WO2019136922A1 - 肺结节探测方法、应用服务器及计算机可读存储介质 - Google Patents

肺结节探测方法、应用服务器及计算机可读存储介质 Download PDF

Info

Publication number
WO2019136922A1
WO2019136922A1 PCT/CN2018/090911 CN2018090911W WO2019136922A1 WO 2019136922 A1 WO2019136922 A1 WO 2019136922A1 CN 2018090911 W CN2018090911 W CN 2018090911W WO 2019136922 A1 WO2019136922 A1 WO 2019136922A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
lung
picture
net network
dimensional
Prior art date
Application number
PCT/CN2018/090911
Other languages
English (en)
French (fr)
Inventor
王健宗
吴天博
刘莉红
刘新卉
肖京
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 WO2019136922A1 publication Critical patent/WO2019136922A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/08—Projecting images onto non-planar surfaces, e.g. geodetic screens
    • 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
    • G06T2207/30064—Lung nodule

Definitions

  • the present application relates to the field of medical image analysis technologies, and in particular, to a pulmonary nodule detection method, an application server, and a computer readable storage medium.
  • the present application proposes a pulmonary nodule detection method, an application server, and a computer readable storage medium to solve the problem of how to conveniently perform nodule detection.
  • the present application provides a lung nodule detecting method, the method comprising the steps of:
  • the pre-processed three-dimensional map is used as an input of the U-NET network structure to train the U-NET network to output an image highlighting a lung nodule region;
  • the image of the highlighted lung nodule region is fed back to the terminal device.
  • the present application further provides an application server, including a memory and a processor, wherein the memory stores a pulmonary nodule detection system operable on the processor, the lung nodule detection system
  • the processor stores a pulmonary nodule detection system operable on the processor, the lung nodule detection system
  • the image of the highlighted lung nodule region is fed back to the terminal device.
  • the present application further provides a computer readable storage medium storing a pulmonary nodule detection system, the lung nodule detection system being executable by at least one processor, The step of causing the at least one processor to perform the pulmonary nodule detection method as described above.
  • the lung nodule detecting method, the application server and the computer readable storage medium proposed by the present application firstly receive a lung picture sent by the terminal device; and then, preprocess the lung picture, Converting the lung picture into a three-dimensional map; further, constructing a U-NET network structure; and then training the pre-processed three-dimensional map as an input of the U-NET network structure to the U-NET network To output an image highlighting the area of the pulmonary nodule; finally, the image of the highlighted lung nodule area is fed back to the terminal device to assist the doctor to quickly locate the lung nodule from the output image, thereby reducing the pressure on the doctor.
  • FIG. 1 is a schematic diagram of an optional application environment of each embodiment of the present application.
  • FIG. 2 is a schematic diagram of an optional hardware architecture of the application server of FIG. 1;
  • FIG. 3 is a schematic diagram of a program module of a first embodiment of a pulmonary nodule detecting system of the present application
  • FIG. 4 is a schematic diagram of a U-NET network structure of an embodiment of a pulmonary nodule detecting system of the present application
  • FIG. 5 is a schematic diagram of a program module of a second embodiment of the pulmonary nodule detecting system of the present application.
  • FIG. 6 is a schematic flow chart of a first embodiment of a pulmonary nodule detecting method of the present application.
  • FIG. 7 is a schematic flow chart of a second embodiment of a pulmonary nodule detecting method of the present application.
  • Terminal Equipment 1 application server 2 The internet 3 Memory 11 processor 12 Network Interface 13 Pulmonary nodule detection system 200 Receiving module 201 Conversion module 202 Building module 203 Training module 204 Feedback module 205 Setting module 206 Judgment module 207
  • FIG. 1 it is a schematic diagram of an optional application environment of each embodiment of the present application.
  • the present application is applicable to an application environment including, but not limited to, the terminal device 1, the application server 2, and the network 3.
  • the terminal device 1 may be a medical scanning device such as an electronic computed tomography device, a nuclear magnetic resonance device, or an X-ray machine.
  • the application server 2 may be a computing device such as a rack server, a blade server, a tower server, or a rack server.
  • the application server 2 may be a stand-alone server or a server cluster composed of multiple servers.
  • the network 3 may be an intranet, an Internet, a Global System of Mobile communication (GSM), a Wideband Code Division Multiple Access (WCDMA), a 4G network, Wireless or wired networks such as 5G networks, Bluetooth, Wi-Fi, and call networks.
  • GSM Global System of Mobile communication
  • WCDMA Wideband Code Division Multiple Access
  • the application server 2 is respectively connected to one or more of the terminal devices 1 through the network 3 for data transmission and interaction.
  • the terminal device 1 includes a terminal device 1 corresponding to a hospital.
  • FIG. 2 it is a schematic diagram of an optional hardware architecture of the application server 2 of the present application.
  • the application server 2 may include, but is not limited to, the memory 11, the processor 12, and the network interface 13 being communicably connected to each other through a system bus. It is pointed out that Figure 1 only shows the application server 2 with components 11-13, but it should be understood that not all illustrated components may be implemented, and more or fewer components may be implemented instead.
  • the application server 2 may be a computing device such as a rack server, a blade server, a tower server, or a rack server.
  • the application server 2 may be an independent server or a server cluster composed of multiple servers. .
  • the memory 11 includes at least one type of readable storage medium including a flash memory, a hard disk, a multimedia card, a card type memory (eg, SD or DX memory, etc.), a random access memory (RAM), a static Random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk, and the like.
  • the memory 11 may be an internal storage unit of the application server 2, such as a hard disk or memory of the application server 2.
  • the memory 11 may also be an external storage device of the application server 2, such as a plug-in hard disk equipped on the application server 2, a smart memory card (SMC), and a secure digital number. (Secure Digital, SD) card, flash card, etc.
  • the memory 11 can also include both the internal storage unit of the application server 2 and its external storage device.
  • the memory 11 is generally used to store an operating system installed in the application server 2 and various types of application software, such as program codes of the pulmonary nodule detection system 200. Further, the memory 11 can also be used to temporarily store various types of data that have been output or are to be output.
  • the processor 12 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments.
