WO2021147218A1 - 医学影像识别检测方法、装置、设备及存储介质 - Google Patents

医学影像识别检测方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2021147218A1
WO2021147218A1 PCT/CN2020/093556 CN2020093556W WO2021147218A1 WO 2021147218 A1 WO2021147218 A1 WO 2021147218A1 CN 2020093556 W CN2020093556 W CN 2020093556W WO 2021147218 A1 WO2021147218 A1 WO 2021147218A1
Authority
WO
WIPO (PCT)
Prior art keywords
pathological
image set
pathological image
enhanced
image
Prior art date
Application number
PCT/CN2020/093556
Other languages
English (en)
French (fr)
Inventor
侯晓帅
李风仪
南洋
谢春梅
王佳平
Original Assignee
平安科技（深圳）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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021147218A1 publication Critical patent/WO2021147218A1/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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • 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
    • G06T5/70—
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/30084—Kidney; Renal

Definitions

  • This application relates to the field of artificial intelligence technology, and in particular to a medical image recognition and detection method, device, equipment, and storage medium.
  • AI technology can help doctors locate lesions and analyze the condition, assisting doctors to make accurate and rapid diagnosis.
  • AI applications are mainly concentrated in lung nodules, fundus, breast, etc.
  • AI technology is also applied in digital pathological diagnosis.
  • the inventor realizes that in the current clinical kidney biopsy pathological diagnosis process, the pathologist observes the pathological morphology of the glomerulus in the pathological slice image through the optical microscope, the distribution of cell proliferation in the glomerulus, and the pathologist’s own experience to perform pathological analysis and give feedback to the pathologist.
  • a pathological diagnosis report is issued. Due to the large number of glomeruli in the slice image, the workload of identifying glomeruli by visual observation is extremely large, the efficiency is relatively low, and the diagnosis results are easily affected by the doctor's subjective factors.
  • This application provides a medical image recognition and detection method, device, equipment, and storage medium, the main purpose of which is to provide an intelligent glomerulus detection solution.
  • a medical image recognition and detection method includes:
  • the enhanced pathological image set is input into a pre-built pathological detection model, the weight parameters of the enhanced pathological image set are obtained through forward propagation in the pathological detection model, and the weight parameters are updated by using a gradient descent algorithm until The pathology detection model tends to converge, and the pathology detection model after training is obtained;
  • Receive the pathological image to be detected input by the user detect the pathological image to be detected through the pathological detection model after the training is completed, and return the detection result to the user.
  • the normalization operation includes:
  • x represents the image data in the pathological image set
  • ⁇ is the mean value of the image data in the pathological image set
  • ⁇ is the standard deviation of the image data in the pathological image set.
  • the image noise reduction processing includes:
  • f(x,y) represents the pathological image after noise reduction
  • g(x,y) represents the pathological image to be processed
  • the obtaining the weight parameter of the enhanced pathological image set through forward propagation in the pathology detection model includes:
  • the front scenic spot of the enhanced pathological image set is obtained, and the border position of the front scenic spot is calculated by a neutral branch algorithm, and a predetermined weight parameter is used according to the border border position
  • the calculation formula calculates the weight parameter of the enhanced pathological image set.
  • the calculating the weight parameter of the enhanced pathological image set according to the position of the border boundary using a predetermined weight parameter calculation formula includes:
  • the weight parameters of the enhanced pathological image set are calculated using the following formula:
  • Loss iou represents the weight parameter
  • l p , r p , t p , b p respectively represent the predicted distance value of the upper, lower, left, and right borders of the scenic spot before the enhanced pathological image set
  • l g , r g , t g , B g respectively represent the actual distance values of the upper, lower, left and right borders of the scenic spot in front of the enhanced pathological image set.
  • the present application also provides a medical image recognition and detection device, which includes:
  • the normalized noise reduction module is used to obtain a pathological image set, perform a normalization operation on the pathological image set, and perform image denoising processing on the pathological image set after the normalization operation to obtain a standard pathological image set ;
  • An image enhancement module for performing image enhancement processing on the standard pathological image set to obtain an enhanced pathological image set
  • the model training module is used to input the enhanced pathological image set into a pre-built pathology detection model, obtain the weight parameters of the enhanced pathological image set through the forward propagation in the pathology detection model, and use the gradient descent algorithm to determine the weight parameters of the enhanced pathological image set.
  • the weight parameters are updated until the pathology detection model tends to converge, and the pathology detection model after training is obtained;
  • the detection module is used to receive the image to be detected input by the user, to detect the image to be detected through the pathological detection model after the training is completed, to obtain the detection result of the image to be detected, and to return the detection result to the user.
  • the present application also provides a medical image recognition and detection device, which includes a memory and a processor.
  • the memory stores a medical image recognition and detection program that can run on the processor.
  • the medical image recognition and detection program is executed by the processor, the following steps are implemented:
  • the enhanced pathological image set is input into a pre-built pathological detection model, the weight parameters of the enhanced pathological image set are obtained through forward propagation in the pathological detection model, and the weight parameters are updated by using a gradient descent algorithm until The pathology detection model tends to converge, and the pathology detection model after training is obtained;
  • Receive the pathological image to be detected input by the user detect the pathological image to be detected through the pathological detection model after the training is completed, and return the detection result to the user.
  • the normalization operation includes:
  • x represents the image data in the pathological image set
  • ⁇ is the mean value of the image data in the pathological image set
  • ⁇ is the standard deviation of the image data in the pathological image set.
  • the image noise reduction processing includes:
  • f(x,y) represents the pathological image after noise reduction
  • g(x,y) represents the pathological image to be processed
  • the obtaining the weight parameter of the enhanced pathological image set through forward propagation in the pathology detection model includes:
  • the front scenic spot of the enhanced pathological image set is obtained, and the border position of the front scenic spot is calculated by a neutral branch algorithm, and a predetermined weight parameter is used according to the border border position
  • the calculation formula calculates the weight parameter of the enhanced pathological image set.
  • the present application also provides a computer-readable storage medium having a medical image recognition and detection program stored on the computer-readable storage medium, and the medical image recognition and detection program can be used by one or more processors. Execute to realize the steps of the medical image recognition and detection method as described above.
  • the medical image recognition and detection method, device, equipment, and storage medium proposed in this application acquire a pathological image set when the user performs glomerular detection, and perform normalization, image denoising and image enhancement processing on the pathological image set, An enhanced pathological cell image set can be obtained, and the pre-built pathological detection model can be obtained by using the enhanced pathological cell image set to obtain a trained pathological detection model, and the pathological image to be detected is detected according to the pathological detection model after the training is completed. , And return the detection results to the user, so that the detection results of diseases such as glomeruli can be intelligently identified from medical images.
  • FIG. 1 is a schematic flowchart of a medical image recognition and detection method provided by an embodiment of the application
