WO2020253048A1 - 基于深度学习的影像识别方法、装置、设备及存储介质 - Google Patents

基于深度学习的影像识别方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2020253048A1
WO2020253048A1 PCT/CN2019/117573 CN2019117573W WO2020253048A1 WO 2020253048 A1 WO2020253048 A1 WO 2020253048A1 CN 2019117573 W CN2019117573 W CN 2019117573W WO 2020253048 A1 WO2020253048 A1 WO 2020253048A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
image
preset
processing
recognized
Prior art date
Application number
PCT/CN2019/117573
Other languages
English (en)
French (fr)
Inventor
吴海萍
吕传峰
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 WO2020253048A1 publication Critical patent/WO2020253048A1/zh

Links

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
    • 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
    • G06T2207/30008—Bone
    • G06T2207/30012—Spine; Backbone
    • 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
    • G06T2207/30056—Liver; Hepatic
    • 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
    • G06T2207/30061—Lung

Definitions

  • This application relates to the technical field of medical image processing, and in particular to an image recognition method, device, equipment, and storage medium based on deep learning.
  • the main purpose of this application is to provide an image recognition method, device, equipment and storage medium based on deep learning, aiming to solve the technical problems of low recognition accuracy and low recognition efficiency of existing clinical images.
  • the image recognition method based on deep learning includes:
  • the medical image to be recognized of the target type is distributed to the target recognition corresponding to the target type In the network model;
  • the medical image to be recognized is subjected to a preset target sign location and corresponding preset target sign category determination processing, and is configured to obtain a processing result and output it.
  • the present application also provides an image recognition device based on deep learning.
  • the image recognition device based on deep learning includes:
  • the first determining module is configured to determine the target type of the medical image to be recognized when the medical image to be recognized is detected;
  • the distribution module is configured to distribute the medical image to be recognized of the target type to the target according to the target type of the medical image to be recognized and the preset association relationship between the type of medical image to be recognized and each recognition network model Target recognition network model corresponding to the type;
  • the processing module is configured to determine the preset target sign position and the corresponding preset target sign category on the medical image to be recognized based on the target recognition network model, and is configured to obtain and output a processing result.
  • the processing module includes:
  • a preprocessing unit configured to preprocess the medical image to be recognized, and configured to obtain a preprocessed image
  • a layering unit configured to perform layered processing on the pre-processed image according to the target recognition network model, and determine the position of the target layered sign corresponding to the medical image to be recognized;
  • the output unit is configured to perform preset target sign category recognition processing on the layered preprocessed images, and output the target sign position and the corresponding preset target sign category as processing results.
  • the image recognition device based on deep learning further includes:
  • the acquisition module is set to acquire preset use cases of various types of images, the use cases corresponding to the preset ratio of the same type of images are selected as the first use case, and other use cases corresponding to the first training exception for the same type of images are set as the second use case ;
  • the training module is configured to use the first use case as a training use case to perform adjustment training of the target recognition network model corresponding to the basic recognition network model, so as to adjust and train the multiple layers of the basic recognition network model for the image
  • the test module is configured to test the basic recognition network model after adjusting and training the second use case as a test case, and is configured to obtain the target recognition type.
  • the output unit includes:
  • An alternate processing subunit configured to perform a preset number of alternate processing of convolution and pooling on the layered preprocessed image to obtain an initial processing result
  • the classification subunit is configured to classify the initial processing result according to a preset classification preset, and is configured to obtain the preset target sign category of the preprocessed image after the layering, and layer the target The sign position and the corresponding preset target sign category are used as the processing result and output.
  • the alternate processing subunit is configured to implement:
  • the pooling processing result is subjected to a corresponding number of convolution and pooling alternate processing again, and the initial processing result is configured to be obtained.
  • the alternate processing subunit is further configured to implement:
  • the image recognition device based on deep learning further includes:
  • the second determining module is configured to determine the email address of the target person receiving the image of the processing result according to the preset target sign category;
  • the sending module is configured to send the processing result to the target image receiving person according to the email address.
  • the present application also provides an image recognition device based on deep learning.
  • the image recognition device based on deep learning includes: a memory, a processor, a communication bus, and a computer readable storage on the memory. instruction,
  • the communication bus is configured to realize a communication connection between the processor and the memory
  • the processor is configured to execute the computer-readable instructions to implement the following steps:
  • the medical image to be recognized of the target type is distributed to the target recognition corresponding to the target type In the network model;
  • the medical image to be recognized is subjected to a preset target sign location and corresponding preset target sign category determination processing, and is configured to obtain a processing result and output it.
  • this application also provides a storage medium that stores one or more computer-readable instructions, and the one or more computer-readable instructions can be used by one or more processors. Execute to set as:
  • the medical image to be recognized of the target type is distributed to the target recognition corresponding to the target type In the network model;
  • the medical image to be recognized is subjected to a preset target sign position and corresponding preset target sign category determination processing, and is configured to obtain a processing result and output it.
  • This application determines the target type of the medical image to be recognized when the medical image to be recognized is detected; according to the target type of the medical image to be recognized, and the preset type of the medical image to be recognized and the identification network model Association relationship, distributing the medical image to be recognized of the target type to the target recognition network model corresponding to the target type; based on the target recognition network model to preset the target sign position and corresponding to the medical image to be recognized
  • the preset processing for determining the target sign category is configured to obtain and output the processing result. In this embodiment, it is no longer necessary to send clinical images to different imaging doctors for artificial recognition. Instead, when a medical image to be recognized is detected, for the different types of medical images to be recognized, the target type is to be recognized.
  • the recognized medical image is distributed to the target recognition network model corresponding to the target type, and the target recognition network model is a model that can accurately recognize the location of the sign and the sign category of the image after training. Therefore, in this embodiment The automatic recognition of the medical image to be recognized is realized, thus, the recognition accuracy of the clinical image is improved, and the recognition efficiency of the existing clinical image is reduced. The technical problems of low recognition accuracy and low recognition efficiency of existing clinical images are solved.
  • FIG. 1 is a schematic flowchart of a first embodiment of an image recognition method based on deep learning in this application;
  • Figure 2 is a process of determining the preset target sign position and the corresponding preset target sign category on the medical image to be recognized based on the target recognition network model in the image recognition method based on deep learning in this application, and is configured to be processed Schematic diagram of detailed flow of results and output steps;
  • FIG. 3 is a schematic diagram of the device structure of the hardware operating environment involved in the method of the embodiment of the present application.
  • the image recognition method based on deep learning includes:
  • Step S10 when the medical image to be recognized is detected, the target type of the medical image to be recognized is determined
  • the deep learning-based image recognition method has recognition functions for many types of images, and many types of image recognition functions refer to the detection of various types of preset images.
  • the many types of images include lung images, lumbar intervertebral disc images, liver images and other types.
  • the corresponding recognition can be performed, and it needs to be based on deep learning images.
  • the deep learning-based image recognition device corresponding to the recognition method is provided with a corresponding type of recognition network model.
  • the recognition network model includes, but is not limited to, convolutional neural network models.
  • a recognition network model of lung images must be set in the image recognition device, and corresponding recognition must be performed when lumbar disc images are detected, and a lumbar intervertebral disc must be set in the image recognition device Image recognition network model, etc.
  • determining the target type of the medical image to be recognized when the medical image to be recognized is detected, determining the target type of the medical image to be recognized, specifically, when the medical image to be recognized is detected, the method of determining the target type of the medical image to be recognized includes :
  • Manner 1 Determine the target type of the medical image to be recognized according to the image identifier pre-carried in the medical image to be recognized.