  • the processor 12 is typically used to control the overall operation of the application server 2.
  • the processor 12 is configured to run program code or process data stored in the memory 11, such as running the pulmonary nodule detection system 200 or the like.
  • the network interface 13 may comprise a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the application server 2 and other electronic devices.
  • the present application proposes a pulmonary nodule detection system 200.
  • FIG. 3 there is shown a block diagram of the first embodiment of the pulmonary nodule detection system 200 of the present application.
  • the pulmonary nodule detection system 200 includes a series of computer program instructions stored on the memory 11, and when the computer program instructions are executed by the processor 12, the pulmonary nodules of the embodiments of the present application can be implemented. Probing operation.
  • the pulmonary nodule detection system 200 can be divided into one or more modules based on the particular operations implemented by the various portions of the computer program instructions. For example, in FIG. 3, the pulmonary nodule detection system 200 can be divided into a receiving module 201, a conversion module 202, a construction module 203, a training module 204, and a feedback module 205. among them:
  • the receiving module 201 is configured to receive a lung picture sent by the terminal device 1;
  • the lung picture is a lung CT picture
  • the application server 2 establishes a long connection with the terminal device 1.
  • the terminal device 1 is a CT imaging device.
  • the CT picture is sent to the application server 2, and the application server 2 receives the lung CT picture through the receiving module 201.
  • the conversion module 202 is configured to preprocess the lung picture to convert the lung picture into a three-dimensional picture.
  • the application server 2 first cuts the slice connected to the outside of the lung CT picture through the conversion module 202; Converting the lung CT picture into a 0-1 three-dimensional image using -600HU as a threshold, and in the present embodiment, calculating an optimal threshold of the lung CT using the Otsu method (OTSU);
  • Otsu method Otsu method
  • the size of the image block is based on the average distance of the center reached by the slice, and all small holes are made up; finally, the image pixel values of the three-dimensional image are modified to [-1200, 600], and then scaled to [0, 255].
  • the pixel of the non-lung area is set to 170.
  • the building module 203 is configured to construct a U-NET network structure
  • FIG. 4 is a schematic structural diagram of a U-NET network according to the present application.
  • the three-dimensional map is used as an input to the U-NET network structure, and the U-NET network structure processes the three-dimensional map.
  • the U-NET network structure forms a U-shaped structure through a shrinking network and an expanding network, and extracts features from the picture.
  • the U-NET network consists of 23 convolutional layers.
  • the shrinking network is mainly responsible for the downsampling work, extracting high-dimensional feature information, each downsampling contains two 3x3 convolution operations, a 2x2 pooling operation, and a rectified linear unit (ReLU) as an activation function.
  • ReLU rectified linear unit
  • each upsampling contains two 3x3 convolution operations, and the linear unit is modified as an activation function.
  • the image size is doubled, and the number of features becomes 1/2.
  • each output feature is merged with the features of the phased contraction network to complement the missing boundary information.
  • the training module 204 is configured to train the U-NET network, and use the pre-processed three-dimensional map as an input of the trained U-NET network structure to output an image that highlights a lung nodule region.
  • the application server further converts the three-dimensional map into a 128*128*128 cube by the conversion module 202 as an input of the U-NET network structure.
  • 70% of the input cubes are guaranteed to have a nodule, and the remaining 30% are randomly cropped so that the sample contains background samples.
  • Large pulmonary nodules are fewer in number than small pulmonary nodules. Therefore, when the sample is taken, the nodule sample with a diameter larger than 30mm and 40mm is expanded by 2, 6 times.
  • the U-NET network output that has been trained to highlight the image of the lung nodule region.
  • the feedback module 205 is configured to feed back an image of the highlighted lung nodule region to the terminal device.
  • the application server 2 feeds back the image of the highlighted lung nodule region to the display interface of the terminal device 1 through the feedback module 205 to assist the doctor to quickly locate the lung nodule from the output image.
  • the pulmonary nodule detection system 200 proposed by the present application first receives a picture of a lung transmitted by the terminal device 1; then, pre-processes the picture of the lung to display the picture of the lung Converting into a three-dimensional map; further, constructing a U-NET network structure; then, training the pre-processed three-dimensional map as an input of the U-NET network structure to train the U-NET network to output a prominent lung nodule The image of the region; finally, the image of the highlighted lung nodule region is fed back to the terminal device to assist the doctor to quickly locate the lung nodule from the output image, reducing the pressure on the doctor.
  • the pulmonary nodule detection system 200 includes a setting module, in addition to the receiving module 201, the conversion module 202, the building module 203, the training module 204, and the feedback module 205 in the first embodiment. 206.
  • the setting module 205 is configured to set a ground truth.
  • the determining module 206 is configured to determine whether the three-dimensional image and the reference value are coincident. When the coincidence degree is greater than the reference value, the image is a positive sample, and the coincidence degree is less than the reference value, and the image is a negative sample.
  • the positive and negative samples are determined by using IOU (coincidence degree), and the coincidence degree is greater than the reference value as an image positive sample, and the coincidence degree is smaller than the reference value as an image negative sample.
  • IOU coincidedence degree
  • the number of positive samples is much smaller than that of negative samples, so there will be many false positives, and these false positives with higher scores are treated as so-called Hard negative.
  • the application server 2 further performs a Hard Negative Mining process on the image that has been distinguished from the positive and negative samples by using the U-net network, thereby enhancing the ability to discriminate false positive lung nodules. .
  • the application server 2 passes the three-dimensional graph of the 128*128*128 input to the U-NET network mechanism after four convolutions and maximum pooling, and then passes through 2
  • the deconvolution yields a probability map of 32*32*32; in turn, new branches are added before each deconvolution, and each branch is output 32* through the corresponding deconvolution layer.