  • FIG. 2 is a schematic diagram of the internal structure of a medical image recognition and detection device provided by an embodiment of the application;
  • FIG. 3 is a schematic diagram of the modules of the medical image recognition and detection device in the medical image recognition and detection device provided by an embodiment of the application.
  • This application provides a medical image recognition and detection method.
  • FIG. 1 it is a schematic flowchart of a medical image recognition and detection method provided by an embodiment of this application.
  • the method can be executed by a device, and the device can be implemented by software and/or hardware.
  • the medical image recognition and detection method includes:
  • the pathological image set may be, for example, a pathological image of cell proliferation in the glomerulus.
  • the acquisition of a pathological image set includes: taking kidney tissue of a preset size, and placing a liquid fixative to fix the kidney tissue, and using different concentrations of ethanol to fix the fixed kidney tissue.
  • the kidney tissue is dehydrated, the dehydrated kidney tissue is stained by the Pap staining method, and the stained kidney tissue is sliced according to preset specifications according to the pathological section scanner, so as to obtain the pathological image set.
  • the predetermined size of the kidney tissue is 2.0cm*2.0cm*0.3cm
  • the dehydration process is automatically controlled by a dehydrator
  • the predetermined size is a pathological image of 1024*1024.
  • the normalization of the data is to scale the data to a small specific interval.
  • the normalization processing method applied in this application is the z-score normalization method, and its core idea is as follows:
  • x represents the image data in the pathological image set
  • ⁇ is the mean value of the image data in the pathological image set
  • ⁇ is the standard deviation of the image data in the pathological image set.
  • the image noise reduction processing is performed on the pathological image set after the normalization operation through a filtering algorithm.
  • the filtering algorithm includes:
  • f(x,y) represents the pathological image after noise reduction
  • g(x,y) represents the pathological image to be processed
  • the embodiment of the present application uses data enhancement technology to perform image enhancement processing on the standard pathological image set.
  • the data enhancement techniques described in this application include random flipping and random tailoring.
  • this application uses the cv2 instruction in the python library to set the corresponding flip parameters for the standard pathological image set, so that the standard pathological image set has a 1/3 probability of horizontal flipping, vertical flipping, and horizontal and vertical flipping, where cv .flip(img,1) represents a horizontal rotation of 180 degrees, cv.flip(img,0) represents a vertical rotation of 180 degrees, and cv.flip(img,-1) represents a horizontal and vertical rotation of 180 degrees at the same time.
  • this application uses the randomCrop(image) function to randomly crop the set of pathological images that are randomly flipped, so that the size of the pathological image after cropping is 2/3 of the width and height of the original image.
  • the data of the pathological cell image is enhanced to obtain the enhanced pathological cell image set.
  • the pre-built kidney proportional glomerulus detection model is implemented based on the Mask R-CNN framework, which is a detection and instance segmentation model framework.
  • the obtaining the weight parameters of the enhanced pathological image set through the forward propagation in the pathology detection model includes: performing classification branch processing on the enhanced pathological image set to obtain the front spot of the enhanced pathological image set, and passing the The sexual branching algorithm calculates the border position of the front scenic spot, and uses a predetermined weight parameter calculation formula to calculate the weight parameter of the enhanced pathological image set according to the border border position.
  • the classification branch algorithm includes:
  • Centerness represents pathologically enhanced border boundary of landscape image set, l p, r p, t p, b p respectively represent the front reinforcing frame boundary pathological landscape image set, down, left, and right from the predicted value.
  • the weight parameter calculation formula is:
  • Loss iou represents the weight parameter
  • l p , r p , t p , b p respectively represent the predicted distance value of the upper, lower, left, and right borders of the scenic spot before the enhanced pathological image set
  • l g , r g , t g , B g respectively represent the actual distance values of the upper, lower, left and right borders of the scenic spot in front of the enhanced pathological image set.
  • the gradient descent algorithm is an optimization algorithm in neural network model training. In order to find the weight parameters that the pre-built pathology detection model tends to converge, it is necessary to follow the gradient vector of the pre-built pathology detection model. The variables are updated in the opposite direction, so that the gradient can be reduced the fastest until the pre-built pathology detection model converges to obtain the pathology detection model after the training is completed.
  • the image to be detected input by the user is received, and the image to be detected input by the user is detected according to the pathology detection model after the above training is completed, and the detection result of the image is obtained, and the detection result is Return to the user.
  • the detection result includes two kinds of normal nos and hardened gs.
  • the application also provides a medical image recognition and detection equipment.
  • FIG. 2 it is a schematic diagram of the internal structure of the medical image recognition detection device 1 provided by an embodiment of this application.
  • the medical image recognition and detection device 1 may be a PC (Personal Computer, personal computer), or a terminal device such as a smart phone, a tablet computer, or a portable computer, or a server.
  • the medical image recognition and detection equipment 1 at least includes a memory 11, a processor 12, a communication bus 13, and a network interface 14.
  • the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, and the like.
  • the memory 11 may be an internal storage unit of the medical image recognition and detection device 1 in some embodiments, such as a hard disk of the medical image recognition and detection device 1.
  • the memory 11 may also be an external storage device of the medical image recognition and detection device 1, for example, a plug-in hard disk equipped on the medical image recognition and detection device 1, a smart media card (SMC), and a secure digital (Secure Digital, SD) card, Flash Card, etc.
  • the memory 11 may also include both an internal storage unit of the medical image recognition and detection device 1 and an external storage device.
  • the memory 11 can not only be used to store application software and various data installed in the medical image recognition and detection device 1, such as the code of the medical image recognition and detection program 01, etc., but also can be used to temporarily store data that has been output or will be output.
  • the processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip, for running program codes or processing stored in the memory 11 Data, such as the execution of medical image recognition and detection program 01, etc.
  • CPU central processing unit
  • controller microcontroller
  • microprocessor or other data processing chip, for running program codes or processing stored in the memory 11 Data, such as the execution of medical image recognition and detection program 01, etc.
  • the communication bus 13 is used to realize the connection and communication between these components.
  • the network interface 14 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface), and is usually used to establish a communication connection between the device 1 and other electronic devices.
  • the device 1 may also include a user interface.
  • the user interface may include a display (Display) and an input unit such as a keyboard (Keyboard).
  • the optional user interface may also include a standard wired interface and a wireless interface.
  • the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc.
  • the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the medical image recognition and detection device 1 and to display a visualized user interface.
  • Figure 2 only shows the medical image recognition and detection device 1 with components 11-14 and the medical image recognition and detection program 01. Those skilled in the art can understand that the structure shown in Figure 1 does not constitute a medical image recognition and detection device.
  • the definition of 1 may include fewer or more components than those shown in the figure, or a combination of certain components, or different component arrangements.
  • the memory 11 stores the medical image recognition and detection program 01; when the processor 12 executes the medical image recognition and detection program 01 stored in the memory 11, the following steps are implemented:
  • Step 1 Obtain a pathological image set, perform a normalization operation on the pathological image set, and perform image denoising processing on the pathological image set after the normalization operation to obtain a standard pathological image set
  • the acquisition of a pathological image set includes: taking kidney tissue of a preset size, and placing a liquid fixative to fix the kidney tissue, and using different concentrations of ethanol to fix the fixed kidney tissue.
  • the kidney tissue is dehydrated, the dehydrated kidney tissue is stained by the Pap staining method, and the stained kidney tissue is sliced according to preset specifications according to the pathological section scanner, so as to obtain the pathological image set.
  • the predetermined size of the kidney tissue is 2.0cm*2.0cm*0.3cm
  • the dehydration process is automatically controlled by a dehydrator
  • the predetermined size is a pathological image of 1024*1024.
  • the normalization of the data is to scale the data to a small specific interval.
  • the normalization processing method applied in this application is the z-score normalization method, and its core idea is as follows:
  • x represents the image data in the pathological image set
  • ⁇ is the mean value of the image data in the pathological image set
  • ⁇ is the standard deviation of the image data in the pathological image set.
  • the image noise reduction processing is performed on the pathological image set after the normalization operation through a filtering algorithm.
  • the filtering algorithm includes:
  • f(x,y) represents the pathological image after noise reduction
  • g(x,y) represents the pathological image to be processed
  • Step 2 Perform image enhancement processing on the standard pathological image to obtain an enhanced pathological image set.
  • the embodiment of the present application uses data enhancement technology to perform image enhancement processing on the standard pathological image set.
  • the data enhancement techniques described in this application include random flipping and random tailoring.
  • this application uses the cv2 instruction in the python library to set the corresponding flip parameters for the standard pathological image set, so that the standard pathological image set has a 1/3 probability of horizontal flipping, vertical flipping, and horizontal and vertical flipping, where cv .flip(img,1) represents a horizontal rotation of 180 degrees, cv.flip(img,0) represents a vertical rotation of 180 degrees, and cv.flip(img,-1) represents a horizontal and vertical rotation of 180 degrees at the same time.
  • this application uses the randomCrop(image) function to randomly crop the set of pathological images that are randomly flipped, so that the size of the pathological image after cropping is 2/3 of the width and height of the original image.
  • the data of the pathological cell image is enhanced to obtain the enhanced pathological cell image set.
  • Step 3 Input the enhanced pathological image set into a pre-built pathology detection model, obtain the weight parameters of the enhanced pathological image set through forward propagation in the pathology detection model, and use a gradient descent algorithm to calculate the weight parameters Update until the pathology detection model tends to converge, and the pathology detection model after training is obtained.
  • the pre-built kidney proportional glomerulus detection model is implemented based on the Mask R-CNN framework, which is a detection and instance segmentation model framework.
  • the obtaining the weight parameters of the enhanced pathological image set through the forward propagation in the pathology detection model includes: performing classification branch processing on the enhanced pathological image set to obtain the front spot of the enhanced pathological image set, and passing the The sexual branching algorithm calculates the border position of the front scenic spot, and uses a predetermined weight parameter calculation formula to calculate the weight parameter of the enhanced pathological image set according to the border border position.
  • the classification branch algorithm includes:
  • Centerness represents pathologically enhanced border boundary of landscape image set, l p, r p, t p, b p respectively represent the front reinforcing frame boundary pathological landscape image set, down, left, and right from the predicted value.
  • the weight parameter calculation formula is:
  • Loss iou represents the weight parameter
  • l p , r p , t p , b p respectively represent the predicted distance value of the upper, lower, left, and right borders of the scenic spot before the enhanced pathological image set
  • l g , r g , t g , B g respectively represent the actual distance values of the upper, lower, left and right borders of the scenic spot in front of the enhanced pathological image set.
  • the gradient descent algorithm is the most commonly used optimization algorithm for neural network model training. In order to find the weight parameters that the pre-built pathology detection model tends to converge, it is necessary to follow the gradient of the pre-built pathology detection model. The variables are updated in the opposite direction of the vector, so that the gradient can be reduced the fastest until the pre-built pathology detection model converges, and the pathology detection model after the training is completed.
  • Step 4 Receive the image to be detected input by the user, detect the image to be detected through the pathology detection model after the training is completed, and return the detection result to the user.
  • the image to be detected input by the user is received, and the image to be detected input by the user is detected according to the pathology detection model after the above training is completed, and the detection result of the image is obtained, and the detection result is Return to the user.
  • the detection result includes two kinds of normal nos and hardened gs.
  • the medical image recognition and detection program 100 may also be divided into one or more modules, and the one or more modules are stored in the memory 11 and are executed by one or more processors (in this embodiment). For example, it is executed by the processor 12) to complete this application.
  • the module referred to in this application refers to a series of computer program instruction segments that can complete specific functions, and is used to describe the medical image recognition and detection program 01 in the medical image recognition and detection device 100 The implementation process.
  • FIG. 3 a framework diagram of the apparatus 100 of the medical image recognition and detection program in an embodiment of the medical image recognition and detection apparatus of this application.
  • the medical image recognition and detection apparatus 100 can be divided into groups.
  • the unified noise reduction module 10, the image enhancement module 20, the model training module 30, and the detection module 40 are exemplary:
  • the normalized noise reduction module 10 is used to: obtain a pathological image set, perform a normalization operation on the pathological image set, and perform image noise reduction processing on the pathological image set after the normalization operation to obtain a standard Pathology image set.
  • the image enhancement module 20 is used to perform image enhancement processing on the standard pathological image to obtain an enhanced pathological cell image set.
  • the model training module 30 is configured to: input the enhanced pathological image set into a pre-built pathological detection model, obtain the weight parameters of the enhanced pathological image set through forward propagation in the pathological detection model, and use gradient descent The algorithm updates the weight parameters until the pathology detection model tends to converge, and the pathology detection model after training is obtained.
  • the detection module 40 is configured to: receive the image to be detected input by the user, detect the image to be detected through the pathology detection model after the training is completed, obtain the detection result of the image, and return the detection result to the user.
  • the embodiment of the present application also proposes a computer-readable storage medium.
  • the computer-readable storage medium may be non-volatile or volatile.
  • a medical image recognition and detection program is stored on the computer-readable storage medium, and the medical image recognition and detection program 01 can be executed by one or more processors 12 to implement the following operations:
  • the enhanced pathological image set is input into a pre-built pathological detection model, the weight parameters of the enhanced pathological image set are obtained through forward propagation in the pathological detection model, and the weight parameters are updated by using a gradient descent algorithm until The pathology detection model tends to converge, and the pathology detection model after training is obtained;
  • the image to be detected input by the user is received, the image to be detected is detected by the pathological detection model after the training is completed, the detection result of the image is obtained, and the detection result is returned to the user.