  • Method 2 Obtain the source channel of the medical image to be identified, and determine the target type of the medical image to be identified according to the source channel, where the source channel refers to the source department or the source machine.
  • the purpose of determining the target type of the medical image to be recognized is to distribute the medical image to be recognized to the target recognition network model corresponding to the target type.
  • Step S20 According to the target type of the medical image to be recognized, and the preset association relationship between the type of medical image to be recognized and each recognition network model, the medical image to be recognized of the target type is distributed to the corresponding target type Target recognition network model;
  • the target type of the medical image to be recognized is obtained, according to the target type of the medical image to be recognized, and the preset association relationship between the type of medical image to be recognized and each recognition network model, the target type of the medical image to be recognized
  • the image is distributed to the target recognition network model corresponding to the target type.
  • the target type of the medical image to be recognized is a lung image
  • the medical image to be recognized is distributed to the first target recognition network model corresponding to the lung image type
  • the target type of the medical image to be recognized is waist Intervertebral disc image
  • distributing the medical image to be identified to the second target recognition network model corresponding to the type of lumbar intervertebral disc image is to perform targeted analysis and processing on the medical image to be recognized.
  • step S30 the medical image to be identified is determined based on the target recognition network model on the preset target sign position and the corresponding preset target sign category, and is configured to obtain and output the processing result.
  • the medical image to be recognized is subjected to a preset target sign position and corresponding preset target sign category determination processing, and is configured to obtain a processing result and output it, specifically,
  • the process of determining the preset target sign position and the corresponding preset target sign category of the medical image to be recognized includes a first determining operation of determining the preset target sign position, and a pre-determining operation.
  • Set the second determination operation for determining the target sign category is set the second determination operation for determining the target sign category.
  • the first determination operation and the second determination operation can be determined simultaneously (the target recognition network model synchronously determines the location of the image and the sign category corresponding to the location).
  • the first determination operation and the second determination The operation can also be determined successively, which means that the target recognition network model first determines the preset target sign position, and then determines the corresponding preset target sign category based on the preset target sign position.
  • the lumbar intervertebral disc includes partition 1 (upper half of waist 1), partition 2 (lower half of waist 1), and partition 3 (upper half of waist 2). ) And zone 4 (the lower half of waist 2), each zone also includes each sub zone, the preset target sign position refers to which specific zone or sub zone the medical image to be recognized belongs to, and the preset target sign
  • the categories include the first (fracture) sign category, the second (lumbar disc bulging) sign category, and the third (lumbar disc herniation sign) category.
  • the step of determining the preset target sign position and corresponding preset target sign category on the medical image to be recognized based on the target recognition network model, and the step of configuring to obtain the processing result and output includes:
  • Step S31 preprocessing the medical image to be recognized and configured to obtain a preprocessed image
  • the medical image to be recognized is preprocessed, and configured to obtain the preprocessed image.
  • the specific preprocessing process includes first performing preset contrast stretching on the image, and preset image size.
  • the change of the preset image size includes operations such as preset image size scaling and expansion.
  • the purpose of preprocessing the medical image to be recognized is to ensure subsequent cutting or image processing. The regularity.
  • Step S32 Perform hierarchical processing on the preprocessed image according to the target recognition network model, and determine the target hierarchical sign position corresponding to the medical image to be recognized;
  • the preprocessed image is processed hierarchically according to the target recognition network model to determine the target hierarchical sign position corresponding to the medical image to be recognized.
  • the hierarchical process can be used
  • the adaptive threshold maximum between-class variance method where the adaptive threshold maximum between-class variance method, is to divide the image to be processed into two types of image and background according to gray-scale features to extract feature information and configure to obtain the tissue area, After the tissue area is obtained, the entire tissue area is hierarchically recognized to determine the target hierarchical sign location corresponding to the medical image to be recognized.
  • Step S33 Perform preset target sign category recognition processing on the layered preprocessed images, and output the target sign position and the corresponding preset target sign category as processing results.
  • the pre-processed images after the stratification are respectively subjected to preset target sign category recognition processing, and the target stratified sign position and the corresponding preset target sign category are taken as the processing result and output. It is explained that, in the process of performing preset target sign category recognition processing on the pre-processed images after layering, the convolution, pooling, activation, and classification processing processes of the image are involved to finally obtain the processing result.
  • the step of determining the preset target sign position and the corresponding preset target sign category on the medical image to be recognized based on the target recognition network model, and the configuration to obtain the processing result and output before the step includes:
  • Step A1 Obtain preset use cases for various types of images, set the use cases corresponding to the preset ratio of the same type of images as the first use case, and set other use cases corresponding to the same type of image with the first training exception as the second use case;
  • use cases for various types of images are pre-stored, for example, use cases for lung images, lumbar disc images, liver images, etc., are pre-stored to obtain preset use cases for various types of images.
  • the use case corresponding to selecting the preset ratio of the same type of image is set as the first use case, and the other use cases corresponding to the first training exception for the same type of images are set as the second use case.
  • Step A2 Use the first use case as a training use case to perform adjustment training of the target recognition network model corresponding to the basic recognition network model, so as to adjust the positions of multiple hierarchical features for the image in the training of the basic recognition network model
  • the first use case as a training use case to perform adjustment training of the target recognition network model corresponding to the basic recognition network model, so as to adjust training the basic recognition network model for multiple hierarchical feature positions of the image
  • the sign category, the image weight matrix of the corresponding sign feature For example, a lung image includes multiple hierarchical sign positions, and each hierarchical sign position in the lung image corresponds to multiple sign categories, such as sign category 1, sign Category 2, etc., to adjust and train the image weight matrix of multiple feature features (specific feature image matrix) corresponding to each feature category, such as the first feature feature, the second feature feature, etc., such as the image of the first feature feature
  • the weight matrix needs to be increased, and the image weight matrix of the second sign feature needs to be increased.
  • Step A3 Use the second use case as a test case to test the basic recognition network model after adjustment and training, and configure to obtain the target recognition type.
  • the second use case is used as a test case to test the basic recognition network model after adjustment training. If the test determines that the test accuracy of the basic recognition network model after adjustment training is greater than expected When the accuracy is set, the basic recognition network model after the adjustment training is used as the target recognition type, wherein if the test determines that the test accuracy of the basic recognition network model after the adjustment training is less than the preset accuracy , Continue to train and adjust the basic recognition network model after the adjustment training, and in the subsequent training and adjustment process, perform the sampling method of the application case with replacement, and configure it to obtain a new first use case and a new second use case. Use case to obtain the target recognition network model with final training.
  • the medical image to be recognized is convolved, pooled, activated, and classified according to the target recognition network model to finally obtain the processing result.
  • This application determines the target type of the medical image to be recognized when the medical image to be recognized is detected; according to the target type of the medical image to be recognized, and the preset type of the medical image to be recognized and the identification network model Association relationship, distributing the medical image to be recognized of the target type to the target recognition network model corresponding to the target type; based on the target recognition network model to preset the target sign position and corresponding to the medical image to be recognized
  • the preset processing for determining the target sign category is configured to obtain and output the processing result. In this embodiment, it is no longer necessary to send clinical images to different imaging doctors for artificial recognition. Instead, when a medical image to be recognized is detected, for the different types of medical images to be recognized, the target type is to be recognized.
  • the recognized medical image is distributed to the target recognition network model corresponding to the target type, and the target recognition network model is a model that can accurately recognize the location of the sign and the sign category of the image after training. Therefore, in this embodiment The automatic recognition of the medical image to be recognized is realized, thus, the recognition accuracy of the clinical image is improved, and the recognition efficiency of the existing clinical image is reduced. The technical problems of low recognition accuracy and low recognition efficiency of existing clinical images are solved.