  • the pulmonary nodule detection system 200 proposed by the present application can also be used to set a ground truth and determine whether the three-dimensional image and the reference value coincide, when the coincidence degree is greater than The reference value is a positive sample of the image, and the coincidence degree is less than the reference value is a negative sample of the image, and the U-net network performs Hard Negative Mining processing on the image that has been distinguished from the positive and negative samples, thereby enhancing the discrimination of the false positive lung nodules. ability.
  • the present application also proposes a pulmonary nodule detection method.
  • FIG. 6 it is a schematic flowchart of the first embodiment of the pulmonary nodule detecting method of the present application.
  • the order of execution of the steps in the flowchart shown in FIG. 6 may be changed according to different requirements, and some steps may be omitted.
  • Step S501 receiving a lung picture sent by the terminal device 1;
  • the lung picture is a lung CT picture
  • the application server 2 establishes a long connection with the terminal device 1.
  • the terminal device 1 is a CT imaging device.
  • the CT picture is sent to the application server 2, and the application server 2 receives the lung CT picture.
  • Step S502 preprocessing the lung picture to convert the lung picture into a three-dimensional picture.
  • the application server 2 first cuts the slice connected to the outside of the lung CT picture; and further, uses -600HU as the threshold. Converting the lung CT picture into a 0-1 three-dimensional image.
  • the optimal threshold of the lung CT is calculated by using the Otsu method (OTSU); then, the size of the three-dimensional map block is Based on the average distance of the center reached by the slice, all small holes are made up; finally, the image pixel values of the three-dimensional image are modified to [-1200, 600], and then scaled to [0, 255].
  • the pixel of the non-lung area is set to 170.
  • Step S503 constructing a U-NET network structure
  • FIG. 4 is a schematic diagram of a U-NET network structure of an embodiment of the pulmonary nodule detecting system 200 of the present application.
  • the three-dimensional map is used as an input to the U-NET network structure, and the U-NET network structure processes the three-dimensional map.
  • the U-NET network structure forms a U-shaped structure through a shrinking network and an expanding network, and extracts features from the picture.
  • the U-NET network consists of 23 convolutional layers.
  • the shrinking network is mainly responsible for the downsampling work, extracting high-dimensional feature information, each downsampling contains two 3x3 convolution operations, a 2x2 pooling operation, and a rectified linear unit (ReLU) as an activation function.
  • ReLU rectified linear unit
  • each upsampling contains two 3x3 convolution operations, and the linear unit is modified as an activation function.
  • the image size is doubled, and the number of features becomes 1/2.
  • each output feature is merged with the features of the phased contraction network to complement the missing boundary information.
  • Step S504 the preprocessed three-dimensional map is used as an input of the trained U-NET network structure to output an image highlighting the lung nodule region.
  • the application server further converts the three-dimensional map into a 128*128*128 cube by the conversion module 202 as an input of the U-NET network structure.
  • 70% of the input cubes are guaranteed to have a nodule, and the remaining 30% are randomly cropped so that the sample contains background samples.
  • Large pulmonary nodules are fewer in number than small pulmonary nodules. Therefore, when going to the sample, the nodule sample with a diameter larger than 30mm and 40mm is expanded by 2, 6 times.
  • the trained U-NET network output highlights the image of the lung nodule area.
  • Step S505 feeding back the image of the highlighted lung nodule region to the terminal device.
  • the application server 2 feeds back the image of the highlighted lung nodule region to the display interface of the terminal device 1 through the feedback module 205 to assist the doctor to quickly locate the lung nodule from the output image.
  • the pulmonary nodule detecting method proposed by the present application first receives the lung picture transmitted by the terminal device 1; then, preprocesses the lung picture to convert the lung picture into a three-dimensional map; further, constructing a U-NET network structure; then, training the pre-processed three-dimensional map as an input of the U-NET network structure to train the U-NET network to output a prominent lung nodule region Image; Finally, the image of the highlighted lung nodule region is fed back to the terminal device to assist the doctor to quickly locate the lung nodule from the output image, reducing the pressure on the doctor.
  • FIG. 7 is a schematic flow chart of a second embodiment of the pulmonary nodule detecting method of the present application.
  • the pre-processed three-dimensional map is used as an input of the U-NET network structure to train the U-NET network to output an image highlighting an image of a lung nodule region. Specifically, the following steps are included:
  • step S601 a ground truth is set.
  • Step S602 determining whether the three-dimensional image and the reference value are coincident.
  • the coincidence degree is greater than the reference value, the image is a positive sample, and the coincidence degree is less than the reference value is an image negative sample.
  • the positive and negative samples are determined by using IOU (coincidence degree), and the coincidence degree is greater than the reference value as an image positive sample, and the coincidence degree is smaller than the reference value as an image negative sample.
  • IOU coincidedence degree
  • the number of positive samples is much smaller than that of negative samples, so there will be many false positives, and these false positives with higher scores are treated as so-called Hard negative.
  • step S603 the image of the positive and negative samples has been subjected to Hard Negative Mining processing by using the U-net network, thereby enhancing the ability to discriminate false positive lung nodules.
  • the application server 2 passes the three-dimensional graph of the 128*128*128 input to the U-NET network mechanism after four convolutions and maximum pooling, and then passes through 2
  • the deconvolution yields a probability map of 32*32*32; in turn, new branches are added before each deconvolution, and each branch is output 32* through the corresponding deconvolution layer.
  • the lung nodule detection method proposed by the present application can also be used to set a ground truth, and determine whether the three-dimensional image and the reference value coincide, when the coincidence is greater than the
  • the reference value is the positive sample of the image, and the coincidence degree is smaller than the reference value.
  • the U-net network performs Hard Negative Mining processing on the image that has been distinguished from the positive and negative samples, thereby enhancing the ability to discriminate false positive lung nodules.
  • the foregoing embodiment method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be through hardware, but in many cases, the former is better.
  • Implementation Based on such understanding, the technical solution of the present application, which is essential or contributes to the prior art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM, disk,
  • the optical disc includes a number of instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the methods described in various embodiments of the present application.