Abstract

本申请涉及一种人工智能技术，揭露了一种医学影像识别检测方法，包括：获取病理图像集，对所述病理图像集进行归一化、图像降噪以及图像增强处理，得到增强病理图像集；将所述增强病理图像集输入至预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，对所述权重参数进行更新直至所述病理检测模型趋于收敛，完成训练；通过训练完成后的所述病理检测模型对待检测的病理图像进行检测，并将检测结果返回给用户。本申请还提出一种医学影像识别检测装置以及一种计算机可读存储介质。本申请实现了医学影像的智能识别及检测。

Description

医学影像识别检测方法、装置、设备及存储介质
本申请要求于2020年01月20日提交中国专利局、申请号为202010069864.4、发明名称为“医学影像识别检测方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及人工智能技术领域，尤其涉及一种医学影像识别检测方法、装置、设备及存储介质。
背景技术
随着深度学习在医学影像领域中的渗透及应用，将AI技术应用于医疗影像分析可以帮助医生定位病灶分析病情，辅助医生精确快速的做出诊断。目前在医疗影像领域，AI应用主要集中在肺结节、眼底、乳腺等，随着AI技术的不断影像及临床需求的日益提高，AI技术在数字病理诊断也得到应用。
发明人意识到目前在临床肾活检病理诊断流程中，病理医生通过光学显微镜观察病理切片图像中肾小球的病理形态、肾小球内细胞增生分布情况以及病理医生的自身经验进行病理分析并给出病理诊断报告。由于切片图像中肾小球个数众多，通过肉眼观察识别肾小球的工作量极大，效率比较低，而且诊断结果容易受到医生主观因素的影响。
发明内容
本申请提供一种医学影像识别检测方法、装置、设备及存储介质，其主要目的在于提供一种智能化的肾小球检测方案。
为实现上述目的，本申请提供的一种医学影像识别检测方法，包括：
获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
对所述标准病理图像进行图像增强处理，得到增强病理图像集；
将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
接收用户输入的待检测病理图像，通过上述训练完成后的病理检测模型对所述待检测病理图像进行检测，并将检测结果返回给所述用户。
可选地，所述归一化操作包括：
x＝(x-μ)/σ
其中，x表示病理图像集中的图像数据，μ为病理图像集中图像数据的均值，σ为病理图像集中图像数据的标准差。
可选地，所述图像降噪处理包括：
Figure PCTCN2020093556-appb-000001
其中，f(x,y)表示降噪后的病理图像，g(x,y)表示待处理的病理图像，
Figure PCTCN2020093556-appb-000002
表示病理图像的噪声方差，
Figure PCTCN2020093556-appb-000003
表示病理图像的像素灰度均值，
Figure PCTCN2020093556-appb-000004
表示病理图像的像素灰度的方差。
可选地，所述通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，包括：
对所述增强病理图像集进行分类分支处理后得到所述增强病理图像集的前景点，通过中性分支算法计算所述前景点的边框界位置，根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数。
可选地，所述根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数，包括：
利用下述公式计算出所述增强病理图像集的权重参数：
Figure PCTCN2020093556-appb-000005
其中，Loss iou表示权重参数，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值，l g、r g、t g、b g分别表示增强病理图像集前景点的边框界上、下、左、右的实际距离值。
此外，为实现上述目的，本申请还提供一种医学影像识别检测装置，所述装置包括：
归一化降噪模块，用于获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
图像增强模块，用于对所述标准病理图像集进行图像增强处理，得到增强病理图像集；
模型训练模块，用于将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
检测模块，用于接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，得到对所述待检测图像的检测结果，并将检测结果返回给所述用户。
此外，为实现上述目的，本申请还提供一种医学影像识别检测设备，该设备包括存储器和处理器，所述存储器中存储有可在所述处理器上运行的医学影像识别检测程序，所述医学影像识别检测程序被所述处理器执行时实现如下步骤：
获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
对所述标准病理图像进行图像增强处理，得到增强病理图像集；
将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
接收用户输入的待检测病理图像，通过上述训练完成后的病理检测模型对所述待检测病理图像进行检测，并将检测结果返回给所述用户。
可选地，所述归一化操作包括：
x＝(x-μ)/σ
其中，x表示病理图像集中的图像数据，μ为病理图像集中图像数据的均值，σ为病理图像集中图像数据的标准差。
可选地，所述图像降噪处理包括：
Figure PCTCN2020093556-appb-000006
其中，f(x,y)表示降噪后的病理图像，g(x,y)表示待处理的病理图像，
Figure PCTCN2020093556-appb-000007
表示病理图像的噪声方差，
Figure PCTCN2020093556-appb-000008
表示病理图像的像素灰度均值，
Figure PCTCN2020093556-appb-000009
表示病理图像的像素灰度的方差。
可选地，所述通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，包括：
对所述增强病理图像集进行分类分支处理后得到所述增强病理图像集的前景点，通过中性分支算法计算所述前景点的边框界位置，根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数。
此外，为实现上述目的，本申请还提供一种计算机可读存储介质，所述计算机可读存储介质上存储有医学影像识别检测程序，所述医学影像识别检测程序可被一个或者多个处理器执行，以实现如上所述的医学影像识别检测方法的步骤。
本申请提出的医学影像识别检测方法、装置、设备及存储介质，当用户进行肾小球检测时，获取病理图像集，对所述病理图像集进行归一化、图像降噪以及图像增强处理，可以得到增强的病理细胞图像集，利用所述增强病理细胞图像集对预先构建的病理检测模型，得到训练完成的病理检测模型，根据训练完成后的所述病理检测模型对待检测的病理图像进行检测，并将检测结果返回给用户，从而可以实现从医学影像中智能识别出肾小球等疾病的检测结果。
附图说明
图1为本申请一实施例提供的医学影像识别检测方法的流程示意图；
图2为本申请一实施例提供的医学影像识别检测设备的内部结构示意图；
图3为本申请一实施例提供的医学影像识别检测装置中医学影像识别检测装置的模块示意图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供一种医学影像识别检测方法。参照图1所示，为本申请一实施例提供的医学影像识别检测方法的流程示意图。该方法可以由一个装置执行，该装置可以由软件和/或硬件实现。
在本实施例中，所述医学影像识别检测方法包括：
S1、获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集。
本申请较佳实施例中，所述病理图像集可以是，例如肾小球内存在细胞增生的病理图像。本申请较佳实施例中所述获取病理图像集包括：取体积为预设尺寸的肾脏组织，并放入液态固定剂对所述肾脏组织进行固定，利用不同种浓度乙醇对固定后的所述肾脏组织进行脱水处理，通过巴氏染色法对脱水处理后的所述肾脏组织进行染色，根据病理切片扫描仪对染色后的所述肾脏组织按预设的规格进行切片，从而得到所述病理图像集。其中，本申请中，所述预设尺寸为2.0cm*2.0cm*0.3cm的肾脏组织，所述脱水处理通过脱水机自控完成，所述预设规格为1024*1024的病理图像。