  • this application provides another embodiment of an image recognition method based on deep learning.
  • the pre-processed images after the layering are respectively subjected to preset target sign category recognition processing, and all The steps of the said target hierarchical sign position and the corresponding preset target sign category as the processing result and output include:
  • Step B1 performing alternate processing of convolution and pooling for a preset number of times on the preprocessed image after the layering to obtain an initial processing result
  • Performing alternate processing of convolution and pooling for a preset number of times on the preprocessed image after the layering, and the preset number of times may be 3 times to obtain an initial processing result.
  • the step of performing alternate processing of convolution and pooling for a preset number of times on the preprocessed image after the layering to obtain an initial processing result includes:
  • Step C1 performing filtering and convolution processing on the layered preprocessed image according to the image weight matrix to obtain a convolution processing result
  • the convolution process can be understood as: the sign feature of one part of the image is the same as the other part, that is, the sign feature learned in this part can also appear on the corresponding other part, so the learned sign feature is used as the detector , Applied to any place of this image, that is, the characteristic features learned from a small-scale image are convolved with the original large-size image.
  • the convolution can be the characteristic matrix of the corresponding image and the pre-multiple The corresponding detection matrices are multiplied by the corresponding sign features, and finally the image weights are summed to obtain the convolution processing result.
  • the pixel matrix corresponding to the layered preprocessed image is multiplied by the detection matrix or the pixel matrix corresponding to the preset feature feature, and finally the image weight is summed, Obtain the convolution processing result.
  • Step C2 performing pooling processing on the convolution processing result, and configuring to obtain a pooling processing result
  • the convolution processing result is pooled and configured to obtain the pooling processing result.
  • the step of performing pooling processing on the convolution processing result and configuring to obtain the pooling processing result includes:
  • Step D1 dividing the convolution processing result into a plurality of image matrices with the same size and preset size;
  • the convolution processing result is divided into a plurality of 3*3-dimensional image matrices.
  • Step D2 Obtain the maximum pixel value or average pixel value in the image matrix of the preset size, and replace the image matrix of the preset size with the maximum pixel value or the average pixel value, and configure to obtain a new image matrix;
  • the maximum pixel value or the average pixel value in the image matrix of the preset size replace the image matrix of the preset size with the maximum pixel value or the average pixel value, and configure to obtain a new image matrix, such as 3*
  • the maximum pixel value in the 3-dimensional image matrix is 1, then 1 is substituted for the 3*3-dimensional image matrix. Since the convolution processing result includes multiple 3*3-dimensional image matrices, a new one can be obtained finally Image matrix.
  • Step D3 Set the new image matrix as the pooling processing result.
  • Step C3 According to the preset number of times, the pooling processing result is again subjected to a corresponding number of convolution and pooling alternate processings, and configured to obtain an initial processing result.
  • C1-C2 are one-time convolution and pooling alternate processing procedures.
  • a preset number of convolution and pooling alternate processing procedures are required to be configured to obtain an initial processing result.
  • Step B2 Classify the initial processing result according to preset classification presets, configure to obtain the preset target sign category of the pre-processed image after stratification, and divide the target sign position and The corresponding preset target sign category is used as the processing result and output.
  • the initial processing result is classified according to the preset classification preset. If the initial processing result is less than the preset classification preset, the preset target sign category can correspond to the first category. When the processing result is greater than or equal to the preset classification preset, the preset target sign category can correspond to the second category. After the preset target sign category of the preprocessed image after the layering is obtained, the target sign category The hierarchical sign position and the corresponding preset target sign category are used as the processing result and output.
  • the initial processing result is obtained by performing a preset number of alternate processing of convolution and pooling on the preprocessed image after the layering; the initial processing result is performed according to a preset classification preset
  • the classification processing is configured to obtain the preset target sign category of the preprocessed image after the layering, and the target layered sign position and the corresponding preset target sign category are used as a processing result and output.
  • the automatic recognition of medical images to be recognized is specifically realized to improve the efficiency of image recognition.
  • the present application provides another embodiment of an image recognition method based on deep learning.
  • the preset target sign position and the corresponding position of the medical image to be recognized are performed based on the target recognition network model.
  • the preset processing for determining the target sign category configured to obtain the processing result and output, includes:
  • Step S40 Determine the email address of the target person receiving the image of the processing result according to the preset target sign category
  • the mail address of the person receiving the image of the target of the processing result is determined according to the preset target sign category, wherein there is a mapping relationship between the mail address of the image person and the sign category.
  • Step S50 according to the email address, send the processing result to the target image receiving person.
  • the processing result is sent to the target image receiving person for further confirmation by the image person.
  • the processing result is sent to the target image recipient according to the email address .
  • the processing result is sent to the corresponding target image receiving personnel instead of sending it randomly, which improves the efficiency of image recognition.
  • FIG. 3 is a schematic diagram of the device structure of the hardware operating environment involved in the solution of the embodiment of the present application.
  • the image recognition device based on deep learning in the embodiments of this application can be a PC, or a smart phone, a tablet computer, an e-book reader, MP3 (Moving Picture Experts Group Audio Layer III, moving picture experts compress standard audio layer 3) player, MP4 (Moving Picture Experts) Group Audio Layer IV, dynamic image experts compress standard audio layer 4) Players, portable computers and other terminal equipment.
  • MP3 Moving Picture Experts Group Audio Layer III, moving picture experts compress standard audio layer 3
  • MP4 Motion Picture Experts
  • Group Audio Layer IV dynamic image experts compress standard audio layer 4
  • the image recognition device based on deep learning may include a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002.
  • the communication bus 1002 is configured to realize connection and communication between the processor 1001 and the memory 1005.
  • the memory 1005 can be a high-speed RAM memory or a stable memory (non-volatile memory), such as disk storage.
  • the memory 1005 may also be a storage device independent of the foregoing processor 1001.
  • the deep learning-based image recognition device may also include a target user interface, a network interface, a camera, RF (Radio Frequency (radio frequency) circuits, sensors, audio circuits, WiFi modules, etc.
  • the target user interface may include a display screen (Display) and an input unit such as a keyboard (Keyboard), and the optional target user interface may also include a standard wired interface and a wireless interface.
  • the optional network interface can include standard wired interface and wireless interface (such as WI-FI interface).
  • FIG. 3 does not constitute a limitation on the image recognition device based on deep learning, and may include more or less components than shown in the figure, or a combination Certain components, or different component arrangements.
  • the memory 1005 as a computer storage medium may include an operating system, a network communication module, and computer readable instructions.
  • the operating system is a computer-readable instruction that manages and controls the hardware and software resources of an image recognition device based on deep learning, and supports the operation of computer-readable instructions and other software and/or computer-readable instructions.
  • the network communication module is configured to realize communication between various components in the memory 1005 and communication with other hardware and software in an image recognition device based on deep learning.
  • the processor 1001 is configured to execute computer-readable instructions stored in the memory 1005 to implement the steps of any one of the above-mentioned deep learning-based image recognition methods.
  • the specific implementation of the deep learning-based image recognition device of the present application is basically the same as the foregoing embodiments of the deep learning-based image recognition method, and will not be repeated here.
  • This application also provides an image recognition device based on deep learning.
  • the specific implementation of the image recognition device based on deep learning of this application is basically the same as the above embodiments of the image recognition method based on deep learning, and will not be repeated here.
  • This application provides a storage medium, which may be a non-volatile readable storage medium, the storage medium stores one or more computer-readable instructions, and the one or more computer-readable instructions also It may be executed by one or more processors to implement the steps of any one of the above-mentioned deep learning-based image recognition methods.
  • the specific implementation of the storage medium of the present application is basically the same as the foregoing embodiments of the deep learning-based image recognition method, and will not be repeated here.