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

一种肺结节探测方法、应用服务器及计算机可读存储介质。该方法包括：接收终端设备发送的肺部图片（S501）；对所述肺部图片预处理，以将所述肺部图片转换成三维图（S502）；构建U-NET网络（S503）；将所述预处理后的三维图作为所述U-NET网络结构的输入训练所述U-NET网络，以输出凸显肺结节区域的图像（S504）；将所述凸显肺结节区域的图像反馈至终端设备（S505）。上述肺结节探测方法、应用服务器及计算机可读存储介质能够快速定位肺结节区域，减轻医生的压力。

Description

肺结节探测方法、应用服务器及计算机可读存储介质
本申请要求于2018年1月12日提交中国专利局，申请号为201810029621.0、发明名称为“肺结节探测方法、应用服务器及计算机可读存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及医学图像分析技术领域，尤其涉及一种肺结节探测方法、应用服务器及计算机可读存储介质。
背景技术
中国是肺癌全球死亡人数最多的国家，肺癌发病率的不断增加与人口老龄化、城市工业化、农村城市化、环境污染化以及生活方式不良化等因素有关。肺癌早期诊断对患者的资料和预后有着重要的作用。肺癌早期主要表现为肺结节，肺部细胞增生或异物都会导致肺结节的产生，在日益变差的环境中，越来越多的人肺部产生了肺结节。现如今，肺结节已经是目前非常常见的一种症状，很多年轻人都会去医院今早摘除肺结节。然而在临床中，结节的检测一般是由医生根据经验进行肉眼判断，工作强度很大。
发明内容
有鉴于此，本申请提出一种肺结节探测方法、应用服务器及计算机可读存储介质，以解决如何方便地进行结节检测的问题。
首先，为实现上述目的，本申请提出一种肺结节探测方法，该方法包括步骤：
接收终端设备发送的肺部图片；
对所述肺部图片预处理，以将所述肺部图片转换成三维图；
构建U-NET网络，并完成所述U-NET网络的训练；
将所述预处理后的三维图作为所述U-NET网络结构的输入训练所述U-NET网络，以输出凸显肺结节区域的图像；
将所述凸显肺结节区域的图像反馈至终端设备。
此外，为实现上述目的，本申请还提供一种应用服务器，包括存储器、处理器，所述存储器上存储有可在所述处理器上运行的肺结节探测系统，所述肺结节探测系统被所述处理器执行时实现如下步骤：
接收终端设备发送的肺部图片；
对所述肺部图片预处理，以将所述肺部图片转换成三维图；
构建U-NET网络，并完成所述U-NET网络的训练；
将所述预处理后的三维图作为所述已经训练完成的U-NET网络结构的输入，以输出凸显肺结节区域的图像；
将所述凸显肺结节区域的图像反馈至终端设备。
进一步地，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存储有肺结节探测系统，所述肺结节探测系统可被至少一个处理器执行，以使所述至少一个处理器执行如上述的肺结节探测方法的步骤。
相较于现有技术，本申请所提出的肺结节探测方法、应用服务器及计算机可读存储介质，，首先，接收终端设备发送的肺部图片；接着，对所述肺部图片预处理，以将所述肺部图片转换成三维图；进一步地，构建U-NET网络结构；然后，将所述预处理后的三维图作为所述U-NET网络结构的输入训练所述U-NET网络，以输出凸显肺结节区域的图像；最后，将所述凸显肺结节区域的图像反馈至终端设备，以协助医生快速从输出图像在中定位到肺结节，减轻医生的压力。
附图说明
图1是本申请各个实施例一可选的应用环境示意图；
图2是图1中应用服务器一可选的硬件架构的示意图；
图3是本申请肺结节探测系统第一实施例的程序模块示意图；
图4是本申请肺结节探测系统一实施例的U-NET网络结构示意图；
图5是本申请肺结节探测系统第二实施例的程序模块示意图；
图6是本申请肺结节探测方法第一实施例的流程示意图；
图7是本申请肺结节探测方法第二实施例的流程示意图；
附图标记：
终端设备 1
应用服务器 2
网络 3
存储器 11
处理器 12
网络接口 13
肺结节探测系统 200
接收模块 201
转换模块 202
构建模块 203
训练模块 204
反馈模块 205
设置模块 206
判断模块 207
处理模块 208
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
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
需要说明的是，在本申请中涉及“第一”、“第二”等的描述仅用于描述目的，而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外，各个实施例之间的技术方案可以相互结合，但是必须是以本领域普通技术人员能够实现为基础，当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在，也不在本申请要求的保护范围之内。
参阅图1所示，是本申请各个实施例一可选的应用环境示意图。
在本实施例中，本申请可应用于包括，但不仅限于，终端设备1、应用服务器2、网络3的应用环境中。其中，所述终端设备1可以是电子计算机断层扫描设备、核磁共振设备、X线机等医学扫描设备。所述应用服务器2可以是机架式服务器、刀片式服务器、塔式服务器或机柜式服务器等计算设备，该应用服务器2可以是独立的服务器，也可以是多个服务器所组成的服务器集群。所述网络3可以是企业内部网(Intranet)、互联网(Internet)、全球移动通讯系统(Global System of Mobile communication，GSM)、宽带码分多址(Wideband Code Division Multiple Access，WCDMA)、4G网络、5G网络、蓝牙(Bluetooth)、 Wi-Fi、通话网络等无线或有线网络。
其中，所述应用服务器2通过所述网络3分别与一个或多个所述终端设备1通信连接，以进行数据传输和交互。
在本实施例中，所述终端设备1包括医院对应的终端设备1。
参阅图2所示，是本申请应用服务器2一可选的硬件架构的示意图。
本实施例中，所述应用服务器2可包括，但不仅限于，可通过系统总线相互通信连接存储器11、处理器12、网络接口13。需要指出的是，图1仅示出了具有组件11-13的应用服务器2，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。
其中，所述应用服务器2可以是机架式服务器、刀片式服务器、塔式服务器或机柜式服务器等计算设备，该应用服务器2可以是独立的服务器，也可以是多个服务器所组成的服务器集群。
所述存储器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)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，所述存储器11可以是所述应用服务器2的内部存储单元，例如该应用服务器2的硬盘或内存。在另一些实施例中，所述存储器11也可以是所述应用服务器2的外部存储设备，例如该应用服务器2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。当然，所述存储器11还可以既包括所述应用服务器2的内部存储单元也包括其外部存储设备。本实施例中，所述存储器11通常用于存储安装于所述应用服务器2的操作系统和各类应用软件，例如肺结节探测系统200的程序代码等。此外，所述存储器11还可以用于暂时地存储已经输出或者将要输出的各类数据。
