WO2019238104A1 - 实现肺结节图像分类检测的计算机装置及方法 - Google Patents

实现肺结节图像分类检测的计算机装置及方法 Download PDF

Info

Publication number
WO2019238104A1
WO2019238104A1 PCT/CN2019/091190 CN2019091190W WO2019238104A1 WO 2019238104 A1 WO2019238104 A1 WO 2019238104A1 CN 2019091190 W CN2019091190 W CN 2019091190W WO 2019238104 A1 WO2019238104 A1 WO 2019238104A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
lung nodule
lung
images
nodule
Prior art date
Application number
PCT/CN2019/091190
Other languages
English (en)
French (fr)
Inventor
姚育东
钱唯
郑斌
马贺
齐守良
赵明芳
Original Assignee
深圳市前海安测信息技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市前海安测信息技术有限公司 filed Critical 深圳市前海安测信息技术有限公司
Publication of WO2019238104A1 publication Critical patent/WO2019238104A1/zh

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • 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
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30061—Lung
    • G06T2207/30064—Lung nodule

Definitions

  • the present invention relates to the technical field of lung nodule image processing, and in particular, to a computer device and method for implementing classification detection of lung nodule images.
  • the main object of the present invention is to provide a computer device and method for implementing lung nodule image classification detection, which aims to solve the lack of completeness of the existing lung nodule classification detection technology based on lung CT image processing and affect the classification of lung nodules.
  • the problem of low detection accuracy is to provide a computer device and method for implementing lung nodule image classification detection, which aims to solve the lack of completeness of the existing lung nodule classification detection technology based on lung CT image processing and affect the classification of lung nodules. The problem of low detection accuracy.
  • the present invention provides a computer device for classification and detection of lung nodule images, and a computer device for classification and detection of lung nodule images.
  • the computer device is remotely connected to a CT scanner through a communication network, and passes a database. The connection is connected with the image database.
  • the computer device includes a processor adapted to implement various computer program instructions and a memory adapted to store a plurality of computer program instructions, the computer program instructions are loaded by the processor and executed as follows: scan by CT The scanner scans the patient's lung CT image, and performs adaptive morphological segmentation of the lung CT image to obtain the lung nodule image; the lung nodule image is marked into different categories according to the degree of danger, and the lung nodule image of the marked category is stored to In the image database; based on the lung nodule images in the image database, the lung nodule image unit library is established; the distance between the pair of image units in the lung nodule image unit library is calculated to obtain the distance matrix; the distance matrix is clustered and the clustering is determined The number of lung nodule image units for each cluster; calculate each lung nodule CT unit density distribution of lung nodules based on image unit; training and classification of lung nodule risk using lung nodule CT value density based on supervised machine learning model; lung nodule CT based on each lung
  • the step of performing adaptive morphological segmentation on the CT images of the lungs to obtain the lung nodule images includes the following steps: preprocessing the CT images of the lungs to determine the boundary of the two lungs and segmenting the parenchyma of the lungs; Rough segmentation of lung CT images using imaging and anatomical features; fine segmentation of each candidate nodule in the lung CT image after coarse segmentation using the FCM method to obtain basic images of candidate lung nodules; determination of candidate lung nodules Whether the nodules are connected to the pleura and blood vessels; the candidate nodules connected to the pleura and blood vessels are segmented using a circular cut to obtain images of lung nodules.
  • the step of establishing a lung nodule image unit library based on the lung nodule images in the image database includes the following steps: Step A, determining whether the CT value of each pixel of each image unit of the lung nodule image is greater than a predetermined minimum gray The degree value, if it is, is saved to the temporary data set; otherwise, the entire lung nodule image is traversed; step B, repeat step A until all image unit pixels are filtered; step C, randomly select the temporary data Several image units are collected and saved to the lung nodule image unit library.
