测骨龄的方法及装置
本申请要求在2018年09月30日提交中国专利局、申请号为201811163210.7、申请名称为“一种检测骨龄的方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明实施例涉及机器学习技术领域，尤其涉及一种检测骨龄的方法及装置。
背景技术
“骨龄”是骨骼年龄的简称，是青少年儿童骨骼发育水平同骨发育标准比较而得出的发育年龄，它比年龄、身高、体重更能精确的反映出身体的成熟程度，更加准确地反映个体的生长发育水平和成熟程度。
临床上通过检测骨龄来判读儿童的生物学年龄，通过生物年龄与日历年龄的差异来评估儿童发育状况，了解儿童性成熟的趋势，预测儿童的成年身高等，并广泛用于影响儿童生长发育疾病的治疗监测，对一些儿科内分泌疾病的诊断有很大帮助。
现有技术中主要通过拍摄骨龄片，由人工查看骨龄片估计骨龄。该方法依赖人工根据骨龄片估计骨龄，精度受人的主观因素影像大，效率较低。
发明内容
由于现有技术中依赖人工根据骨龄片估计骨龄，精度受人的主观因素影像大，效率较低的问题，本申请实施例提供了一种检测骨龄的方法及装置。
第一方面，本申请实施例提供了一种检测骨龄的方法，包括：
获取待检测用户的骨龄片，并调整所述骨龄片中的各骨头至基准位置；
从所述骨龄片中确定腕骨区域；
采用分类模型确定所述腕骨区域中骨化中心的数量，所述分类模型是以预先标记骨化中心的数量的腕骨区域的骨龄片为训练样本，对深度残差网络进行训练后确定的；
根据所述腕骨区域中骨化中心的数量确定所述待检测用户的骨龄。
本申请实施例中，由于腕骨区域中骨化中心的数量代表着不同的骨龄阶段，因此本申请实施例中采用分类模型自动确定腕骨区域中骨化中心的数量，然后根据腕骨区域中骨化中心的数量确定待检测用户的骨龄，而不需要人工主观根据骨龄片判断骨龄，从而一方面提高了检测骨龄的精度，另一方面提高了检测骨龄的效率。相较于通过识别腕骨形态来确定骨龄的方法来说，根据腕骨区域中骨化中心的数量确定待检测用户的骨龄的方法，其效率更高。
可选地，所述从所述骨龄片中确定腕骨区域，包括：
采用分割模型确定所述骨龄片中腕骨区域对应的关键点的坐标，所述分割模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的；
根据所述腕骨区域对应的关键点的坐标确定所述腕骨区域。
可选地，所述采用分类模型确定所述腕骨区域中骨化中心的数量，包括：
将所述腕骨区域的骨龄片经N个连续的卷积特征提取块处理，得到所述腕骨区域的图像特征，N大于0，所述卷积特征提取块包括L个卷积模块，L大于0，任意一个卷积模块中包括卷积层、BN层及激励层；针对所述N个卷积特征提取块中任意两个连续的第一卷积特征提取块和第二卷积特征提取块，所述第二卷积特征提取块输出的第二图像特征与所述第一卷积特征提取块输出的第一图像特征相加后作为第三积特征提取块的输入或者N个连续的卷积特征提取块的输出；所述第三卷积特征提取块为位于所述第二卷积特征提取块之后且与所述第二卷积特征提取块连续的卷积特征提取块；
将所述腕骨区域的图像特征输入全连接层，输出所述腕骨区域的类型；
根据所述腕骨区域的类型确定所述腕骨区域中骨化中心的数量。
由于在采用分类模型确定腕骨区域中骨化中心的数量时，先采用分割模 型从骨龄片中分割出腕骨区域，缩小检测范围，然后再从腕骨区域中确定骨化中心的数量，根据腕骨区域中骨化中心的数量确定骨龄，从而提高检测骨龄的精度。
可选地，所述调整所述骨龄片中的各骨头至基准位置，包括：
获取预设基准点的坐标；
采用调整模型确定所述骨龄片中关键点的坐标，所述调整模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的，所述关键点的坐标和所述预设基准点的坐标为同一坐标系中的坐标；
根据所述预设基准点的坐标和所述关键点的坐标确定所述骨龄片各骨头的当前位置与基准位置之间的对应关系；
根据所述对应关系将所述骨龄片中的各骨头调整至基准位置。
在确定腕骨区域之前，先将待检测用户的骨龄片中的各骨头调整至基准位置，从而提高确定腕骨区域的精度，进一步提高了检测骨龄的精度。
第二方面，本申请实施例提供了一种检测骨龄的装置，包括：
获取模块，用于获取待检测用户的骨龄片，并调整所述骨龄片中的各骨头至基准位置；
分割模块，用于从所述骨龄片中确定腕骨区域；
分类模块，用于采用分类模型确定所述腕骨区域中骨化中心的数量，所述分类模型是以预先标记骨化中心的数量的腕骨区域的骨龄片为训练样本，对深度残差网络进行训练后确定的；