所述处理器12在一些实施例中可以是中央处理器(Central Processing  Unit，CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器12通常用于控制所述应用服务器2的总体操作。本实施例中，所述处理器12用于运行所述存储器11中存储的程序代码或者处理数据，例如运行所述的肺结节探测系统200等。
所述网络接口13可包括无线网络接口或有线网络接口，该网络接口13通常用于在所述应用服务器2与其他电子设备之间建立通信连接。
至此，己经详细介绍了本申请相关设备的硬件结构和功能。下面，将基于上述介绍提出本申请的各个实施例。
首先，本申请提出一种肺结节探测系统200。
参阅图3所示，是本申请肺结节探测系统200第一实施例的程序模块图。
本实施例中，所述肺结节探测系统200包括一系列的存储于存储器11上的计算机程序指令，当该计算机程序指令被处理器12执行时，可以实现本申请各实施例的肺结节探测操作。在一些实施例中，基于该计算机程序指令各部分所实现的特定的操作，肺结节探测系统200可以被划分为一个或多个模块。例如，在图3中，所述肺结节探测系统200可以被分割成接收模块201、转换模块202、构建模块203、训练模块204及反馈模块205。其中：
所述接收模块201，用于接收终端设备1发送的肺部图片；
具体地，所述肺部图片为肺部CT图片，所述应用服务器2与终端设备1建立长连接，在本实施例中，所述终端设备1为CT拍摄设备。当终端设备1拍摄完患者的肺部CT图片时，将所述CT图片发送至所述应用服务器2，所述应用服务器2通过所述接收模块201接收所述肺部CT图片。
所述转换模块202，用于对所述肺部图片预处理，以将所述肺部图片转换成三维图。
具体地，由于CT图片一般会有些切片与外界相连接，为了避免外界图片的干扰，所述应用服务器2通过所述转换模块202，首先切除所述肺部CT图片 上下与外界连接的切片；进而，使用-600HU作为阈值将所述肺部CT图片转换成0-1三维图像，在本实施例中，采用大津法(OTSU)计算所述肺部CT的最优阈值；接着，以所述三维图区块的大小即切片达到的中心的平均距离为依据，弥补所有小的凹陷洞；最后将所述三维图的图像像素值修改至[-1200,600]，再缩放至[0,255]。非肺部区域的像素点设置为170。通过处理可以消除后续图片处理中的噪音，例如骨头的亮斑，CT床的金属线条。
所述构建模块203，用于构建U-NET网络结构；
具体地，请一并参阅附图4，附图4为本申请U-NET网络结构示意图。所述三维图作为U-NET网络结构的输入，U-NET网络结构对所述三维图进行处理。具体地，U-NET网络结构通过一个收缩网络以及一个扩张网络，构成了一个U型结构，对图片进行特征提取，U-NET网络一共由23个卷积层构成。收缩网络主要负责下采样的工作，提取高维特征信息，每一次下采样包含两个的3x3的卷积操作，一个2x2的池化操作，通过修正线性单元(rectified linear unit,ReLU)作为激活函数，每一次下采样，图片大小变为原来的1/2，特征数量变为原来的2倍。扩张网络主要负责上采样的工作，每一次上采样包含两个3x3的卷积操作，通过修正线性单元作为激活函数。每一次上采样，图片大小变为原来的2倍，特征数量变为原来的1/2。在上采样操作中，将每一次的输出特征与相映射的收缩网络的特征合并在一起，补全中间丢失的边界信息。
所述训练模块204，用于训练所述U-NET网络，并将所述预处理后的三维图作为已经训练完成的U-NET网络结构的输入，以输出凸显肺结节区域的图像。
具体地，在本实施例中，限于GPU存储大小，所述应用服务器还通过所述转换模块202将所述三维图转换成128*128*128的正方体作为所述U-NET网络结构的输入。为了能够在CT图中快速探测肺结节，70％的输入正方体中保证拥有一个结节，剩下的30％采取随机裁剪，使样本中包含背景样本。大型肺结节相较于小型肺结节而言数量较少。所以在去样本的时候，对于直径大于 30mm，40mm的结节样本扩充2,6倍，最后经过已经训练完成的U-NET网络输出凸显肺结节区域的图像。
所述反馈模块205，用于将所述凸显肺结节区域的图像反馈至终端设备。
具体地，所述应用服务器2通过所述反馈模块205将所述凸显肺结节区域的图像反馈至终端设备1的显示界面，以协助医生快速从输出图像在中定位到肺结节。
通过上述程序模块201-204，本申请所提出的肺结节探测系统200，首先，接收终端设备1发送的肺部图片；接着，对所述肺部图片预处理，以将所述肺部图片转换成三维图；进一步地，构建U-NET网络结构；然后，将所述预处理后的三维图作为所述U-NET网络结构的输入训练所述U-NET网络，以输出凸显肺结节区域的图像；最后，将所述凸显肺结节区域的图像反馈至终端设备，以协助医生快速从输出图像在中定位到肺结节，减轻医生的压力。
参阅图5所示，是本申请肺结节探测系统200第二实施例的程序模块图。本实施例中，所述的肺结节探测系统200除了包括第一实施例中的所述接收模块201、转换模块202、构建模块203、训练模块204及反馈模块205之外，还包括设置模块206、判断模块207及处理模块208。
所述设置模块205，用于设置参考值(ground truth)。
所述判断模块206，用于判断所述三维图像与所述参考值是否重合，当重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本。
具体地，在本实施例中，利用IOU(重合度)决定正负样本，重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本。在实际应用在中，正样本的数量远远小于负样本，这样会出现许多false positive，把其中得分较高的这些false positive当做所谓的Hard negative。
因此，所述应用服务器2还通过所述处理模块208，利用U-net网络对已经 区分正负样本的图像进行难分样本挖掘(Hard Negative Mining)处理，从而加强判别假阳性肺结节的能力。
具体地，所述应用服务器2将所述输入至所述U-NET网络机构的128*128*128的正方体三维图经过4次卷积(convolution)及最大池化(max pooling)之后再经过2次去卷积(deconvolution)得到32*32*32的概率图；进而在每个去卷积(deconvolution)前加入新的分支，每个分支经过相应的去卷积(deconvolution)层也输出32*32*32的概率图；最后将所有概率图同时进行反向传播，以减小相同敏感度下假阳性的大小，凸显肺结节结构。
通过上述程序模块206-208，本申请所提出的肺结节探测系统200，还能够用于设置参考值(ground truth)，并判断所述三维图像与所述参考值是否重合，当重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本，U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理，从而加强判别假阳性肺结节的能力。
此外，本申请还提出一种肺结节探测方法。
参阅图6所示，是本申请肺结节探测方法第一实施例的流程示意图。在本实施例中，根据不同的需求，图6所示的流程图中的步骤的执行顺序可以改变，某些步骤可以省略。
步骤S501，接收终端设备1发送的肺部图片；