Landscapes

  • Engineering & Computer Science (AREA)
  • Quality & Reliability (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

本申请公开了一种基于深度学习的影像识别方法、装置、设备及存储介质，所述方法包括：在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。本申请基于智能决策方式解决现有临床影像的识别准确度低，识别效率低的技术问题。

Description

基于深度学习的影像识别方法、装置、设备及存储介质
本申请要求于2019年06月18日提交中国专利局、申请号为201910529344.4、发明名称为“基于深度学习的影像识别方法、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
本申请涉及医学影像处理技术领域，尤其涉及一种基于深度学习的影像识别方法、装置、设备及存储介质。
背景技术
在临床影像识别中，不同病灶的征象不同，目前，常常需要不同的影像医生人为结合征象，患者年龄等多种信息综合判断，才能给出精准的识别结果，然而，由于影像医生的人手以及经验严重不足，尤其是针对不同影像都进行分别识别的影像医生的人手以及经验不足，降低了现有临床影像的识别准确度，且降低了现有临床影像的识别效率。
发明内容
本申请的主要目的在于提供一种基于深度学习的影像识别方法、装置、设备及存储介质，旨在解决现有临床影像的识别准确度低，识别效率低的技术问题。
为实现上述目的，本申请提供一种基于深度学习的影像识别方法，所述基于深度学习的影像识别方法包括：
在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。
本申请还提供一种基于深度学习的影像识别装置，所述基于深度学习的影像识别装置包括：
第一确定模块，设置为在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
分发模块，设置为根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
处理模块，设置为基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。
可选地，所述处理模块包括：
预处理单元，设置为对所述待识别医学影像进行预处理，配置为得到预处理图像；
分层单元，设置为根据所述目标识别网络模型对所述预处理图像进行分层处理，确定所述待识别医学影像对应的目标分层征象位置；
输出单元，设置为对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
可选地，所述基于深度学习的影像识别装置还包括：
获取模块，设置为获取预设的各类型影像的用例，对应挑选同一类型影像预设比例的所述用例设为第一用例，对应同一类型影像第一训练用例外的其他用例设为第二用例；
训练模块，设置为将所述第一用例作为训练用例进行所述目标识别网络模型对应基础识别网络模型的调整训练，以调整训练所述基础识别网络模型中针对所述影像的的多个分层征象位置的多个征象类别，所分别对应的征象特征的图像权值矩阵；
测试模块，设置为将所述第二用例作为测试用例进行调整训练后的所述基础识别网络模型的测试，配置为得到所述目标识别型。
可选地，所述输出单元包括：
交替处理子单元，设置为对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果；
分类子单元，设置为根据预设分类预置对所述初始处理结果进行分类处理，配置为得到所述分层后的所述预处理图像的预设的目标征象类别，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
可选地，所述交替处理子单元设置为实现：
根据所述图像权值矩阵，对所述分层后的所述预处理图像进行滤波卷积处理，得到卷积处理结果；
对所述卷积处理结果进行池化处理，配置为得到池化处理结果；
根据所述预设次数，对所述池化处理结果再次进行相应次数的卷积与池化交替处理，配置为得到初始处理结果。
可选地，所述交替处理子单元还设置为实现：
将所述卷积处理结果分割为多个大小一致的预设尺寸的图像矩阵；
获取所述预设尺寸的图像矩阵中的最大像素值或者平均像素值，将所述最大像素值或者平均像素值代替所述预设尺寸的图像矩阵，配置为得到新的图像矩阵；
将所述新的图像矩阵设为所述池化处理结果。
可选地，所述基于深度学习的影像识别装置还包括：
第二确定模块，设置为根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址；
发送模块，设置为根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员。
此外，为实现上述目的，本申请还提供一种基于深度学习的影像识别设备，所述基于深度学习的影像识别设备包括：存储器、处理器，通信总线以及存储在所述存储器上的计算机可读指令，
所述通信总线设置为实现处理器与存储器间的通信连接；
所述处理器设置为执行所述计算机可读指令，以实现以下步骤：
在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。
此外，为实现上述目的，本申请还提供一种存储介质，所述存储介质存储有一个或者一个以上计算机可读指令，所述一个或者一个以上计算机可读指令可被一个或者一个以上的处理器执行以设置为：
在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。
本申请通过在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。在本实施例中，不再需要将临床影像发送给不同的影像医生进行人为识别，而是在检测到待识别医学影像时，针对该不同类型的待识别医学影像，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中，而目标识别网络模型都是经过训练后能够对应对影像进行征象位置以及征象类别的准确识别的模型，因而，在本实施例中实现自动对所述待识别医学影像进行识别，因而，提升了临床影像的识别准确度，降低了现有临床影像的识别效率。解决了现有临床影像的识别准确度低，识别效率低的技术问题。
附图说明
图1为本申请基于深度学习的影像识别方法第一实施例的流程示意图；
图2为本申请基于深度学习的影像识别方法中基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤的细化流程示意图；
图3是本申请实施例方法涉及的硬件运行环境的设备结构示意图。
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，并不设置为限定本申请。
本申请提供一种基于深度学习的影像识别方法，在本申请基于深度学习的影像识别方法的第一实施例中，参照图1，所述基于深度学习的影像识别方法包括：
步骤S10，在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
需要说明的是，在本实施例中，基于深度学习的影像识别方法是具有众多类型影像的识别功能的，众多类型影像识别功能指的是在检测到预设的各种类型影像时，都能相应的识别，该众多类型影像包括肺部影像、腰椎间盘影像、肝部影像等类型，要在检测到预设的各种类型影像时，都能进行相应的识别，需要在基于深度学习的影像识别方法对应的基于深度学习的影像识别装置中设置有相应类型的识别网络模型，该识别网络模型包括但不限于卷积神经网络模型等类型，具体地，要在检测到肺部影像时，能进行相应的识别，要在所述影像识别装置中设置有肺部影像的识别网络模型，要在检测到腰椎间盘影像时，能进行相应的识别，要在所述影像识别装置中设置有腰椎间盘影像的识别网络模型等。