检测模块，用于根据所述腕骨区域中骨化中心的数量确定所述待检测用户的骨龄。
可选地，所述分割模块具体用于：
采用分割模型确定所述骨龄片中腕骨区域对应的关键点的坐标，所述分割模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的；
根据所述腕骨区域对应的关键点的坐标确定所述腕骨区域。
可选地，所述分类模块具体用于：
将所述腕骨区域的骨龄片经N个连续的卷积特征提取块处理，得到所述腕骨区域的图像特征，N大于0，所述卷积特征提取块包括L个卷积模块，L大于0，任意一个卷积模块中包括卷积层、BN层及激励层；针对所述N个卷积特征提取块中任意两个连续的第一卷积特征提取块和第二卷积特征提取块，所述第二卷积特征提取块输出的第二图像特征与所述第一卷积特征提取块输出的第一图像特征相加后作为第三积特征提取块的输入或者N个连续的卷积特征提取块的输出；所述第三卷积特征提取块为位于所述第二卷积特征提取块之后且与所述第二卷积特征提取块连续的卷积特征提取块；
将所述腕骨区域的图像特征输入全连接层，输出所述腕骨区域的类型；
根据所述腕骨区域的类型确定所述腕骨区域中骨化中心的数量。
可选地，所述获取模块具体用于：
获取预设基准点的坐标；
采用调整模型确定所述骨龄片中关键点的坐标，所述调整模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的，所述关键点的坐标和所述预设基准点的坐标为同一坐标系中的坐标；
根据所述预设基准点的坐标和所述关键点的坐标确定所述骨龄片各骨头的当前位置与基准位置之间的对应关系；
根据所述对应关系将所述骨龄片中的各骨头调整至基准位置。
第三方面，本申请实施例提供了一种检测骨龄的设备，包括至少一个处理器、以及至少一个存储器，其中，所述存储单元存储有计算机程序，当所述程序被所述处理器执行时，使得所述处理器执行第一方面所述方法的步骤。
第四方面，本申请实施例提供了一种计算机可读介质，其存储有可由检测骨龄的设备执行的计算机程序，当所述程序在检测骨龄的设备上运行时，使得所述检测骨龄的设备执行第一方面所述方法的步骤。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简要介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域的普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的一种检测骨龄的方法的流程示意图；
图2为本申请实施例提供的一种骨龄片的示意图；
图3为本申请实施例提供的一种腕骨区域的示意图；
图4为本申请实施例提供的一种调整骨龄片的方法的流程示意图；
图5为本申请实施例提供的一种深度残差网络的结构示意图；
图6为本申请实施例提供的一种卷积特征提取块的结构示意图；
图7为本申请实施例提供的一种检测骨化中心数量的方法的流程示意图；
图8为本申请实施例提供的一种检测骨龄的装置的结构示意图；
图9为本申请实施例提供的一种检测骨龄的设备的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及有益效果更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
骨龄：骨骼年龄的简称，需要借助于骨骼在X光摄像中的特定图像来确定。通常要拍摄人左手手腕部的X光片，医生通过X光片观察左手掌指骨、腕骨及桡尺骨下端的骨化中心的发育程度，来确定骨龄。
本申请实施例中检测骨龄的技术方案适用于医院检测青少年的骨龄，为医生判定青少年生长发育情况提供参考，同时帮助医生诊断染色体异常、遗传性疾病、内分泌疾病等疾病。
图1示例性示出了本申请实施例提供的一种检测骨龄的方法的流程示意图，该流程可以由检测骨龄的装置执行，具体包括以下步骤：
步骤S101，获取待检测用户的骨龄片，并调整骨龄片中的各骨头至基准 位置。
骨龄片指采用X光拍摄的特定图像，通常拍摄人左手的X光片作为骨龄片，示例性地，如图2所示。
获取待检测用户的骨龄片之后，对骨龄片进行预处理，预处理过程主要包括腺体分割以及图像归一化。
腺体分割包括以下步骤：首先对整张骨龄片使用高斯滤波，将滤波的结果二值化，二值化的阈值通过求图像灰度直方图的最大类间距方法获得。然后将二值化的结果膨胀后通过漫水法(flood fill)获得一个个独立的区域块，统计每个区域块的面积。将面积最大的区域块保留，分割出来的手骨图像。将分割出的手骨图像粘贴到一张和手骨图像的长宽相符的纯黑图像上。