具体地，所述肺部图片为肺部CT图片，所述应用服务器2与终端设备1建立长连接，在本实施例中，所述终端设备1为CT拍摄设备。当终端设备1拍摄完患者的肺部CT图片时，将所述CT图片发送至所述应用服务器2，所述应用服务器2接收所述肺部CT图片。
步骤S502，对所述肺部图片预处理，以将所述肺部图片转换成三维图。
具体地，由于CT图片一般会有些切片与外界相连接，为了避免外界图片的干扰，所述应用服务器2，首先切除所述肺部CT图片上下与外界连接的切片； 进而，使用-600HU作为阈值将所述肺部CT图片转换成0-1三维图像，在本实施例中，采用大津法(OTSU)计算所述肺部CT的最优阈值；接着，以所述三维图区块的大小即切片达到的中心的平均距离为依据，弥补所有小的凹陷洞；最后将所述三维图的图像像素值修改至[-1200,600]，再缩放至[0,255]。非肺部区域的像素点设置为170。通过处理可以消除后续图片处理中的噪音，例如骨头的亮斑，CT床的金属线条。
步骤S503，构建U-NET网络结构；
具体地，请一并参阅附图4，附图4为本申请肺结节探测系统200一实施例的U-NET网络结构示意图。所述三维图作为U-NET网络结构的输入，U-NET网络结构对所述三维图进行处理。具体地，U-NET网络结构通过一个收缩网络以及一个扩张网络，构成了一个U型结构，对图片进行特征提取，U-NET网络一共由23个卷积层构成。收缩网络主要负责下采样的工作，提取高维特征信息，每一次下采样包含两个的3x3的卷积操作，一个2x2的池化操作，通过修正线性单元(rectified linear unit,ReLU)作为激活函数，每一次下采样，图片大小变为原来的1/2，特征数量变为原来的2倍。扩张网络主要负责上采样的工作，每一次上采样包含两个3x3的卷积操作，通过修正线性单元作为激活函数。每一次上采样，图片大小变为原来的2倍，特征数量变为原来的1/2。在上采样操作中，将每一次的输出特征与相映射的收缩网络的特征合并在一起，补全中间丢失的边界信息。
步骤S504，将所述预处理后的三维图作为已经训练完成的U-NET网络结构的输入，以输出凸显肺结节区域的图像。
具体地，在本实施例中，限于GPU存储大小，所述应用服务器还通过所述转换模块202将所述三维图转换成128*128*128的正方体作为所述U-NET网络结构的输入。为了能够在CT图中快速探测肺结节，70％的输入正方体中保证拥有一个结节，剩下的30％采取随机裁剪，使样本中包含背景样本。大型肺结节相较于小型肺结节而言数量较少。所以在去样本的时候，对于直径大于 30mm，40mm的结节样本扩充2,6倍，最后经过训练的U-NET网络输出凸显肺结节区域的图像。
步骤S505，将所述凸显肺结节区域的图像反馈至终端设备。
具体地，所述应用服务器2通过所述反馈模块205将所述凸显肺结节区域的图像反馈至终端设备1的显示界面，以协助医生快速从输出图像在中定位到肺结节。
通过上述步骤S501-S505，本申请所提出的肺结节探测方法，首先，接收终端设备1发送的肺部图片；接着，对所述肺部图片预处理，以将所述肺部图片转换成三维图；进一步地，构建U-NET网络结构；然后，将所述预处理后的三维图作为所述U-NET网络结构的输入训练所述U-NET网络，以输出凸显肺结节区域的图像；最后，将所述凸显肺结节区域的图像反馈至终端设备，以协助医生快速从输出图像在中定位到肺结节，减轻医生的压力。
如图7所示，是本申请肺结节探测方法的第二实施例的流程示意图。本实施例中，第一实施例中的将所述预处理后的三维图作为所述U-NET网络结构的输入训练所述U-NET网络，以输出凸显肺结节区域的图像的步骤，具体包括如下步骤：
步骤S601，设置参考值(ground truth)。
步骤S602，判断所述三维图像与所述参考值是否重合，当重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本。
具体地，在本实施例中，利用IOU(重合度)决定正负样本，重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本。在实际应用在中，正样本的数量远远小于负样本，这样会出现许多false positive，把其中得分较高的这些false positive当做所谓的Hard negative。
步骤S603，利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理，从而加强判别假阳性肺结节的能力。
具体地，所述应用服务器2将所述输入至所述U-NET网络机构的128*128*128的正方体三维图经过4次卷积(convolution)及最大池化(max pooling)之后再经过2次去卷积(deconvolution)得到32*32*32的概率图；进而在每个去卷积(deconvolution)前加入新的分支，每个分支经过相应的去卷积(deconvolution)层也输出32*32*32的概率图；最后将所有概率图同时进行反向传播，以减小相同敏感度下假阳性的大小，凸显肺结节结构。
通过上述步骤S601-S603，本申请所提出的肺结节探测方法，还能够用于设置参考值(ground truth)，并判断所述三维图像与所述参考值是否重合，当重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本，U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理，从而加强判别假阳性肺结节的能力。
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。
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，空调器，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. 一种肺结节探测方法，应用于应用服务器，其特征在于，所述方法包括步骤：
    接收终端设备发送的肺部图片；
    对所述肺部图片预处理，以将所述肺部图片转换成三维图；
    构建U-NET网络，并完成所述U-NET网络的训练；
    将所述预处理后的三维图作为训练完成的U-NET网络结构的输入，以输出凸显肺结节区域的图像；
    将所述凸显肺结节区域的图像反馈至终端设备。
  2. 如权利要求1所述的肺结节探测方法，其特征在于，所述对所述肺部图片预处理，以将所述肺部图片转换成三维图的步骤，具体包括：
    切除所述图片上下与外界连接的切片；
    使用-600HU作为阈值将所述肺部图片转换成0-1三维图，其中所述0-1三维图包含有凹陷洞；
    以所述切片达到的中心的平均距离为依据，弥补所有凹陷洞；
    将所述三维图的图像像素值修改至[-1200,600]，再缩放至[0,255]。
  3. 如权利要求1所述的肺结节探测方法，其特征在于，还包括如下步骤：
    将所述三维图转换成128*128*128的正方体三维图。
  4. 如权利要求2所述的肺结节探测方法，其特征在于，还包括如下步骤：
    将所述三维图转换成128*128*128的正方体三维图。
  5. 如权利要求3所述的肺结节探测方法，其特征在于，将所述预处理后的三维图作为所述已经完成训练的U-NET网络结构的输入，以输出凸显肺结节区域的图像的步骤，具体包括如下步骤：
    设置参考值(ground truth)；
    判断三维图像与所述参考值是否重合；
    重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本；
    利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理。