  • the step of calculating the density distribution characteristics of the lung nodule CT value of each clustered lung nodule image unit includes the following steps: calculating the density distribution level of non-zero pixels of CT values in each lung nodule image one by one, Take the CT value non-zero pixel as the center to extract a small square image of size b ⁇ b, where b is the side length of the small square image; match the small image with each lung nodule image in the lung nodule image unit library Unit, the category of the matched lung nodule image unit will be used as the density distribution level of the test pixel; the entire lung nodule image is matched ergonomically to obtain the CT value density distribution image; class point statistics are collected and normalized in the density distribution image The multidimensional CT value density distribution feature vector is obtained, and the number of dimensions of the CT value density distribution feature vector is equal to the number of clusters.
  • the step of training and classifying the degree of lung nodule risk based on the characteristics of the CT nodal density distribution of the lung nodule based on a supervised machine learning model includes the steps of calculating all the lung nodule images in the image database to obtain the CT value density. Distributing the feature vector to generate a feature set, and dividing the feature set into a training set and a test set; randomly extracting not less than a predetermined proportion of feature data from the feature set as a trained feature set and input a supervised machine learning model for training, and perform parameters Optimization; using the remaining feature data in the feature set except the training set as the test feature set, and inputting the test feature vector into the supervised machine learning model to classify and identify the degree of lung nodule danger.
  • the present invention also provides a method for classifying and detecting lung nodule images, which is applied to a computer device that is remotely connected to a CT scanner through a communication network and connected to an image database through a database connection.
  • the method includes the following steps: scanning a patient's lung CT image with a CT scanner, and adaptively segmenting the lung CT image to obtain a lung nodule image; marking the lung nodule image into different categories according to the degree of danger, and labeling the categories
  • the lung nodule images are stored in the image database; the lung nodule image unit library is established based on the lung nodule images in the image database; the distance matrix between the two image units in the lung nodule image unit library is calculated to obtain the distance matrix; the distance matrix Perform clustering and determine the number of clusters to obtain lung nodule image units of each cluster; calculate the density distribution characteristics of lung nodule CT values for each lung nodule image unit; use lung nodule CT based on a supervised machine learning model Value density distribution characteristics to achieve training and classification
  • the step of performing adaptive morphological segmentation on the CT images of the lungs to obtain the lung nodule images includes the following steps: preprocessing the CT images of the lungs to determine the boundary of the two lungs and segmenting the parenchyma of the lungs; Rough segmentation of lung CT images using imaging and anatomical features; fine segmentation of each candidate nodule in the lung CT image after coarse segmentation using the FCM method to obtain basic images of candidate lung nodules; determination of candidate lung nodules Whether the nodules are connected to the pleura and blood vessels; the candidate nodules connected to the pleura and blood vessels are segmented using a circular cut to obtain images of lung nodules.
  • the step of establishing a lung nodule image unit library based on the lung nodule images in the image database includes the following steps: Step A, determining whether the CT value of each pixel of each image unit of the lung nodule image is greater than a predetermined minimum gray The degree value, if it is, is saved to the temporary data set; otherwise, the entire lung nodule image is traversed; step B, repeat step A until all image unit pixels are filtered; step C, randomly select the temporary data Several image units are collected and saved to the lung nodule image unit library.
  • the step of calculating the density distribution characteristic of the lung nodule CT value density of each clustered lung nodule image unit includes the following steps: calculating the density of non-zero pixels of CT values in each lung nodule image one by one Distribution level, extracting small square images of size b ⁇ b with the CT value non-zero pixels as the center, where b is the side length of the small square image; matching the small image with each lung in the lung nodule image unit library Nodule image unit, the category of the matched lung nodule image unit will be used as the density distribution level of the test pixel; traversely match the entire lung nodule image to obtain the CT value density distribution image; classify the number of points in the density distribution image A normalized multi-dimensional CT value density distribution feature vector is obtained by normalization, and the number of dimensions of the CT value density distribution feature vector is equal to the number of clusters.
  • the step of training and classifying the degree of lung nodule risk based on the characteristics of the CT nodal density distribution of the lung nodule based on a supervised machine learning model includes the steps of calculating all the nodule images in the image database to obtain the CT value density Distributing the feature vector to generate a feature set, and dividing the feature set into a training set and a test set; randomly extracting not less than a predetermined proportion of feature data from the feature set as a trained feature set and input a supervised machine learning model for training, and perform parameters Optimization; using the remaining feature data in the feature set except the training set as the test feature set, and inputting the test feature vector into the supervised machine learning model to classify and identify the degree of lung nodule danger.