图像归一化包括以下步骤：骨龄片为dicom格式的图像，先根据dicom信息，选取窗宽窗位，转为png格式的骨龄片图像。通过在骨龄片图像上侧或两侧添加黑边，将骨龄片图像长宽比调整为1:1，最后将骨龄片图像缩放到512*512大小。
步骤S102，从骨龄片中确定腕骨区域。
腕骨区域位于手腕部，由8块小骨组成，排列成两排，近侧排自桡侧向尺侧为手舟骨、月骨、三角骨及豌豆骨，除豌豆骨外，均参与桡腕关节的组成。远侧排自桡侧向尺侧为大多角骨、小多角骨、头状骨及钩骨，均参与腕掌关节的组成，示例性地，如图3所示。
步骤S103，采用分类模型确定腕骨区域中骨化中心的数量。
骨发育过程中，首先骨化的部位称为骨化中心。腕骨区域中各小骨并不是同时生长出来的，而是依次生长，因此检测腕骨区域中各小骨的骨化中心，然后根据骨化中心的数量可以判断骨龄。
分类模型是以预先标记骨化中心的数量的腕骨区域的骨龄片为训练样本，对深度残差网络进行训练后确定的。
步骤S104，根据腕骨区域中骨化中心的数量确定待检测用户的骨龄。
具体地，预先设置腕骨区域中骨化中心的数量与骨龄之间的对照参考表， 在分类模型输出骨化中心的数量后，通过查表确定待检测用户的骨龄。
由于腕骨区域中骨化中心的数量代表着不同的骨龄阶段，因此本申请实施例中采用分类模型自动确定腕骨区域中骨化中心的数量，然后根据腕骨区域中骨化中心的数量确定待检测用户的骨龄，而不需要人工主观根据骨龄片判断骨龄，从而一方面提高了检测骨龄的精度，另一方面提高了检测骨龄的效率。相较于通过识别腕骨形态来确定骨龄的方法来说，根据腕骨区域中骨化中心的数量确定待检测用户的骨龄的方法，其效率更高。
可选地，在上述步骤S101中，调整骨龄片中的各骨头至基准位置，具体包括以下步骤，如图4所示：
步骤S401，获取预设基准点的坐标。
预设基准点的坐标可以是预设的部分骨头的坐标，用于表示部分骨头的基准位置，比如预设基准点的坐标可以是与中指相关的点的预设坐标，用于表示中指的基准位置，预设基准点的坐标也可以是与小指相关的点的预设坐标，用于表示小指的基准位置。
步骤S402，采用调整模型确定骨龄片中关键点的坐标，调整模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的。
关键点的坐标和预设基准点的坐标为同一坐标系中的坐标。
示例性地，获取多张骨龄片，然后人工标记骨龄片中小指周围的关键点，然后将标记关键点的骨龄片输入深度残差网络进行训练，当深度残差网络的目标函数满足预设条件时，确定调整模型。当获取待检测用户的骨龄片时，将骨龄片输入调整模型，确定骨龄片中小指周围的关键点。
步骤S403，根据预设基准点的坐标和关键点的坐标确定骨龄片各骨头的当前位置与基准位置之间的对应关系。
步骤S404，根据对应关系将骨龄片中的各骨头调整至基准位置。
当预设基准点和关键点都是与小指相关的点时，根据预设基准点的坐标和关键点的坐标可以确定小指在骨龄片中的当前位置与小指在骨龄片中的基 准位置之间的对应关系，进一步也可以得到骨龄片中其他骨头的当前位置与基准位置之间的对应关系，其中对应关系包括平移关系以及旋转关系。然后根据对应关系将骨龄片中的各骨头调整至基准位置。在确定腕骨区域之前，先将待检测用户的骨龄片中的各骨头调整至基准位置，从而提高确定腕骨区域的精度，进一步提高了检测骨龄的精度。
可选地，在上述步骤S102中，可以采用分割模型从未调整的骨龄片中确定腕骨区域，也可以采用分割模型从调整后的骨龄片中确定腕骨区域，其中，调整骨龄片指调整骨龄片中的各骨头至基准位置。分割模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的。当然，也可以采用其他神经网络模型进行训练，以获得分割模型，用于从调整后的骨龄片中确定腕骨区域，在此不做限定。
下面具体介绍分割模型的训练过程：获取多张骨龄片作为训练样本。针对每一张骨龄片，对骨龄片进行预处理，再将骨龄片中的各骨头调整至基准位置，其中，调整骨龄片的位置以及对骨龄片进行预处理的过程已在前文中描述，此处不再赘述。由标注人员在每张骨龄片中标注关键点，关键点为骨龄片中腕骨区域附近的点。然后对训练样本进行数据增强，将数据量增强至原来数据量的10倍，数据增强的方法包括但不限于：
1、随机旋转一定角度。
2、随机上下左右平移0～30像素。