  6. 如权利要求4所述的肺结节探测方法，其特征在于，将所述预处理后的三维图作为所述已经完成训练的U-NET网络结构的输入，以输出凸显肺结节区域的图像的步骤，具体包括如下步骤：
    设置参考值(ground truth)；
    判断三维图像与所述参考值是否重合；
    重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本；
    利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理。
  7. 如权利要求5或6所述的肺结节探测方法，其特征在于，利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理的步骤，具体包括如下步骤：
    将所述输入的128*128*128的正方体三维图经过4次卷积及池化之后再经过2次去卷积得到32*32*32的概率图；
    在每个去卷积前加入新的分支，每个分支经过相应的去卷积层也输出32*32*32的概率图；
    将所有概率图同时进行反向传播，以减小相同敏感度下假阳性的大小，凸显肺结节结构。
  8. 一种应用服务器，其特征在于，所述应用服务器包括存储器、处理器，所述存储器上存储有可在所述处理器上运行的肺结节探测系统，所述肺结节探测系统被所述处理器执行时实现如下步骤：
    接收终端设备发送的肺部图片；
    对所述肺部图片预处理，以将所述肺部图片转换成三维图；
    构建U-NET网络，并完成所述U-NET网络的训练；
    将所述预处理后的三维图作为训练完成的U-NET网络结构的输入，以输出凸显肺结节区域的图像；
    将所述凸显肺结节区域的图像反馈至终端设备。
  9. 如权利要求8所述的应用服务器，其特征在于，所述对所述肺部图片预处理，以将所述肺部图片转换成三维图的步骤，具体包括：
    切除所述图片上下与外界连接的切片；
    使用-600HU作为阈值将所述肺部图片转换成0-1三维图，其中所述0-1三维图包含有凹陷洞；
    以所述切片达到的中心的平均距离为依据，弥补所有凹陷洞；
    将所述三维图的图像像素值修改至[-1200,600]，再缩放至[0,255]。
  10. 如权利要求8所述的应用服务器，其特征在于，还包括如下步骤：
    将所述三维图转换成128*128*128的正方体三维图。
  11. 如权利要求9所述的应用服务器，其特征在于，还包括如下步骤：
    将所述三维图转换成128*128*128的正方体三维图。
  12. 如权利要求10所述的应用服务器，其特征在于，将所述预处理后的三维图作为所述已经完成训练的U-NET网络结构的输入，以输出凸显肺结节区域的图像的步骤，具体包括如下步骤：
    设置参考值(ground truth)；
    判断三维图像与所述参考值是否重合；
    重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本；
    利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理。
  13. 如权利要求11所述的应用服务器，其特征在于，将所述预处理后的 三维图作为所述已经完成训练的U-NET网络结构的输入，以输出凸显肺结节区域的图像的步骤，具体包括如下步骤：
    设置参考值(ground truth)；
    判断三维图像与所述参考值是否重合；
    重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本；
    利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理。
  14. 如权利要求12或13所述的应用服务器，其特征在于，利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理的步骤，具体包括如下步骤：
    将所述输入的128*128*128的正方体三维图经过4次卷积及池化之后再经过2次去卷积得到32*32*32的概率图；
    在每个去卷积前加入新的分支，每个分支经过相应的去卷积层也输出32*32*32的概率图；
    将所有概率图同时进行反向传播，以减小相同敏感度下假阳性的大小，凸显肺结节结构。
  15. 一种计算机可读存储介质，所述计算机可读存储介质存储有肺结节探测系统，所述肺结节探测系统可被至少一个处理器执行，所述肺结节探测系统被所述处理器执行时实现如下步骤：
    接收终端设备发送的肺部图片；
    对所述肺部图片预处理，以将所述肺部图片转换成三维图；
    构建U-NET网络，并完成所述U-NET网络的训练；
    将所述预处理后的三维图作为训练完成的U-NET网络结构的输入，以输出凸显肺结节区域的图像；
    将所述凸显肺结节区域的图像反馈至终端设备。
  16. 如权利要求15所述的计算机可读存储介质，其特征在于，所述对所述肺部图片预处理，以将所述肺部图片转换成三维图的步骤，具体包括：
    切除所述图片上下与外界连接的切片；
    使用-600HU作为阈值将所述肺部图片转换成0-1三维图，其中所述0-1三维图包含有凹陷洞；
    以所述切片达到的中心的平均距离为依据，弥补所有凹陷洞；
    将所述三维图的图像像素值修改至[-1200,600]，再缩放至[0,255]。
  17. 如权利要求15所述的计算机可读存储介质，其特征在于，还包括如下步骤：
    将所述三维图转换成128*128*128的正方体三维图。
  18. 如权利要求16所述的计算机可读存储介质，其特征在于，还包括如下步骤：
    将所述三维图转换成128*128*128的正方体三维图。
  19. 如权利要求17所述的计算机可读存储介质，其特征在于，将所述预处理后的三维图作为所述已经完成训练的U-NET网络结构的输入，以输出凸显肺结节区域的图像的步骤，具体包括如下步骤：
    设置参考值(ground truth)；
    判断三维图像与所述参考值是否重合；
    重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本；
    利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理。
  20. 如权利要求18所述的计算机可读存储介质，其特征在于，将所述预处理后的三维图作为所述已经完成训练的U-NET网络结构的输入，以输出凸显肺结节区域的图像的步骤，具体包括如下步骤：
    设置参考值(ground truth)；
    判断三维图像与所述参考值是否重合；
    重合度大于所述参考值的为图像正样本，重合度小于所述参考值的为图像负样本；
    利用U-net网络对已经区分正负样本的图像进行Hard Negative Mining处理。
PCT/CN2018/090911 2018-01-12 2018-06-12 肺结节探测方法、应用服务器及计算机可读存储介质 WO2019136922A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810029621.0A CN108154468B (zh) 2018-01-12 2018-01-12 肺结节探测方法、应用服务器及计算机可读存储介质
CN201810029621.0 2018-01-12