在本实施例中，在检测到待识别医学影像时，确定所述待识别医学影像的目标类型，具体地，在检测到待识别医学影像时，确定所述待识别医学影像的目标类型方式包括：
方式一：根据待识别医学影像中预先携带的影像标识，确定所述待识别医学影像的目标类型。
方式二：获取待识别医学影像的来源渠道，根据该来源渠道确定所述待识别医学影像的目标类型，其中，来源渠道指的是来源科室或者是来源机器等。
确定待识别医学影像的目标类型的目的在于将待识别医学影像分发至所述目标类型对应的目标识别网络模型中。
步骤S20，根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
在得到待识别医学影像的目标类型后，根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中。例如，若待识别医学影像的目标类型为肺部影像，将所述待识别医学影像分发至所述肺部影像类型对应的第一目标识别网络模型中，若待识别医学影像的目标类型为腰椎间盘影像，将所述待识别医学影像分发至所述腰椎间盘影像类型对应的第二目标识别网络模型中。将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中的目的在于对待识别医学影像进行针对性的分析处理。
步骤S30，基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。
在本实施例中，基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出，具体地，基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理包括对预设的目标征象位置确定的第一确定操作，以及对预设的目标征象类别确定的第二确定操作。其中，该第一确定操作以及第二确定操作是可以同步进行确定的（目标识别网络模型同步确定图像所处的位置以及该位置对应的征象类别），另外，该第一确定操作以及第二确定操作也可以是先后进行确定的，先后进行确定的指的是目标识别网络模型先进行预设的目标征象位置的确定，后基于预设的目标征象位置进行对应预设的目标征象类别的确定。
具体地，以待识别医学影像为腰椎间盘影像为例进行具体说明，由于腰椎间盘包括分区1（腰1上半部位）、分区2（腰1下半部位）、分区3（腰2上半部位）以及分区4（腰2下半部位），该各个分区还包括各个子分区，预设的目标征象位置指的是该待识别医学影像属于哪个具体的分区或者子分区，该预设的目标征象类别包括第一（骨折）征象类别、第二（腰椎间盘膨出）征象类别、第三(腰椎间盘突出征象)类别等。
参照图2，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤包括：
步骤S31，对所述待识别医学影像进行预处理，配置为得到预处理图像；
在本实施例中，在得到处理结果前，对所述待识别医学影像进行预处理，配置为得到预处理图像，具体预处理过程包括首先对图像进行预设对比度拉伸，预设图像大小尺寸的改变，或者预设位移的平移等，预设图像大小尺寸的改变包括如预设图像大小缩放、扩大等操作，对所述待识别医学影像进行预处理的目的在于确保后续切割或者图像处理时的规整度。
步骤S32，根据所述目标识别网络模型对所述预处理图像进行分层处理，确定所述待识别医学影像对应的目标分层征象位置；
在得到预处理图像后，根据所述目标识别网络模型对所述预处理图像进行分层处理，确定所述待识别医学影像对应的目标分层征象位置，具体地，分层处理过程中可以使用自适应阈值最大类间方差法，其中，自适应阈值最大类间方差法，即是将所述待处理图像按照灰度特征分成图像和背景两类，以提取特征信息，配置为得到组织区域，在得到组织区域后，将整个组织区域进行分层识别，以确定所述待识别医学影像对应的目标分层征象位置。
步骤S33，对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
在分层后，对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出，需要说明的是，在对分层后的所述预处理图像分别进行预设的目标征象类别识别处理过程中，涉及对图像的卷积、池化、激活以及分类处理过程，以最终得到处理结果。
在得到处理结果前，是需要得到准确的目标识别网络模型的。
具体地，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤之前包括：
步骤A1，获取预设的各类型影像的用例，对应挑选同一类型影像预设比例的所述用例设为第一用例，对应同一类型影像第一训练用例外的其他用例设为第二用例；
在本实施例中，预先存储有各类型影像的用例，例如，预先存储有该肺部影像的用例、腰椎间盘影像的用例、肝部影像的用例等，获取预设的各类型影像的用例，对应挑选同一类型影像预设比例的所述用例设为第一用例，对应同一类型影像第一训练用例外的其他用例设为第二用例。
步骤A2，将所述第一用例作为训练用例进行所述目标识别网络模型对应基础识别网络模型的调整训练，以调整训练所述基础识别网络模型中针对所述影像的的多个分层征象位置的多个征象类别，所分别对应的征象特征的图像权值矩阵；
将所述第一用例作为训练用例进行所述目标识别网络模型对应基础识别网络模型的调整训练，以调整训练所述基础识别网络模型中针对所述影像的的多个分层征象位置的多个征象类别，所分别对应的征象特征的图像权值矩阵，例如，肺部影像包括多个分层征象位置，该肺部影像的每个分层征象位置对应多个征象类别如征象类别1、征象类别2等，对该每个征象类别对应的多个征象特征（具体的特征图像矩阵）如第一征象特征、第二征象特征等的图像权值矩阵进行调整训练，如第一征象特征的图像权值矩阵需要增大，而第二征象特征的图像权值矩阵需要增大等。
步骤A3，将所述第二用例作为测试用例进行调整训练后的所述基础识别网络模型的测试，配置为得到所述目标识别型。
在训练完成后，将所述第二用例作为测试用例进行调整训练后的所述基础识别网络模型的测试，若测试确定该调整训练后的所述基础识别网络模型的测试的测试准确度大于预设准确度时，将所述调整训练后的所述基础识别网络模型作为目标识别型，其中，若测试确定该调整训练后的所述基础识别网络模型的测试的测试准确度小于预设准确度时，继续训练调整所述调整训练后的所述基础识别网络模型，在后续的训练调整过程中，进行对应用例的有放回抽样方式，配置为得到新的第一用例以及新的第二用例，以最终训练得到目标识别网络模型。
在得到目标识别网络模型后，根据该目标识别网络模型对待识别医学影像进行卷积、池化、激活以及分类处理过程，以最终得到处理结果。
本申请通过在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出。在本实施例中，不再需要将临床影像发送给不同的影像医生进行人为识别，而是在检测到待识别医学影像时，针对该不同类型的待识别医学影像，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中，而目标识别网络模型都是经过训练后能够对应对影像进行征象位置以及征象类别的准确识别的模型，因而，在本实施例中实现自动对所述待识别医学影像进行识别，因而，提升了临床影像的识别准确度，降低了现有临床影像的识别效率。解决了现有临床影像的识别准确度低，识别效率低的技术问题。
进一步地，本申请提供基于深度学习的影像识别方法的另一实施例，在该实施例中，所述对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出步骤包括：
步骤B1，对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果；