进一步地，所述数据的归一化是将数据按比例缩放，使之落入一个小的特定区间。在本申请实施例中，本申请应用的归一化处理方法是z-score归一化方法，其核心思想如下所示：
x＝(x-μ)/σ
其中，x表示病理图像集中的图像数据，μ为病理图像集中图像数据的均值，σ为病理图像集中图像数据的标准差。
进一步地，由于图像降噪处理的效果直接影响到后续的处理步骤以及最终的识别结果，在本申请实施例中通过滤波算法对归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集。其中，所述滤波算法包括：
Figure PCTCN2020093556-appb-000010
其中，f(x,y)表示降噪后的病理图像，g(x,y)表示待处理的病理图像，
Figure PCTCN2020093556-appb-000011
表示病理图像的噪声方差，
Figure PCTCN2020093556-appb-000012
表示病理图像的像素灰度均值，
Figure PCTCN2020093556-appb-000013
表示病理图像的像素灰度的方差。
S2、对所述标准病理图像进行图像增强处理，得到增强病理图像集。
由于过度的数据增强会破坏原来数据的分布，影响网络训练效果，因此本申请实施例利用数据增强技术对所述标准病理图像集进行图像增强处理。其中，本申请所述数据增强技术包括随机翻转和随机剪裁。
详细地，本申请利用python库中的cv2指令对所述标准病理图像集设置相应翻转参数，使所述标准病理图像集各有1/3概率发生水平翻转、垂直翻转、水平垂直翻转，其中cv.flip(img,1)代表水平方向旋转180度，cv.flip(img,0)代表垂直方向旋转180度，cv.flip(img,-1)代表水平和垂直方向同时旋转180度。
进一步地，本申请使用randomCrop(image)函数对随机翻转后的所述病理图像集进行随意裁剪，使得裁剪后的所述病理图像大小为原始图像宽和高的2/3，据此本申请实现对病理细胞图像的数据增强，得到所述增强病理细胞图像集。
S3、将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型。
本申请较佳实施例中，所述预先构建的肾脏比例肾小球检测模型基于Mask R-CNN框架实现，所述Mask R-CNN是一种检测及实例分割模型框架。所述通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数包括：对所述增强病理图像集进行分类分支处理后得到所述增强病理图像集的前景点，通过中性分支算法计算所述前景点的边框界位置，根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数。
所述分类分支算法包括：
Figure PCTCN2020093556-appb-000014
其中，Centerness表示增强病理图像集前景点的边框界位置，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值。
所述权重参数计算公式为：
Figure PCTCN2020093556-appb-000015
其中，Loss iou表示权重参数，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值，l g、r g、t g、b g分别表示增强病理图像集前景点的边框界上、下、左、右的实际距离值。
进一步地，所述梯度下降算法是神经网络模型训练中的优化算法，为找到所述预先构建 的病理检测模型趋于收敛的权重参数，需要沿着与所述预先构建的病理检测模型的梯度向量相反的方向更新变量，这样可以使得梯度减少最快，直至所述预先构建的病理检测模型进行收敛，得到所述训练完成后的病理检测模型。
S4、接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，得到所述图像的检测结果，并将检测结果返回给所述用户。
本申请较佳实施例中接收用户输入的待检测图像，根据上述训练完成后的所述病理检测模型对所述用户输入的待检测图像进行检测，得到所述图像的检测结果，并将检测结果返回给所述用户。其中，所述检测结果包括正常nos、硬化gs两种。
本申请还提供一种医学影像识别检测设备。参照图2所示，为本申请一实施例提供的医学影像识别检测设备1的内部结构示意图。
在本实施例中，所述医学影像识别检测设备1可以是PC(Personal Computer，个人电脑)，或者是智能手机、平板电脑、便携计算机等终端设备，也可以是一种服务器等。该医学影像识别检测设备1至少包括存储器11、处理器12，通信总线13，以及网络接口14。
其中，存储器11至少包括一种类型的可读存储介质，所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、磁性存储器、磁盘、光盘等。存储器11在一些实施例中可以是医学影像识别检测设备1的内部存储单元，例如该医学影像识别检测设备1的硬盘。存储器11在另一些实施例中也可以是医学影像识别检测设备1的外部存储设备，例如医学影像识别检测设备1上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。进一步地，存储器11还可以既包括医学影像识别检测设备1的内部存储单元也包括外部存储设备。存储器11不仅可以用于存储安装于医学影像识别检测设备1的应用软件及各类数据，例如医学影像识别检测程序01的代码等，还可以用于暂时地存储已经输出或者将要输出的数据。
处理器12在一些实施例中可以是一中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器或其他数据处理芯片，用于运行存储器11中存储的程序代码或处理数据，例如执行医学影像识别检测程序01等。
通信总线13用于实现这些组件之间的连接通信。
网络接口14可选的可以包括标准的有线接口、无线接口(如WI-FI接口)，通常用于在该设备1与其他电子设备之间建立通信连接。
可选地，该设备1还可以包括用户接口，用户接口可以包括显示器(Display)、输入单元比如键盘(Keyboard)，可选的用户接口还可以包括标准的有线接口、无线接口。可选地，在一些实施例中，显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(Organic Light-Emitting Diode，有机发光二极管)触摸器等。其中，显示器也可以适当的称为显示屏或显示单元，用于显示在医学影像识别检测设备1中处理的信息以及用于显示可视化的用户界面。
图2仅示出了具有组件11-14以及医学影像识别检测程序01的医学影像识别检测设备1，本领域技术人员可以理解的是，图1示出的结构并不构成对医学影像识别检测设备1的限定，可以包括比图示更少或者更多的部件，或者组合某些部件，或者不同的部件布置。
在图2所示的设备1实施例中，存储器11中存储有医学影像识别检测程序01；处理器12执行存储器11中存储的医学影像识别检测程序01时实现如下步骤：
步骤一、获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集
本申请较佳实施例中所述获取病理图像集包括：取体积为预设尺寸的肾脏组织，并放入液态固定剂对所述肾脏组织进行固定，利用不同种浓度乙醇对固定后的所述肾脏组织进行脱水处理，通过巴氏染色法对脱水处理后的所述肾脏组织进行染色，根据病理切片扫描仪对染色后 的所述肾脏组织按预设的规格进行切片，从而得到所述病理图像集。其中，本申请中，所述预设尺寸为2.0cm*2.0cm*0.3cm的肾脏组织，所述脱水处理通过脱水机自控完成，所述预设规格为1024*1024的病理图像。
进一步地，所述数据的归一化是将数据按比例缩放，使之落入一个小的特定区间。在本申请实施例中，本申请应用的归一化处理方法是z-score归一化方法，其核心思想如下所示：
x＝(x-μ)/σ
其中，x表示病理图像集中的图像数据，μ为病理图像集中图像数据的均值，σ为病理图像集中图像数据的标准差。
进一步地，由于图像降噪处理的效果直接影响到后续的处理步骤以及最终的识别结果，在本申请实施例中通过滤波算法对归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集。其中，所述滤波算法包括：
Figure PCTCN2020093556-appb-000016
其中，f(x,y)表示降噪后的病理图像，g(x,y)表示待处理的病理图像，
Figure PCTCN2020093556-appb-000017
表示病理图像的噪声方差，
Figure PCTCN2020093556-appb-000018
表示病理图像的像素灰度均值，
Figure PCTCN2020093556-appb-000019