  • the computer device and method for classifying and detecting lung nodule images described in the present invention can calculate the CT value density distribution of lung nodule images based on unsupervised clustering and extract the density distribution characteristics of the images.
  • the clustering method is used to analyze the lung nodule data, and the clustering evaluation parameters of the contour index are used to evaluate the clustering effect, which improves the processing efficiency, the degree of adaptation and the robustness of the method.
  • the present invention uses a CT value density distribution calculation method to calculate a CT value density distribution characteristic of a lung nodule image. This feature is used to characterize the denseness of the CT value of the lung nodule, thereby converting the physiology of the lung nodule into an image in the information field. Density distribution characteristics;
  • the present invention can also learn and classify the differences in density distribution characteristics of different lung nodule categories based on a supervised machine learning model, thereby improving the accuracy of lung nodule image classification detection, and has a wide range of applications.
  • FIG. 1 is a schematic block diagram of a preferred embodiment of a computer device for classification and detection of lung nodule images according to the present invention
  • FIG. 2 is a flowchart of a preferred embodiment of a method for classifying and detecting lung nodule images according to the present invention.
  • FIG. 1 is a schematic block diagram of a preferred embodiment of a computer device for implementing lung nodule image classification detection according to the present invention.
  • the computer device 1 is installed with a pulmonary nodule image classification detection system 10.
  • the computer device 1 performs remote information communication with the CT scanner 2 through the communication network 3.
  • the computer device 1 obtains the information from the CT scanner 2.
  • the computer device 1 may be a computing device having data processing and communication functions, such as a personal computer, a mainframe computer, a workstation computer, a server, a cloud platform server, and the like.
  • the CT scanner 2 is installed in a medical institution such as a medical examination institution or a large hospital, and can scan a patient's lung CT image.
  • the communication network 3 may be a wireless network (such as a communication network such as GPRS, WIFI, Bluetooth) or an Internet network (such as a network such as the Internet).
  • the computer device 1 is also connected to an image database 4 through a database connection 5 for storing lung nodule images of each patient.
  • the database connection 5 may be an Open Database Connectivity (ODBC) And Java database connection (Java Data Base Connectivity, JDBC).
  • the computer device 1 for implementing classification detection of lung nodule images includes, but is not limited to, a lung nodule image classification detection system 10, a memory 11 adapted to store a plurality of computer program instructions, and execution of various computer programs An instruction processor 12 and a communication unit 13.
  • the memory 11 may be a read-only memory ROM, a random access memory RAM, an electrically erasable memory EEPROM, a flash memory FLASH, a magnetic disk, or an optical disk.
  • the processor 12 is a central processing unit (CPU), a microcontroller (MCU), a data processing chip, or an information processing unit having a data processing function.
  • the communication unit 13 is a wired or wireless communication interface with a remote communication function, for example, a communication interface supporting communication technologies such as GSM, GPRS, WCDMA, CDMA, WIFI, and Bluetooth.
  • the lung nodule image classification detection system 10 is composed of program modules composed of various computer program instructions, including, but not limited to, a lung nodule image acquisition module 101 and a lung nodule element clustering module. 102.
  • the module referred to in the present invention refers to a series of computer program instruction segments that can be executed by the processor 12 of the computer device 1 and can complete fixed functions, and is stored in the memory 11 of the computer device 1, which will be described in detail below with reference to FIG. 2. Specific functions of each module.
  • FIG. 2 a flowchart of a preferred embodiment of a method for classifying and detecting lung nodule images according to the present invention is shown.
  • various method steps of the method for implementing classification detection of lung nodule images are implemented by a computer software program, and the computer software program is stored in a computer-readable storage medium (for example, memory 11) in the form of computer program instructions.
  • the computer-readable storage medium may include: a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.
  • the computer program instructions can be loaded by a processor (for example, the processor 12) and execute steps S21 to S28 as follows.