3、随机缩放0.85～1.15倍。
4、对图像对比度和亮度进行少量抖动。
之后再将训练样本输入深度残差网络进行训练。训练时根据标注的关键点的坐标和网络预测的关键点的坐标计算loss函数，通过反向传播的方法训练，训练的优化算法使用带有动量和阶梯衰减的sgd算法。
进一步地，训练好分割模型后，采用分割模型确定骨龄片中腕骨区域对应的关键点的坐标，根据腕骨区域对应的关键点的坐标确定腕骨区域。
由于在采用分类模型确定腕骨区域中骨化中心的数量时，先采用分割模 型从骨龄片中分割出腕骨区域，缩小检测范围，然后再从腕骨区域中确定骨化中心的数量，根据腕骨区域中骨化中心的数量确定骨龄，从而提高检测骨龄的精度。
可选地，在上述步骤S103中，分类模型的训练过程如下：
获取从多张骨龄片分割出的腕骨区域的骨龄片作为训练样本。针对每一张腕骨区域的骨龄片。由标注人员在每张腕骨区域的骨龄片中标注骨化中心的数量。然后对训练样本进行数据增强，数据增强的方法包括但不限于：
1、随机旋转一定角度。
2、随机上下左右平移0～30像素。
3、随机缩放0.85～1.15倍。
4、对图像对比度和亮度进行少量抖动。
之后再将训练样本输入深度残差网络进行训练。训练时根据标注的标注骨化中心的数量和网络预测的标注骨化中心的数量计算loss函数，通过反向传播的方法训练，训练的优化算法使用带有动量和阶梯衰减的sgd算法。
可选地，上述深度残差网络的结构如图5所示，包括N个连续的卷积特征提取块以及一个全连接层，针对N个卷积特征提取块中任意两个连续的第一卷积特征提取块和第二卷积特征提取块，第二卷积特征提取块输出的第二图像特征与第一卷积特征提取块输出的第一图像特征相加后作为第三积特征提取块的输入或者N个连续的卷积特征提取块的输出。第三卷积特征提取块为位于第二卷积特征提取块之后且与第二卷积特征提取块连续的卷积特征提取块。卷积特征提取块包括L个卷积模块，L大于0，任意一个卷积模块中包括卷积层、BN层及激励层，具体如图6所示。
进一步地，训练好分类模型后，采用分类模型确定腕骨区域中骨化中心的数量，包括以下步骤，如图7所示：
步骤S701，将腕骨区域的骨龄片经N个连续的卷积特征提取块处理，得到腕骨区域的图像特征。
步骤S702，将腕骨区域的图像特征输入全连接层，输出腕骨区域的类型。
步骤S703，根据腕骨区域的类型确定腕骨区域中骨化中心的数量。
示例性地，全连接层输出的腕骨区域的类型为0～8类九个类型，其中0类表示腕骨区域中骨化中心的数量为0，1表示腕骨区域中骨化中心的数量为1，依次类推。然后进一步根据腕骨区域中骨化中心的数量查询对照表，确定待检测用户的骨龄。
由于腕骨区域中骨化中心的数量代表着不同的骨龄阶段，因此本申请实施例中采用分类模型自动确定腕骨区域中骨化中心的数量，然后根据腕骨区域中骨化中心的数量确定待检测用户的骨龄，而不需要人工主观根据骨龄片判断骨龄，从而一方面提高了检测骨龄的精度，另一方面提高了检测骨龄的效率。相较于通过识别腕骨形态来确定骨龄的方法来说，根据腕骨区域中骨化中心的数量确定待检测用户的骨龄的方法，其效率更高。
基于相同的技术构思，本申请实施例提供了一种检测骨龄的装置，如图8所示，该装置800包括：
获取模块801，用于获取待检测用户的骨龄片，并调整所述骨龄片中的各骨头至基准位置；
分割模块802，用于从所述骨龄片中确定腕骨区域；
分类模块803，用于采用分类模型确定所述腕骨区域中骨化中心的数量，所述分类模型是以预先标记骨化中心的数量的腕骨区域的骨龄片为训练样本，对深度残差网络进行训练后确定的；
检测模块804，用于根据所述腕骨区域中骨化中心的数量确定所述待检测用户的骨龄。
可选地，所述分割模块802具体用于：
采用分割模型确定所述骨龄片中腕骨区域对应的关键点的坐标，所述分割模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的；
根据所述腕骨区域对应的关键点的坐标确定所述腕骨区域。
可选地，所述分类模块803具体用于：
将所述腕骨区域的骨龄片经N个连续的卷积特征提取块处理，得到所述腕骨区域的图像特征，N大于0，所述卷积特征提取块包括L个卷积模块，L大于0，任意一个卷积模块中包括卷积层、BN层及激励层；针对所述N个卷积特征提取块中任意两个连续的第一卷积特征提取块和第二卷积特征提取块，所述第二卷积特征提取块输出的第二图像特征与所述第一卷积特征提取块输出的第一图像特征相加后作为第三积特征提取块的输入或者N个连续的卷积特征提取块的输出；所述第三卷积特征提取块为位于所述第二卷积特征提取块之后且与所述第二卷积特征提取块连续的卷积特征提取块；