对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，该预设次数可以为3次，得到初始处理结果。
所述对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果步骤包括：
步骤C1，根据所述图像权值矩阵，对所述分层后的所述预处理图像进行滤波卷积处理，得到卷积处理结果；
其中，卷积过程可以理解为：图像的一部分的征象特征与其他部分是一样的，即是在这一部分学习的征象特征也能出现在相应另一部分上，因而将学习到的征象特征作为探测器，应用到这个图像的任意地方中去，即通过小范围图像所学习到的征象特征跟原本的大尺寸的图像作卷积，在数学上，卷积可以是相应图像的特性矩阵与预先的多个征象特征对应探测矩阵相乘最后再图像权值求和，得到卷积处理结果。
在本实施例中，根据所述图像权值矩阵，将分层后的所述预处理图像对应的像素矩阵与预设征象特征对应的探测矩阵或者像素矩阵相乘，最后图像权值求和，得到卷积处理结果。
步骤C2，对所述卷积处理结果进行池化处理，配置为得到池化处理结果；
在得到卷积处理结果后，对所述卷积处理结果进行池化处理，配置为得到池化处理结果。
具体地，所述对所述卷积处理结果进行池化处理，配置为得到池化处理结果步骤包括：
步骤D1，将所述卷积处理结果分割为多个大小一致的预设尺寸的图像矩阵；
例如将所述卷积处理结果分割为多个3*3维的图像矩阵。
步骤D2，获取所述预设尺寸的图像矩阵中的最大像素值或者平均像素值，将所述最大像素值或者平均像素值代替所述预设尺寸的图像矩阵，配置为得到新的图像矩阵；
获取所述预设尺寸的图像矩阵中的最大像素值或者平均像素值，将所述最大像素值或者平均像素值代替所述预设尺寸的图像矩阵，配置为得到新的图像矩阵，如3*3维的图像矩阵中最大像素值为1，则将1代替所述3*3维的图像矩阵，由于卷积处理结果中包括多个3*3维的图像矩阵，因而，最后能够得到新的图像矩阵。
步骤D3，将所述新的图像矩阵设为所述池化处理结果。
将所述新的图像矩阵设为所述池化处理结果。
步骤C3，根据所述预设次数，对所述池化处理结果再次进行相应次数的卷积与池化交替处理，配置为得到初始处理结果。
上述C1-C2为一次卷积以及池化交替处理过程，在本实施例中，需要进行预设次数的卷积以及池化的交替处理过程，配置为得到初始处理结果。
步骤B2，根据预设分类预置对所述初始处理结果进行分类处理，配置为得到所述分层后的所述预处理图像的预设的目标征象类别，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
在得到初始处理结果后，根据预设分类预置对所述初始处理结果进行分类处理，若初始处理结果小于预设分类预置时，预设的目标征象类别可对应为第一类，若初始处理结果大于等于预设分类预置时，预设的目标征象类别可对应为第二类，在得到所述分层后的所述预处理图像的预设的目标征象类别后，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
在本实施例中，通过对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果；根据预设分类预置对所述初始处理结果进行分类处理，配置为得到所述分层后的所述预处理图像的预设的目标征象类别，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。在本实施例中，具体实现自动识别待识别医学影像，以提升影像识别效率。
进一步地，本申请提供基于深度学习的影像识别方法的另一实施例，在该实施例中，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤之后包括：
步骤S40，根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址；
在本实施例中，在输出处理结果后，根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址，其中，影像人员的邮件地址与征象类别存在映射关系。
步骤S50，根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员。
根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员，以供所述影像人员进行后续的进一步确认。
在本实施例中，若根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址；根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员。在本实施例中，实现将所述处理结果发送给对应的目标接收影像人员，而不是随意发送，提升了影像识别效率。
参照图3，图3是本申请实施例方案涉及的硬件运行环境的设备结构示意图。
本申请实施例基于深度学习的影像识别设备可以是PC，也可以是智能手机、平板电脑、电子书阅读器、MP3（Moving Picture Experts Group Audio Layer III，动态影像专家压缩标准音频层面3）播放器、MP4（Moving Picture Experts Group Audio Layer IV，动态影像专家压缩标准音频层面4）播放器、便携计算机等终端设备。
如图3所示，该基于深度学习的影像识别设备可以包括：处理器1001，例如CPU，存储器1005，通信总线1002。其中，通信总线1002设置为实现处理器1001和存储器1005之间的连接通信。存储器1005可以是高速RAM存储器，也可以是稳定的存储器（non-volatile memory），例如磁盘存储器。存储器1005可选的还可以是独立于前述处理器1001的存储设备。
可选地，该基于深度学习的影像识别设备还可以包括目标用户接口、网络接口、摄像头、RF（Radio Frequency，射频）电路，传感器、音频电路、WiFi模块等等。目标用户接口可以包括显示屏（Display）、输入单元比如键盘（Keyboard），可选目标用户接口还可以包括标准的有线接口、无线接口。网络接口可选的可以包括标准的有线接口、无线接口（如WI-FI接口）。
本领域技术人员可以理解，图3中示出的基于深度学习的影像识别设备结构并不构成对基于深度学习的影像识别设备的限定，可以包括比图示更多或更少的部件，或者组合某些部件，或者不同的部件布置。
如图3所示，作为一种计算机存储介质的存储器1005中可以包括操作系统、网络通信模块以及计算机可读指令。操作系统是管理和控制基于深度学习的影像识别设备硬件和软件资源的计算机可读指令，支持计算机可读指令以及其它软件和/或计算机可读指令的运行。网络通信模块设置为实现存储器1005内部各组件之间的通信，以及与基于深度学习的影像识别设备中其它硬件和软件之间通信。
在图3所示的基于深度学习的影像识别设备中，处理器1001设置为执行存储器1005中存储的计算机可读指令，实现上述任一项所述的基于深度学习的影像识别方法的步骤。
本申请基于深度学习的影像识别设备具体实施方式与上述基于深度学习的影像识别方法各实施例基本相同，在此不再赘述。
本申请还提供一种基于深度学习的影像识别装置，本申请基于深度学习的影像识别装置具体实施方式与上述基于深度学习的影像识别方法各实施例基本相同，在此不再赘述。
本申请提供了一种存储介质，所述存储介质可以为非易失性可读存储介质，所述存储介质存储有一个或者一个以上计算机可读指令，所述一个或者一个以上计算机可读指令还可被一个或者一个以上的处理器执行以设置为实现上述任一项所述的基于深度学习的影像识别方法的步骤。
本申请存储介质具体实施方式与上述基于深度学习的影像识别方法各实施例基本相同，在此不再赘述。
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处理范围内。