  • a CT image of a lung of a patient is scanned by a CT scanner, and an adaptive morphological segmentation of the CT image of the lung is performed to obtain a lung nodule image.
  • the CT scanner 2 scans a lung region of the patient to obtain a lung
  • the CT image and the lung nodule image acquisition module 101 acquires the lung CT image from the CT scanner 2 through the communication unit 13 and performs adaptive morphological segmentation of the lung CT image to obtain the lung nodule image.
  • the lung nodule image acquisition module 101 further includes the following steps before performing adaptive morphological segmentation on the CT images of the lungs: Step 1.
  • Step 2 According to the imaging and anatomical features of the CT image of the lung, coarsely segment the CT image of the lung. For the situation that the left and right lungs are not completely separated after segmentation, an adaptive morphological segmentation is used to obtain a lung nodule image.
  • CAD Computer Aided Diagnosis
  • the main segmentation basis of the present invention for adaptively segmenting lung nodules is: the center of the lung nodule has a high CT value, and the boundary of the lung nodule is an irregular closed or semi-closed curve; step 3, FCM (Fuzzy C -Means) method to finely segment each candidate nodule in the CT image of the lung after coarse segmentation to obtain a basic image of the candidate lung nodule; step 4, determine whether the candidate nodule is connected to the pleura and blood vessels; step 5, respectively Candidate nodules connected to the pleura and blood vessels are segmented using the circular cut method to obtain a lung nodule image, which achieves high-quality segmentation of the adaptive morphology of the nodule image.
  • FCM Fuzzy C -Means
  • Step S22 Mark the lung nodule images into different categories according to the degree of danger, and store the lung nodule images of the marked category in the image database 4.
  • the lung nodule image acquisition module 101 The lung nodule images are marked into different categories, and the lung nodule images of the marked categories are stored in the image database 4, and the marked categories include two types of marks: a parenchymal mark and a suspected lung nodule mark.
  • a lung nodule image unit library is established based on the lung nodule images in the image database.
  • the lung nodule element clustering module 102 establishes a lung nodule image unit library based on the lung nodule images in the image database.
  • the method includes the following steps: Step A, determining whether the CT value of each pixel of each image unit of the lung nodule image is greater than a predetermined minimum gray value, and if so, saving it to a temporary data set; otherwise, traversing the entire lung nodule image Step B, repeat step A until all the image unit pixels have been screened; step C, randomly select a number of image units in the temporary data set and save them to the lung nodule image unit library.
  • Step S24 Calculate the distance between two image units in the lung nodule image unit library to obtain a distance matrix.
  • the lung nodule element clustering module 102 calculates two or two image units in the lung nodule image unit library The distance between them gives the distance matrix.
  • the step S24 includes the following steps: Let u and v be any two image units in the lung nodule image unit library, and sequentially convert u and v from a ⁇ a to 1 ⁇
  • the distance matrix is clustered and the number of clusters is determined to obtain the lung nodule image units of each cluster.
  • the lung nodule primitive clustering module 102 uses an unsupervised clustering algorithm to calculate the distance of the lungs.
  • the matrix is clustered and the number of clusters is determined to obtain lung nodule image units of each cluster.
  • the unsupervised clustering algorithm is a clustering algorithm in existing machine learning technologies, and includes clustering algorithms such as K-Means clustering, hierarchical clustering, t-SNE clustering, and DBSCAN clustering.
  • Step S26 Calculate the density distribution characteristics of the lung nodule CT value of each clustered lung nodule image unit.
  • the CT value density distribution characteristics include the degree of danger of lung nodules, and different types of lung nodules have different image features.
  • the lung nodule feature extraction module 103 calculates the density distribution level of non-zero pixels of CT values in each lung nodule image one by one, and extracts small square images of size b ⁇ b with the non-zero pixels of CT values as the center, where b is a small The side length of the square image; match the small block image with each lung nodule image unit in the lung nodule image unit library, and the category of the matched lung nodule image unit will be used as the density distribution level of the test pixel; iteratively match An image of the entire lung nodule is obtained to obtain a CT value density distribution image; class points are counted and normalized in the density distribution image to obtain a multidimensional CT value density distribution feature vector, and the number of dimensions of the CT value density distribution feature vector is equal to The number of classes.