表示病理图像的像素灰度的方差。
步骤二、对所述标准病理图像进行图像增强处理，得到增强病理图像集。
由于过度的数据增强会破坏原来数据的分布，影响网络训练效果，因此本申请实施例利用数据增强技术对所述标准病理图像集进行图像增强处理。其中，本申请所述数据增强技术包括随机翻转和随机剪裁。
详细地，本申请利用python库中的cv2指令对所述标准病理图像集设置相应翻转参数，使所述标准病理图像集各有1/3概率发生水平翻转、垂直翻转、水平垂直翻转，其中cv.flip(img,1)代表水平方向旋转180度，cv.flip(img,0)代表垂直方向旋转180度，cv.flip(img,-1)代表水平和垂直方向同时旋转180度。
进一步地，本申请使用randomCrop(image)函数对随机翻转后的所述病理图像集进行随意裁剪，使得裁剪后的所述病理图像大小为原始图像宽和高的2/3，据此本申请实现对病理细胞图像的数据增强，得到所述增强病理细胞图像集。
步骤三、将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型。
本申请较佳实施例中，所述预先构建的肾脏比例肾小球检测模型基于Mask R-CNN框架实现，所述Mask R-CNN是一种检测及实例分割模型框架。所述通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数包括：对所述增强病理图像集进行分类分支处理后得到所述增强病理图像集的前景点，通过中性分支算法计算所述前景点的边框界位置，根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数。
所述分类分支算法包括：
Figure PCTCN2020093556-appb-000020
其中，Centerness表示增强病理图像集前景点的边框界位置，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值。
所述权重参数计算公式为：
Figure PCTCN2020093556-appb-000021
其中，Loss iou表示权重参数，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值，l g、r g、t g、b g分别表示增强病理图像集前景点的边框界上、下、左、右的实际距离值。
进一步地，所述梯度下降算法是神经网络模型训练最常用的优化算法，为找到所述预先构建的病理检测模型趋于收敛的权重参数，需要沿着与所述预先构建的病理检测模型的梯度向量相反的方向更新变量，这样可以使得梯度减少最快，直至所述预先构建的病理检测模型进行收敛，得到所述训练完成后的病理检测模型。
步骤四、接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，并将检测结果返回给所述用户。
本申请较佳实施例中接收用户输入的待检测图像，根据上述训练完成后的所述病理检测模型对所述用户输入的待检测图像进行检测，得到所述图像的检测结果，并将检测结果返回给所述用户。其中，所述检测结果包括正常nos、硬化gs两种。
可选地，在其他实施例中，医学影像识别检测程序100还可以被分割为一个或者多个模块，一个或者多个模块被存储于存储器11中，并由一个或多个处理器(本实施例为处理器12)所执行以完成本申请，本申请所称的模块是指能够完成特定功能的一系列计算机程序指令段，用于描述医学影像识别检测程序01在医学影像识别检测装置100中的执行过程。
例如，参照图3所示，为本申请医学影像识别检测装置一实施例中的医学影像识别检测程序的装置100框架图，该实施例中，所述医学影像识别检测装置100可以被分割为归一化降噪模块10、图像增强模块20、模型训练模块30以及检测模块40，示例性地：
所述归一化降噪模块10用于：获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集。
所述图像增强模块20用于：对所述标准病理图像进行图像增强处理，得到增强病理细胞图像集。
所述模型训练模块30用于：将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型。
所述检测模块40用于：接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，得到所述图像的检测结果，并将检测结果返回给所述用户。
上述归一化降噪模块10、图像增强模块20、模型训练模块30以及检测模块40等程序模块被执行时所实现的功能或操作步骤与上述实施例大体相同，在此不再赘述。
此外，本申请实施例还提出一种计算机可读存储介质，所述计算机可读存储介质可以是非易失性，也可以是易失性。所述计算机可读存储介质上存储有医学影像识别检测程序，所述医学影像识别检测程序01可被一个或多个处理器12执行，以实现如下操作：
获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
对所述标准病理图像进行图像增强处理，得到增强病理图像集；
将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，得到所述图像的检测结果，并将检测结果返回给所述用户。
本申请计算机可读存储介质具体实施方式与上述医学影像识别检测装置和方法各实施例基本相同，在此不作累述。
需要说明的是，上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。并且本文中的术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、装置、物品或者方法不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、装置、物品或者方法所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、装置、物品或者方法中还存在另外的相同要素。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在如上所述的一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本申请各个实施例所述的方法。
以上仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种医学影像识别检测方法，其中，所述方法包括：
    获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
    对所述标准病理图像集进行图像增强处理，得到增强病理图像集；
    将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
    接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，得到对所述待检测图像的检测结果，并将检测结果返回给所述用户。
  2. 如权利要求1所述的医学影像识别检测方法，其中，所述归一化操作包括：
    x＝(x-μ)/σ
    其中，x表示病理图像集中的图像数据，μ为病理图像集中图像数据的均值，σ为病理图像集中图像数据的标准差。
  3. 如权利要求1所述的医学影像识别检测方法，其中，所述图像降噪处理包括：
    Figure PCTCN2020093556-appb-100001
    其中，f(x,y)表示降噪后的病理图像，g(x,y)表示待处理的病理图像，
    Figure PCTCN2020093556-appb-100002
    表示病理图像的噪声方差，
    Figure PCTCN2020093556-appb-100003
    表示病理图像的像素灰度均值，
    Figure PCTCN2020093556-appb-100004
    表示病理图像的像素灰度的方差。
  4. 如权利要求1所述的医学影像识别检测方法，其中，所述通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，包括：
    对所述增强病理图像集进行分类分支处理后得到所述增强病理图像集的前景点，通过中性分支算法计算所述前景点的边框界位置，根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数。
  5. 如权利要求4中所述的医学影像识别检测方法，其中，所述根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数包括：
    利用下述公式计算出所述增强病理图像集的权重参数：