将所述腕骨区域的图像特征输入全连接层，输出所述腕骨区域的类型；
根据所述腕骨区域的类型确定所述腕骨区域中骨化中心的数量。
可选地，所述获取模块801具体用于：
获取预设基准点的坐标；
采用调整模型确定所述骨龄片中关键点的坐标，所述调整模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的，所述关键点的坐标和所述预设基准点的坐标为同一坐标系中的坐标；
根据所述预设基准点的坐标和所述关键点的坐标确定所述骨龄片各骨头的当前位置与基准位置之间的对应关系；
根据所述对应关系将所述骨龄片中的各骨头调整至基准位置。
基于相同的技术构思，本申请实施例提供了一种检测骨龄的设备，如图9所示，包括至少一个处理器901，以及与至少一个处理器连接的存储器902，本申请实施例中不限定处理器901与存储器902之间的具体连接介质，图9中处理器901和存储器902之间通过总线连接为例。总线可以分为地址总线、数据总线、控制总线等。
在本申请实施例中，存储器902存储有可被至少一个处理器901执行的指令，至少一个处理器901通过执行存储器902存储的指令，可以执行前述的检测骨龄的方法中所包括的步骤。
其中，处理器901是检测骨龄的设备的控制中心，可以利用各种接口和 线路连接检测骨龄的设备的各个部分，通过运行或执行存储在存储器902内的指令以及调用存储在存储器902内的数据，从而实现检测骨龄。可选的，处理器901可包括一个或多个处理单元，处理器901可集成应用处理器和调制解调处理器，其中，应用处理器主要处理操作系统、用户界面和应用程序等，调制解调处理器主要处理无线通信。可以理解的是，上述调制解调处理器也可以不集成到处理器901中。在一些实施例中，处理器901和存储器902可以在同一芯片上实现，在一些实施例中，它们也可以在独立的芯片上分别实现。
处理器901可以是通用处理器，例如中央处理器(CPU)、数字信号处理器、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件，可以实现或者执行本申请实施例中公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者任何常规的处理器等。结合本申请实施例所公开的方法的步骤可以直接体现为硬件处理器执行完成，或者用处理器中的硬件及软件模块组合执行完成。
存储器902作为一种非易失性计算机可读存储介质，可用于存储非易失性软件程序、非易失性计算机可执行程序以及模块。存储器902可以包括至少一种类型的存储介质，例如可以包括闪存、硬盘、多媒体卡、卡型存储器、随机访问存储器(Random Access Memory，RAM)、静态随机访问存储器(Static Random Access Memory，SRAM)、可编程只读存储器(Programmable Read Only Memory，PROM)、只读存储器(Read Only Memory，ROM)、带电可擦除可编程只读存储器(Electrically Erasable Programmable Read-Only Memory，EEPROM)、磁性存储器、磁盘、光盘等等。存储器902是能够用于携带或存储具有指令或数据结构形式的期望的程序代码并能够由计算机存取的任何其他介质，但不限于此。本申请实施例中的存储器902还可以是电路或者其它任意能够实现存储功能的装置，用于存储程序指令和/或数据。
基于同一发明构思，本申请实施例还提供了一种计算机可读介质，其存 储有可由检测骨龄的设备执行的计算机程序，当所述程序在检测骨龄的设备上运行时，使得所述检测骨龄的设备执行检测骨龄的方法的步骤。
本领域内的技术人员应明白，本发明的实施例可提供为方法、或计算机程序产品。因此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
尽管已描述了本发明的优选实施例，但本领域内的技术人员一旦得知了基本创造性概念，则可对这些实施例作出另外的变更和修改。所以，所附权利要求意欲解释为包括优选实施例以及落入本发明范围的所有变更和修改。
显然，本领域的技术人员可以对本发明进行各种改动和变型而不脱离本 发明的精神和范围。这样，倘若本发明的这些修改和变型属于本发明权利要求及其等同技术的范围之内，则本发明也意图包含这些改动和变型在内。