Publications (1)

Publication Number Publication Date
WO2019136922A1 true WO2019136922A1 (zh) 2019-07-18

Family

ID=62461459

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/090911 WO2019136922A1 (zh) 2018-01-12 2018-06-12 肺结节探测方法、应用服务器及计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN108154468B (zh)
WO (1) WO2019136922A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109285157A (zh) * 2018-07-24 2019-01-29 深圳先进技术研究院 左心室心肌分割方法、装置及计算机可读存储介质
CN109308695A (zh) * 2018-09-13 2019-02-05 镇江纳兰随思信息科技有限公司 基于改进U-net卷积神经网络模型的癌细胞识别方法
CN109271992A (zh) * 2018-09-26 2019-01-25 上海联影智能医疗科技有限公司 一种医学图像处理方法、系统、装置和计算机可读存储介质
CN109242962A (zh) * 2018-09-28 2019-01-18 北京青燕祥云科技有限公司 肺结节的3d显示方法、装置及电子设备
KR102378746B1 (ko) * 2019-08-16 2022-03-25 서울여자대학교 산학협력단 의료 영상에서 딥러닝에 기반한 복부 장기 자동분할 방법

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160308907A1 (en) * 2015-04-20 2016-10-20 Shoelace Wireless, Inc. Systems for improved mobile internet speed and security
CN107016665A (zh) * 2017-02-16 2017-08-04 浙江大学 一种基于深度卷积神经网络的ct肺结节检测方法
WO2017210690A1 (en) * 2016-06-03 2017-12-07 Lu Le Spatial aggregation of holistically-nested convolutional neural networks for automated organ localization and segmentation in 3d medical scans
CN107527069A (zh) * 2017-08-22 2017-12-29 京东方科技集团股份有限公司 图像处理方法、装置、电子设备及计算机可读介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7386165B2 (en) * 2004-02-06 2008-06-10 Siemens Medical Solutions Usa, Inc. System and method for a sparse kernel expansion for a Bayes classifier
CN106940816B (zh) * 2017-03-22 2020-06-09 杭州健培科技有限公司 基于3d全卷积神经网络的ct图像肺结节检测系统
GB201709672D0 (en) * 2017-06-16 2017-08-02 Ucl Business Plc A system and computer-implemented method for segmenting an image

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160308907A1 (en) * 2015-04-20 2016-10-20 Shoelace Wireless, Inc. Systems for improved mobile internet speed and security
WO2017210690A1 (en) * 2016-06-03 2017-12-07 Lu Le Spatial aggregation of holistically-nested convolutional neural networks for automated organ localization and segmentation in 3d medical scans
CN107016665A (zh) * 2017-02-16 2017-08-04 浙江大学 一种基于深度卷积神经网络的ct肺结节检测方法
CN107527069A (zh) * 2017-08-22 2017-12-29 京东方科技集团股份有限公司 图像处理方法、装置、电子设备及计算机可读介质

Also Published As

Publication number Publication date
CN108154468A (zh) 2018-06-12
CN108154468B (zh) 2022-03-01

Similar Documents

Publication Publication Date Title
WO2019136922A1 (zh) 肺结节探测方法、应用服务器及计算机可读存储介质
CN111107783B (zh) 用于计算机辅助分诊的方法和系统
WO2021217857A1 (zh) 切片缺陷检测方法、装置、电子设备及可读存储介质
CN111291825B (zh) 病灶分类模型训练方法、装置、计算机设备和存储介质
CN110969245B (zh) 医学图像的目标检测模型训练方法和装置
Rahman et al. Deep learning–driven automated detection of Covid-19 from radiography images: A comparative analysis
CN110276408B (zh) 3d图像的分类方法、装置、设备及存储介质
WO2021051547A1 (zh) 暴力行为检测方法及系统
US11967181B2 (en) Method and device for retinal image recognition, electronic equipment, and storage medium
CN110969623B (zh) 一种肺部ct多征象自动检测方法、系统、终端及存储介质
WO2019140857A1 (zh) 易损斑块识别方法、应用服务器及计算机可读存储介质
EP3373194B1 (en) Image retrieval apparatus and image retrieval method
US11684333B2 (en) Medical image analyzing system and method thereof
CN111104925A (zh) 图像处理方法、装置、存储介质和电子设备
CN111462201A (zh) 基于新型冠状病毒肺炎ct影像的随访分析系统及方法
EP3847665A1 (en) Determination of a growth rate of an object in 3d data sets using deep learning
CN110895812A (zh) Ct图像的检测方法、装置、存储介质及电子设备
CN110415237A (zh) 皮肤瑕疵检测方法、检测装置、终端设备及可读存储介质
CN111028940B (zh) 肺结节多尺度检测方法、装置、设备及介质
CN114758137A (zh) 超声图像分割方法、装置及计算机可读存储介质
CN115331071A (zh) 一种基于多尺度特征图的结核性脑膜脑炎预测方法及系统
Li et al. Developing an image-based deep learning framework for automatic scoring of the pentagon drawing test
US12026877B2 (en) Device and method for pneumonia detection based on deep learning
CN111192320A (zh) 一种位置信息确定方法、装置、设备和存储介质
WO2019071663A1 (zh) 电子装置、虚拟样本生成方法及存储介质

Legal Events

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

Ref document number: 18900133

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

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

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 12.10.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18900133

Country of ref document: EP

Kind code of ref document: A1