    Figure PCTCN2020093556-appb-100005
    其中，Loss ipu表示权重参数，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值，l g、r g、t g、b g分别表示增强病理图像集前景点的边框界上、下、左、右的实际距离值。
  6. 如权利要求4中所述的医学影像识别检测方法，其中，所述分类分支算法包括：
    Figure PCTCN2020093556-appb-100006
    其中，Centerness表示增强病理图像集前景点的边框界位置，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值。
  7. 如权利要求1中所述的医学影像识别检测方法，其中，所述病理检测模型基于Mask R-CNN框架实现。
  8. 一种医学影像识别检测装置，其中，所述装置包括：
    归一化降噪模块，用于获取病理图像集，对所述病理图像集进行归一化操作，并将归一化 操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
    图像增强模块，用于对所述标准病理图像进行图像增强处理，得到增强病理图像集；
    模型训练模块，用于将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
    检测模块，用于接收用户输入的待检测病理图像，通过上述训练完成后的病理检测模型对所述待检测病理图像进行检测，并将检测结果返回给所述用户。
  9. 一种医学影像识别检测设备，其中，所述设备包括存储器和处理器，所述存储器上存储有可在所述处理器上运行的医学影像识别检测程序，所述医学影像识别检测程序被所述处理器执行时实现如下步骤：
    获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
    对所述标准病理图像集进行图像增强处理，得到增强病理图像集；
    将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
    接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，得到对所述待检测图像的检测结果，并将检测结果返回给所述用户。
  10. 如权利要求9所述的医学影像识别检测设备，其中，所述归一化操作包括：
    x＝(x-μ)/σ
    其中，x表示病理图像集中的图像数据，μ为病理图像集中图像数据的均值，σ为病理图像集中图像数据的标准差。
  11. 如权利要求9所述的医学影像识别检测设备，其中，所述图像降噪处理包括：
    Figure PCTCN2020093556-appb-100007
    其中，f(x,y)表示降噪后的病理图像，g(x,y)表示待处理的病理图像，
    Figure PCTCN2020093556-appb-100008
    表示病理图像的噪声方差，
    Figure PCTCN2020093556-appb-100009
    表示病理图像的像素灰度均值，
    Figure PCTCN2020093556-appb-100010
    表示病理图像的像素灰度的方差。
  12. 如权利要求9所述的医学影像识别检测设备，其中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，包括：
    对所述增强病理图像集进行分类分支处理后得到所述增强病理图像集的前景点，通过中性分支算法计算所述前景点的边框界位置，根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数。
  13. 如权利要求11中所述的医学影像识别检测设备，其中，所述根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数包括：
    利用下述公式计算出所述增强病理图像集的权重参数：
    Figure PCTCN2020093556-appb-100011
    其中，Loss iou表示权重参数，l p、r g、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值，l g、r g、t g、b g分别表示增强病理图像集前景点的边框界上、下、左、右的实际距离值。
  14. 如权利要求11中所述的医学影像识别检测设备，其中，所述分类分支算法包括：
    Figure PCTCN2020093556-appb-100012
    其中，Centerness表示增强病理图像集前景点的边框界位置，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值。
  15. 一种计算机可读存储介质，其中，所述计算机可读存储介质上存储有医学影像识别检测程序，所述医学影像识别检测程序可被一个或者多个处理器执行，以实现如权利要求1至7中任一项所述的医学影像识别检测方法的步骤：
    获取病理图像集，对所述病理图像集进行归一化操作，并将归一化操作后的所述病理图像集进行图像降噪处理，得到标准病理图像集；
    对所述标准病理图像集进行图像增强处理，得到增强病理图像集；
    将所述增强病理图像集输入预先构建的病理检测模型中，通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，利用梯度下降算法对所述权重参数进行更新直至所述病理检测模型趋于收敛，得到训练完成后的病理检测模型；
    接收用户输入的待检测图像，通过上述训练完成后的病理检测模型对所述待检测图像进行检测，得到对所述待检测图像的检测结果，并将检测结果返回给所述用户。
  16. 如权利要求15所述的计算机可读存储介质，其中，所述归一化操作包括：
    x＝(x-μ)/σ
    其中，x表示病理图像集中的图像数据，μ为病理图像集中图像数据的均值，σ为病理图像集中图像数据的标准差。
  17. 如权利要求15所述的计算机可读存储介质，其中，所述图像降噪处理包括：
    Figure PCTCN2020093556-appb-100013
    其中，f(x,y)表示降噪后的病理图像，g(x,y)表示待处理的病理图像，
    Figure PCTCN2020093556-appb-100014
    表示病理图像的噪声方差，
    Figure PCTCN2020093556-appb-100015
    表示病理图像的像素灰度均值，
    Figure PCTCN2020093556-appb-100016
    表示病理图像的像素灰度的方差。
  18. 如权利要求15所述的计算机可读存储介质，其中，所述通过所述病理检测模型中的正向传播得到所述增强病理图像集的权重参数，包括：
    对所述增强病理图像集进行分类分支处理后得到所述增强病理图像集的前景点，通过中性分支算法计算所述前景点的边框界位置，根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数。
  19. 如权利要求18中所述的计算机可读存储介质，其中，所述根据所述边框界位置利用预先确定的权重参数计算公式计算出所述增强病理图像集的权重参数包括：
    利用下述公式计算出所述增强病理图像集的权重参数：
    Figure PCTCN2020093556-appb-100017
    其中，Loss iou表示权重参数，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值，l g、r g、t g、b g分别表示增强病理图像集前景点的边框界上、下、左、右的实际距离值。
  20. 如权利要求18中所述的计算机可读存储介质，其中，所述分类分支算法包括：
    Figure PCTCN2020093556-appb-100018
    其中，Centerness表示增强病理图像集前景点的边框界位置，l p、r p、t p、b p分别表示增强病理图像集前景点的边框界上、下、左、右的预测距离值。
PCT/CN2020/093556 2020-01-20 2020-05-30 医学影像识别检测方法、装置、设备及存储介质 WO2021147218A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010069864.4A CN111275689A (zh) 2020-01-20 2020-01-20 医学影像识别检测方法、装置及计算机可读存储介质
CN202010069864.4 2020-01-20