Claims (20)

  1. 一种基于深度学习的影像识别方法，其中，所述基于深度学习的影像识别方法包括：
    在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
    根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
    基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出；
    根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址；
    根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员。
  2. 如权利要求1所述的基于深度学习的影像识别方法，其中，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤包括：
    对所述待识别医学影像进行预处理，配置为得到预处理图像；
    根据所述目标识别网络模型对所述预处理图像进行分层处理，确定所述待识别医学影像对应的目标分层征象位置；
    对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
  3. 如权利要求2所述的基于深度学习的影像识别方法，其中，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤之前包括：
    获取预设的各类型影像的用例，对应挑选同一类型影像预设比例的所述用例设为第一用例，对应同一类型影像第一训练用例外的其他用例设为第二用例；
    将所述第一用例作为训练用例进行所述目标识别网络模型对应基础识别网络模型的调整训练，以调整训练所述基础识别网络模型中针对所述影像的的多个分层征象位置的多个征象类别，所分别对应的征象特征的图像权值矩阵；
    将所述第二用例作为测试用例进行调整训练后的所述基础识别网络模型的测试，配置为得到所述目标识别型。
  4. 如权利要求3所述的基于深度学习的影像识别方法，其中，所述对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出步骤包括：
    对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果；
    根据预设分类预置对所述初始处理结果进行分类处理，配置为得到所述分层后的所述预处理图像的预设的目标征象类别，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
  5. 如权利要求4所述的基于深度学习的影像识别方法，其中，所述对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果步骤包括：
    根据所述图像权值矩阵，对所述分层后的所述预处理图像进行滤波卷积处理，得到卷积处理结果；
    对所述卷积处理结果进行池化处理，配置为得到池化处理结果；
    根据所述预设次数，对所述池化处理结果再次进行相应次数的卷积与池化交替处理，配置为得到初始处理结果。
  6. 如权利要求5所述的基于深度学习的影像识别方法，其中，所述对所述卷积处理结果进行池化处理，配置为得到池化处理结果步骤包括：
    将所述卷积处理结果分割为多个大小一致的预设尺寸的图像矩阵；
    获取所述预设尺寸的图像矩阵中的最大像素值或者平均像素值，将所述最大像素值或者平均像素值代替所述预设尺寸的图像矩阵，配置为得到新的图像矩阵；
    将所述新的图像矩阵设为所述池化处理结果。
  7. 一种基于深度学习的影像识别装置，其中，所述基于深度学习的影像识别装置包括：
    第一确定模块，设置为在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
    分发模块，设置为根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
    处理模块，设置为基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出；
    第二确定模块，设置为根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址；
    发送模块，设置为根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员。
  8. 如权利要求7所述的基于深度学习的影像识别装置，其中，所述处理模块包括：
    预处理单元，设置为对所述待识别医学影像进行预处理，配置为得到预处理图像；
    分层单元，设置为根据所述目标识别网络模型对所述预处理图像进行分层处理，确定所述待识别医学影像对应的目标分层征象位置；
    输出单元，设置为对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
  9. 一种基于深度学习的影像识别设备，其中，所述基于深度学习的影像识别设备包括：存储器、处理器，通信总线以及存储在所述存储器上的计算机可读指令，
    所述通信总线设置为实现处理器与存储器间的通信连接；
    所述处理器设置为执行所述计算机可读指令，以实现如下步骤：
    在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
    根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
    基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出；
    根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址；
    根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员。
  10. 如权利要求9所述的基于深度学习的影像识别设备，其中，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤包括：
    对所述待识别医学影像进行预处理，配置为得到预处理图像；
    根据所述目标识别网络模型对所述预处理图像进行分层处理，确定所述待识别医学影像对应的目标分层征象位置；
    对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
  11. 如权利要求10所述的基于深度学习的影像识别设备，其中，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤之前包括：
    获取预设的各类型影像的用例，对应挑选同一类型影像预设比例的所述用例设为第一用例，对应同一类型影像第一训练用例外的其他用例设为第二用例；
    将所述第一用例作为训练用例进行所述目标识别网络模型对应基础识别网络模型的调整训练，以调整训练所述基础识别网络模型中针对所述影像的的多个分层征象位置的多个征象类别，所分别对应的征象特征的图像权值矩阵；
    将所述第二用例作为测试用例进行调整训练后的所述基础识别网络模型的测试，配置为得到所述目标识别型。
  12. 如权利要求11所述的基于深度学习的影像识别设备，其中，所述对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出步骤包括：
    对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果；
    根据预设分类预置对所述初始处理结果进行分类处理，配置为得到所述分层后的所述预处理图像的预设的目标征象类别，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
  13. 如权利要求12所述的基于深度学习的影像识别设备，其中，所述对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果步骤包括：
    根据所述图像权值矩阵，对所述分层后的所述预处理图像进行滤波卷积处理，得到卷积处理结果；
    对所述卷积处理结果进行池化处理，配置为得到池化处理结果；
    根据所述预设次数，对所述池化处理结果再次进行相应次数的卷积与池化交替处理，配置为得到初始处理结果。
  14. 如权利要求13所述的基于深度学习的影像识别设备，其中，所述对所述卷积处理结果进行池化处理，配置为得到池化处理结果步骤包括：
    将所述卷积处理结果分割为多个大小一致的预设尺寸的图像矩阵；
    获取所述预设尺寸的图像矩阵中的最大像素值或者平均像素值，将所述最大像素值或者平均像素值代替所述预设尺寸的图像矩阵，配置为得到新的图像矩阵；
    将所述新的图像矩阵设为所述池化处理结果。
  15. 一种存储介质，其中，所述存储介质上存储有计算机可读指令，所述计算机可读指令被处理器执行时时实现如下步骤：
    在检测到待识别医学影像时，确定所述待识别医学影像的目标类型；
    根据所述待识别医学影像的目标类型，以及预设的待识别医学影像的类型与各个识别网络模型的关联关系，将所述目标类型的待识别医学影像分发至所述目标类型对应的目标识别网络模型中；
    基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出；
    根据所述预设的目标征象类别，确定所述处理结果的目标接收影像人员的邮件地址；
    根据所述邮件地址，将所述处理结果发送给所述目标接收影像人员。
  16. 如权利要求15所述的存储介质，其中，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤包括：
    对所述待识别医学影像进行预处理，配置为得到预处理图像；
    根据所述目标识别网络模型对所述预处理图像进行分层处理，确定所述待识别医学影像对应的目标分层征象位置；
    对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
  17. 如权利要求16所述的存储介质，其中，所述基于所述目标识别网络模型对所述待识别医学影像进行预设的目标征象位置以及对应预设的目标征象类别的确定处理，配置为得到处理结果并输出步骤之前包括：
    获取预设的各类型影像的用例，对应挑选同一类型影像预设比例的所述用例设为第一用例，对应同一类型影像第一训练用例外的其他用例设为第二用例；
    将所述第一用例作为训练用例进行所述目标识别网络模型对应基础识别网络模型的调整训练，以调整训练所述基础识别网络模型中针对所述影像的的多个分层征象位置的多个征象类别，所分别对应的征象特征的图像权值矩阵；
    将所述第二用例作为测试用例进行调整训练后的所述基础识别网络模型的测试，配置为得到所述目标识别型。
  18. 如权利要求17所述的存储介质，其中，所述对分层后的所述预处理图像分别进行预设的目标征象类别识别处理，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出步骤包括：
    对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果；
    根据预设分类预置对所述初始处理结果进行分类处理，配置为得到所述分层后的所述预处理图像的预设的目标征象类别，将所述目标分层征象位置以及对应的预设的目标征象类别作为处理结果并输出。
  19. 如权利要求18所述的存储介质，其中，所述对所述分层后的所述预处理图像进行预设次数的卷积与池化交替处理，得到初始处理结果步骤包括：
    根据所述图像权值矩阵，对所述分层后的所述预处理图像进行滤波卷积处理，得到卷积处理结果；
    对所述卷积处理结果进行池化处理，配置为得到池化处理结果；
    根据所述预设次数，对所述池化处理结果再次进行相应次数的卷积与池化交替处理，配置为得到初始处理结果。
  20. 如权利要求19所述的存储介质，其中，所述对所述卷积处理结果进行池化处理，配置为得到池化处理结果步骤包括：
    将所述卷积处理结果分割为多个大小一致的预设尺寸的图像矩阵；
    获取所述预设尺寸的图像矩阵中的最大像素值或者平均像素值，将所述最大像素值或者平均像素值代替所述预设尺寸的图像矩阵，配置为得到新的图像矩阵；
    将所述新的图像矩阵设为所述池化处理结果。
PCT/CN2019/117573 2019-06-18 2019-11-12 基于深度学习的影像识别方法、装置、设备及存储介质 WO2020253048A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910529344.4A CN110378876A (zh) 2019-06-18 2019-06-18 基于深度学习的影像识别方法、装置、设备及存储介质
CN201910529344.4 2019-06-18