Claims (10)

  1. 一种检测骨龄的方法，其特征在于，包括：
    获取待检测用户的骨龄片，并调整所述骨龄片中的各骨头至基准位置；
    从所述骨龄片中确定腕骨区域；
    采用分类模型确定所述腕骨区域中骨化中心的数量，所述分类模型是以预先标记骨化中心的数量的腕骨区域的骨龄片为训练样本，对深度残差网络进行训练后确定的；
    根据所述腕骨区域中骨化中心的数量确定所述待检测用户的骨龄。
  2. 如权利要求1所述的方法，其特征在于，所述从所述骨龄片中确定腕骨区域，包括：
    采用分割模型确定所述骨龄片中腕骨区域对应的关键点的坐标，所述分割模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的；
    根据所述腕骨区域对应的关键点的坐标确定所述腕骨区域。
  3. 如权利要求1所述的方法，其特征在于，所述采用分类模型确定所述腕骨区域中骨化中心的数量，包括：
    将所述腕骨区域的骨龄片经N个连续的卷积特征提取块处理，得到所述腕骨区域的图像特征，N大于0，所述卷积特征提取块包括L个卷积模块，L大于0，任意一个卷积模块中包括卷积层、BN层及激励层；针对所述N个卷积特征提取块中任意两个连续的第一卷积特征提取块和第二卷积特征提取块，所述第二卷积特征提取块输出的第二图像特征与所述第一卷积特征提取块输出的第一图像特征相加后作为第三积特征提取块的输入或者N个连续的卷积特征提取块的输出；所述第三卷积特征提取块为位于所述第二卷积特征提取块之后且与所述第二卷积特征提取块连续的卷积特征提取块；
    将所述腕骨区域的图像特征输入全连接层，输出所述腕骨区域的类型；
    根据所述腕骨区域的类型确定所述腕骨区域中骨化中心的数量。
  4. 如权利要求1至3任一所述的方法，其特征在于，所述调整所述骨龄片中的各骨头至基准位置，包括：
    获取预设基准点的坐标；
    采用调整模型确定所述骨龄片中关键点的坐标，所述调整模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的，所述关键点的坐标和所述预设基准点的坐标为同一坐标系中的坐标；
    根据所述预设基准点的坐标和所述关键点的坐标确定所述骨龄片各骨头的当前位置与基准位置之间的对应关系；
    根据所述对应关系将所述骨龄片中的各骨头调整至基准位置。
  5. 一种检测骨龄的装置，其特征在于，包括：
    获取模块，用于获取待检测用户的骨龄片，并调整所述骨龄片中的各骨头至基准位置；
    分割模块，用于从所述骨龄片中确定腕骨区域；
    分类模块，用于采用分类模型确定所述腕骨区域中骨化中心的数量，所述分类模型是以预先标记骨化中心的数量的腕骨区域的骨龄片为训练样本，对深度残差网络进行训练后确定的；
    检测模块，用于根据所述腕骨区域中骨化中心的数量确定所述待检测用户的骨龄。
  6. 如权利要求5所述的装置，其特征在于，所述分割模块具体用于：
    采用分割模型确定所述骨龄片中腕骨区域对应的关键点的坐标，所述分割模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的；
    根据所述腕骨区域对应的关键点的坐标确定所述腕骨区域。
  7. 如权利要求5所述的装置，其特征在于，所述分类模块具体用于：
    将所述腕骨区域的骨龄片经N个连续的卷积特征提取块处理，得到所述腕骨区域的图像特征，N大于0，所述卷积特征提取块包括L个卷积模块，L大于0，任意一个卷积模块中包括卷积层、BN层及激励层；针对所述N个卷 积特征提取块中任意两个连续的第一卷积特征提取块和第二卷积特征提取块，所述第二卷积特征提取块输出的第二图像特征与所述第一卷积特征提取块输出的第一图像特征相加后作为第三积特征提取块的输入或者N个连续的卷积特征提取块的输出；所述第三卷积特征提取块为位于所述第二卷积特征提取块之后且与所述第二卷积特征提取块连续的卷积特征提取块；
    将所述腕骨区域的图像特征输入全连接层，输出所述腕骨区域的类型；
    根据所述腕骨区域的类型确定所述腕骨区域中骨化中心的数量。
  8. 如权利要求5至7任一所述的装置，其特征在于，所述获取模块具体用于：
    获取预设基准点的坐标；
    采用调整模型确定所述骨龄片中关键点的坐标，所述调整模型是以预先标记关键点的坐标的多张骨龄片为训练样本，对深度残差网络进行训练后确定的，所述关键点的坐标和所述预设基准点的坐标为同一坐标系中的坐标；
    根据所述预设基准点的坐标和所述关键点的坐标确定所述骨龄片各骨头的当前位置与基准位置之间的对应关系；
    根据所述对应关系将所述骨龄片中的各骨头调整至基准位置。
  9. 一种检测骨龄的设备，其特征在于，包括至少一个处理器、以及至少一个存储器，其中，所述存储单元存储有计算机程序，当所述程序被所述处理器执行时，使得所述处理器执行权利要求1～4任一权利要求所述方法的步骤。
  10. 一种计算机可读介质，其特征在于，其存储有可由检测骨龄的设备执行的计算机程序，当所述程序在检测骨龄的设备上运行时，使得所述检测骨龄的设备执行权利要求1～4任一所述方法的步骤。
PCT/CN2019/082682 2018-09-30 2019-04-15 一种检测骨龄的方法及装置 WO2020062840A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
SG11202002140VA SG11202002140VA (en) 2018-09-30 2019-04-15 Method and device for detecting a bone age