Publications (1)

Publication Number Publication Date
WO2021147218A1 true WO2021147218A1 (zh) 2021-07-29

Family

ID=71001959

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/093556 WO2021147218A1 (zh) 2020-01-20 2020-05-30 医学影像识别检测方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN111275689A (zh)
WO (1) WO2021147218A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115760858B (zh) * 2023-01-10 2023-05-02 东南大学附属中大医院 基于深度学习的肾脏病理切片细胞识别方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110158474A1 (en) * 2009-12-31 2011-06-30 Indian Institute Of Technology Bombay Image object tracking and segmentation using active contours
US20130034289A1 (en) * 2011-08-02 2013-02-07 Siemens Corporation Filter approach to catheter electrode detection and tracking
CN109671062A (zh) * 2018-12-11 2019-04-23 成都智能迭迦科技合伙企业(有限合伙) 超声图像检测方法、装置、电子设备及可读存储介质
CN109886992A (zh) * 2017-12-06 2019-06-14 深圳博脑医疗科技有限公司 用于分割mri图像中异常信号区的全卷积网络模型训练方法

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109670510B (zh) * 2018-12-21 2023-05-26 万达信息股份有限公司 一种基于深度学习的胃镜活检病理数据筛查系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110158474A1 (en) * 2009-12-31 2011-06-30 Indian Institute Of Technology Bombay Image object tracking and segmentation using active contours
US20130034289A1 (en) * 2011-08-02 2013-02-07 Siemens Corporation Filter approach to catheter electrode detection and tracking
CN109886992A (zh) * 2017-12-06 2019-06-14 深圳博脑医疗科技有限公司 用于分割mri图像中异常信号区的全卷积网络模型训练方法
CN109671062A (zh) * 2018-12-11 2019-04-23 成都智能迭迦科技合伙企业(有限合伙) 超声图像检测方法、装置、电子设备及可读存储介质

Also Published As

Publication number Publication date
CN111275689A (zh) 2020-06-12

Similar Documents

Publication Publication Date Title
CN107895367B (zh) 一种骨龄识别方法、系统及电子设备
WO2019109526A1 (zh) 人脸图像的年龄识别方法、装置及存储介质
JP2021530061A (ja) 画像処理方法及びその装置、電子機器並びにコンピュータ可読記憶媒体
WO2021082691A1 (zh) 眼部oct图像病灶区域的分割方法、装置及终端设备
CN110598782B (zh) 一种训练用于医学影像的分类网络的方法与装置
WO2022088581A1 (zh) 图像检测模型的训练方法及相关装置、设备、存储介质
CN109117857B (zh) 一种生物属性的识别方法、装置及设备
WO2020140370A1 (zh) 眼底出血点的自动检测方法、装置及计算机可读存储介质
WO2018107371A1 (zh) 图像搜索系统及方法
WO2020253508A1 (zh) 异常细胞检测方法、装置及计算机可读存储介质
US11501431B2 (en) Image processing method and apparatus and neural network model training method
WO2021136368A1 (zh) 钼靶图像中胸大肌区域自动检测方法及装置
US11967181B2 (en) Method and device for retinal image recognition, electronic equipment, and storage medium
WO2020151149A1 (zh) 微动脉瘤自动检测方法、装置及计算机可读存储介质
Cao et al. Efficient saliency detection using convolutional neural networks with feature selection
EP4181059A1 (en) Medical image processing method, apparatus, device, storage medium, and product
CN110232318A (zh) 穴位识别方法、装置、电子设备及存储介质
Mei et al. A curve evolution approach for unsupervised segmentation of images with low depth of field
CN114092450A (zh) 一种基于胃镜检查视频的实时图像分割方法、系统、装置
CN114332132A (zh) 图像分割方法、装置和计算机设备
WO2021147218A1 (zh) 医学影像识别检测方法、装置、设备及存储介质
CN109685830A (zh) 目标跟踪方法、装置和设备及计算机存储介质
CN113240661A (zh) 基于深度学习的腰椎骨分析方法、装置、设备及存储介质
CN111080592A (zh) 一种基于深度学习的肋骨提取方法及装置
CN117274278B (zh) 基于模拟感受野的视网膜图像病灶部位分割方法及系统

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: 20915334

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: 20915334

Country of ref document: EP

Kind code of ref document: A1