Publications (1)

Publication Number Publication Date
WO2020253048A1 true WO2020253048A1 (zh) 2020-12-24

Family

ID=68249200

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/117573 WO2020253048A1 (zh) 2019-06-18 2019-11-12 基于深度学习的影像识别方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN110378876A (zh)
WO (1) WO2020253048A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113658175A (zh) * 2021-09-02 2021-11-16 数坤(北京)网络科技股份有限公司 一种征象数据的确定方法及装置

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110378876A (zh) * 2019-06-18 2019-10-25 平安科技（深圳）有限公司 基于深度学习的影像识别方法、装置、设备及存储介质
CN111325745B (zh) * 2020-03-09 2023-08-25 北京深睿博联科技有限责任公司 骨折区域分析方法和装置、电子设备及可读存储介质
CN113555088A (zh) * 2020-04-23 2021-10-26 上海联影智能医疗科技有限公司 图像处理方法、装置、系统和存储介质
CN113807146A (zh) * 2020-06-15 2021-12-17 深圳绿米联创科技有限公司 门锁识别方法、装置及电子设备
CN112766314B (zh) * 2020-12-31 2024-05-28 上海联影智能医疗科技有限公司 解剖结构的识别方法、电子设备及存储介质
CN114549482A (zh) * 2022-02-25 2022-05-27 数坤(北京)网络科技股份有限公司 一种图像关联方法、装置、电子设备和存储介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108230289A (zh) * 2016-12-22 2018-06-29 上海哲康医学科技有限公司 基于x线体检正位胸片的计算机辅助诊断系统和方法
CN108305671A (zh) * 2018-01-23 2018-07-20 深圳科亚医疗科技有限公司 由计算机实现的医学图像调度方法、调度系统及存储介质
US20190080195A1 (en) * 2016-03-14 2019-03-14 Baidu Online Network Technology (Beijing) Co., Ltd . Image recognition method and apparatus, device and non-volatile computer storage medium
CN110378876A (zh) * 2019-06-18 2019-10-25 平安科技（深圳）有限公司 基于深度学习的影像识别方法、装置、设备及存储介质

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109363698B (zh) * 2018-10-16 2022-07-12 杭州依图医疗技术有限公司 一种乳腺影像征象识别的方法及装置
CN109859216B (zh) * 2019-02-16 2021-06-25 深圳市未来感知科技有限公司 基于深度学习的测距方法、装置、设备及存储介质

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190080195A1 (en) * 2016-03-14 2019-03-14 Baidu Online Network Technology (Beijing) Co., Ltd . Image recognition method and apparatus, device and non-volatile computer storage medium
CN108230289A (zh) * 2016-12-22 2018-06-29 上海哲康医学科技有限公司 基于x线体检正位胸片的计算机辅助诊断系统和方法
CN108305671A (zh) * 2018-01-23 2018-07-20 深圳科亚医疗科技有限公司 由计算机实现的医学图像调度方法、调度系统及存储介质
CN110378876A (zh) * 2019-06-18 2019-10-25 平安科技（深圳）有限公司 基于深度学习的影像识别方法、装置、设备及存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113658175A (zh) * 2021-09-02 2021-11-16 数坤(北京)网络科技股份有限公司 一种征象数据的确定方法及装置
CN113658175B (zh) * 2021-09-02 2023-08-18 数坤(北京)网络科技股份有限公司 一种征象数据的确定方法及装置

Also Published As

Publication number Publication date
CN110378876A (zh) 2019-10-25

Similar Documents

Publication Publication Date Title
WO2020253048A1 (zh) 基于深度学习的影像识别方法、装置、设备及存储介质
WO2021012508A1 (zh) Ai影像识别方法、装置、设备及存储介质
WO2020000643A1 (zh) Ct图像肺结节检测方法、设备及可读存储介质
WO2019132168A1 (ko) 수술영상데이터 학습시스템
WO2020251217A1 (ko) 사용자 맞춤형 피부 진단 시스템 및 방법
WO2014104472A1 (en) Method and apparatus for dual display
WO2019054638A1 (ko) 영상 분석 방법, 장치 및 컴퓨터 프로그램
WO2020222341A1 (ko) 사용자의 행동 패턴을 분석하여 공기조화기의 동작을 제어하는 방법 및 공기조화기
WO2020207038A1 (zh) 基于人脸识别的人数统计方法、装置、设备及存储介质
WO2020107761A1 (zh) 广告文案处理方法、装置、设备及计算机可读存储介质
WO2020017875A1 (en) Electronic apparatus, method for processing image and computer-readable recording medium
WO2019235828A1 (ko) 투 페이스 질병 진단 시스템 및 그 방법
WO2013105815A1 (en) Fetus modeling method and image processing apparatus therefor
WO2021045367A1 (ko) 상담 대상자의 드로잉 과정을 통해 심리 상태를 판단하는 방법 및 컴퓨터 프로그램
WO2019168264A1 (ko) 전자 장치 및 그 제어 방법
WO2020024534A1 (zh) 基于电量预测的用电控制方法、装置、设备和存储介质
WO2021153858A1 (ko) 비정형 피부질환 영상데이터를 활용한 판독보조장치
WO2016179879A1 (zh) 基于网络医院的远程辅助诊疗系统及远程辅助诊疗方法
WO2019000118A1 (zh) 工业流程的视觉检测方法、控制设备、系统及存储装置
WO2019045480A1 (ko) 딥러닝 기반 결핵 검사 방법
WO2016104990A1 (ko) 컨텐츠 제공 장치, 디스플레이 장치 및 그 제어 방법
WO2022014798A1 (ko) 고객 만족도 조사 수행 시스템
WO2014073939A1 (en) Method and apparatus for capturing and displaying an image
CN116108391B (zh) 一种基于无监督学习的人体姿态分类识别系统
WO2021091052A1 (ko) 가중 퍼지 소속함수 기반 심층 신경망을 통한 하위 패턴 학습 및 세부 학습을 이용한 클래스 분류 방법 및 장치

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: 19933891

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: 19933891

Country of ref document: EP

Kind code of ref document: A1