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811163210.7 2018-09-30
CN201811163210.7A CN109146879B (zh) 2018-09-30 2018-09-30 一种检测骨龄的方法及装置

Publications (1)

Publication Number Publication Date
WO2020062840A1 true WO2020062840A1 (zh) 2020-04-02

Family

ID=64810680

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/082682 WO2020062840A1 (zh) 2018-09-30 2019-04-15 一种检测骨龄的方法及装置

Country Status (3)

Country Link
CN (1) CN109146879B (zh)
SG (1) SG11202002140VA (zh)
WO (1) WO2020062840A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111507953A (zh) * 2020-04-13 2020-08-07 武汉华晨酷神智能科技有限公司 一种ai骨龄快速识别方法
CN111709874A (zh) * 2020-06-16 2020-09-25 北京百度网讯科技有限公司 图像调整的方法、装置、电子设备及存储介质
CN112102305A (zh) * 2020-09-24 2020-12-18 石家庄喜高科技有限责任公司 多骨骼发育等级检测方法及终端设备
CN112132788A (zh) * 2020-08-29 2020-12-25 浙江工业大学 一种基于特征区域等级识别的骨龄评估方法
CN112801994A (zh) * 2021-02-05 2021-05-14 广东顺德工业设计研究院(广东顺德创新设计研究院) 骨龄评估方法和系统
CN113592820A (zh) * 2021-07-30 2021-11-02 瓴域影诺(北京)科技有限公司 一种股骨区域及关键点检测方法和系统
CN114663426A (zh) * 2022-04-21 2022-06-24 重庆邮电大学 一种基于关键骨区定位的骨龄评估方法
CN115527235A (zh) * 2022-09-27 2022-12-27 内蒙古工业大学 一种基于图像处理识别蒙医手部穴位的方法与装置
CN117423472A (zh) * 2023-12-18 2024-01-19 天津医科大学总医院 一种基于骨龄识别的多要素身高预测方法及系统

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109146879B (zh) * 2018-09-30 2021-05-18 杭州依图医疗技术有限公司 一种检测骨龄的方法及装置
CN109949280B (zh) * 2019-03-06 2021-06-11 上海联影智能医疗科技有限公司 图像处理方法、装置、设备存储介质及生长发育评估系统
US11367181B2 (en) 2018-12-29 2022-06-21 Shanghai United Imaging Intelligence Co., Ltd. Systems and methods for ossification center detection and bone age assessment
CN110516715B (zh) * 2019-08-05 2022-02-11 杭州依图医疗技术有限公司 一种手骨分类方法及装置
CN111553412A (zh) * 2020-04-27 2020-08-18 广州市妇女儿童医疗中心（广州市妇幼保健院、广州市儿童医院、广州市妇婴医院、广州市妇幼保健计划生育服务中心） 对性早熟分类模型进行训练的方法、装置和设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107591200A (zh) * 2017-08-25 2018-01-16 卫宁健康科技集团股份有限公司 基于深度学习及影像组学的骨龄标记识别评估方法及系统
CN107895367A (zh) * 2017-11-14 2018-04-10 中国科学院深圳先进技术研究院 一种骨龄识别方法、系统及电子设备
CN107944496A (zh) * 2017-12-06 2018-04-20 电子科技大学 基于改进后的残差网络的骨龄自动化识别系统
CN108056786A (zh) * 2017-12-08 2018-05-22 浙江大学医学院附属儿童医院 一种基于深度学习的骨龄检测方法和装置
CN109146879A (zh) * 2018-09-30 2019-01-04 杭州依图医疗技术有限公司 一种检测骨龄的方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104573615B (zh) * 2013-10-24 2018-03-16 华为技术有限公司 掌纹采集方法及装置
CN104504376A (zh) * 2014-12-22 2015-04-08 厦门美图之家科技有限公司 一种人脸图像的年龄分类方法和系统
CN106339680B (zh) * 2016-08-25 2019-07-23 北京小米移动软件有限公司 人脸关键点定位方法及装置
CN107066983B (zh) * 2017-04-20 2022-08-09 腾讯科技（上海）有限公司 一种身份验证方法及装置
CN107247949B (zh) * 2017-08-02 2020-06-19 智慧眼科技股份有限公司 基于深度学习的人脸识别方法、装置和电子设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107591200A (zh) * 2017-08-25 2018-01-16 卫宁健康科技集团股份有限公司 基于深度学习及影像组学的骨龄标记识别评估方法及系统
CN107895367A (zh) * 2017-11-14 2018-04-10 中国科学院深圳先进技术研究院 一种骨龄识别方法、系统及电子设备
CN107944496A (zh) * 2017-12-06 2018-04-20 电子科技大学 基于改进后的残差网络的骨龄自动化识别系统
CN108056786A (zh) * 2017-12-08 2018-05-22 浙江大学医学院附属儿童医院 一种基于深度学习的骨龄检测方法和装置
CN109146879A (zh) * 2018-09-30 2019-01-04 杭州依图医疗技术有限公司 一种检测骨龄的方法及装置

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111507953A (zh) * 2020-04-13 2020-08-07 武汉华晨酷神智能科技有限公司 一种ai骨龄快速识别方法
CN111709874B (zh) * 2020-06-16 2023-09-08 北京百度网讯科技有限公司 图像调整的方法、装置、电子设备及存储介质
CN111709874A (zh) * 2020-06-16 2020-09-25 北京百度网讯科技有限公司 图像调整的方法、装置、电子设备及存储介质
CN112132788A (zh) * 2020-08-29 2020-12-25 浙江工业大学 一种基于特征区域等级识别的骨龄评估方法
CN112132788B (zh) * 2020-08-29 2024-04-16 浙江工业大学 一种基于特征区域等级识别的骨龄评估方法
CN112102305A (zh) * 2020-09-24 2020-12-18 石家庄喜高科技有限责任公司 多骨骼发育等级检测方法及终端设备
CN112801994B (zh) * 2021-02-05 2023-12-26 广东顺德工业设计研究院(广东顺德创新设计研究院) 骨龄评估方法和系统
CN112801994A (zh) * 2021-02-05 2021-05-14 广东顺德工业设计研究院(广东顺德创新设计研究院) 骨龄评估方法和系统
CN113592820A (zh) * 2021-07-30 2021-11-02 瓴域影诺(北京)科技有限公司 一种股骨区域及关键点检测方法和系统
CN114663426A (zh) * 2022-04-21 2022-06-24 重庆邮电大学 一种基于关键骨区定位的骨龄评估方法
CN115527235A (zh) * 2022-09-27 2022-12-27 内蒙古工业大学 一种基于图像处理识别蒙医手部穴位的方法与装置
CN117423472A (zh) * 2023-12-18 2024-01-19 天津医科大学总医院 一种基于骨龄识别的多要素身高预测方法及系统
CN117423472B (zh) * 2023-12-18 2024-03-08 天津医科大学总医院 一种基于骨龄识别的多要素身高预测方法及系统

Also Published As

Publication number Publication date
CN109146879A (zh) 2019-01-04
SG11202002140VA (en) 2020-05-28
CN109146879B (zh) 2021-05-18

Similar Documents

Publication Publication Date Title
WO2020062840A1 (zh) 一种检测骨龄的方法及装置
WO2020062839A1 (zh) 一种检测骨龄的方法及装置
Oh et al. Deep anatomical context feature learning for cephalometric landmark detection
CN109741309B (zh) 一种基于深度回归网络的骨龄预测方法及装置
CN109215013B (zh) 自动骨龄预测方法、系统、计算机设备和存储介质
CN109272002B (zh) 一种骨龄片的分类方法及装置
CN107945870B (zh) 基于深度神经网络的早产儿视网膜病变检测方法及装置
CN109509177B (zh) 一种脑部影像识别的方法及装置
Almazroa et al. Optic disc segmentation for glaucoma screening system using fundus images
CN109255786B (zh) 一种检测骨龄的方法及装置
Van Steenkiste et al. Automated assessment of bone age using deep learning and Gaussian process regression
CN110689551B (zh) 四肢骨分割方法、装置、电子设备及可读存储介质
CN112150472A (zh) 基于cbct的三维颌骨图像分割方法、装置及终端设备
Lee et al. Deep learning based cephalometric landmark identification using landmark-dependent multi-scale patches
CN112529918B (zh) 一种脑部ct图像中脑室区域分割的方法、装置及设备
WO2023047117A1 (en) A computer-implemented method, data processing apparatus, and computer program for active learning for computer vision in digital images
CN110555846A (zh) 基于卷积神经网络的全自动骨龄评定方法
CN113425248B (zh) 医疗影像评估方法、装置、设备及计算机存储介质
Goutham et al. Automatic localization of landmarks in cephalometric images via modified U-Net
CN117153394A (zh) 一种预测恒牙发育的方法及系统
CN115880358A (zh) 定位模型的构建方法、影像标志点的定位方法及电子设备
CN110516715B (zh) 一种手骨分类方法及装置
CN113920088B (zh) 一种基于深度学习的桡骨密度检测方法、系统、装置
CN114549523A (zh) 基于单步深度网络的曲面体层图正中多生牙自动检测方法
CN114557670A (zh) 生理年龄预测方法、装置、设备及介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19867911

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19867911

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 29.09.2021)

122 Ep: pct application non-entry in european phase

Ref document number: 19867911

Country of ref document: EP

Kind code of ref document: A1