  • step S27 training and classification of lung nodule risk levels are performed based on the density distribution characteristics of lung nodule CT values based on a supervised machine learning model.
  • the lung nodule risk degree classification module 104 uses lung nodules based on a supervised machine learning model.
  • CT value density distribution features are used to train and classify lung nodules at a dangerous level, including the following steps: all lung nodule images in the image database 4 are obtained from CT value density distribution feature vectors, a feature set is generated, and the feature set is segmented Is the training set and the test set; randomly extract feature data of not less than a predetermined proportion (such as 70%, 80%, or other appropriate percentages) from the feature set as the trained feature set and input a supervised machine learning model for training and parameter optimization ; Use the remaining feature data in the feature set except the training set as the test feature set, and enter the test feature vector into the above machine learning model to classify and identify the lung nodule risk level.
  • a predetermined proportion such as 70%, 80%, or other appropriate percentages
  • the lung nodule CT value density distribution of each lung nodule image is used to eliminate false positive lung nodule images and retain suspected lung nodule images.
  • the lung nodule risk classification module 104 according to each lung Density distribution of lung nodule CT values in the nodule image excludes false positive lung nodule images and retains suspected lung nodule images for doctors to use as a reference for whether or not the patient's lungs have lesions, thereby improving lung cancer screening, detection, Accuracy of diagnosis.
  • the computer device and method for classifying and detecting lung nodule images of the present invention can calculate the CT value density distribution of lung nodule images based on unsupervised clustering and extract the density distribution characteristics of the images, and use common clustering methods for lung nodule data. Analyze and use the clustering evaluation parameters of the contour index to evaluate the clustering effect, improve the processing efficiency, the degree of adaptation, and the robustness of the method.
  • the present invention uses a CT value density distribution calculation method to calculate a CT value density distribution characteristic of a lung nodule image. This feature is used to characterize the denseness of the CT value of the lung nodule, thereby converting the physiology of the lung nodule into an image in the information field. Density distribution characteristics; The present invention also learns and classifies the differences in density distribution characteristics of different lung nodule categories based on a supervised machine learning model, thereby improving the accuracy of lung nodule image classification detection, and has a wide range of applications.
  • the program may be stored in a computer-readable storage medium.
  • the storage medium may include a read-only memory, a random access memory, Disk or CD, etc.
  • the computer device and method for classifying and detecting lung nodule images described in the present invention can calculate the CT value density distribution of lung nodule images based on unsupervised clustering and extract the density distribution characteristics of the images.
  • the clustering method is used to analyze the lung nodule data, and the clustering evaluation parameters of the contour index are used to evaluate the clustering effect, which improves the processing efficiency, the degree of adaptation and the robustness of the method.
  • the present invention uses a CT value density distribution calculation method to calculate a CT value density distribution characteristic of a lung nodule image. This feature is used to characterize the denseness of the CT value of the lung nodule, thereby converting the physiology of the lung nodule into an image in the information field. Density distribution characteristics;
  • the present invention can also learn and classify the differences in density distribution characteristics of different lung nodule categories based on a supervised machine learning model, thereby improving the accuracy of lung nodule image classification detection, and has a wide range of applications.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Radiology & Medical Imaging (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Quality & Reliability (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Analysis (AREA)

Abstract

本发明提供一种实现肺结节图像分类检测的计算机装置及方法，该方法包括步骤：通过CT扫描仪扫描患者的肺部CT图像，对肺部CT图像进行自适应形态分割得到肺结节图像；将肺结节图像标记成不同类别，并将标记的肺结节图像存储至图像数据库中；基于图像数据库中肺结节图像建立肺结节图像单元库；计算肺结节图像单元库中两两图像单元之间的距离得到距离矩阵；对距离矩阵进行聚类；计算每个肺结节图像单元的肺结节CT值密度分布特征；基于有监督机器学习模型使用CT值密度分布特征实现肺结节危险程度的训练和分类;根据每个肺结节图像的CT值密度分布剔除假阳性的肺结节图像。本发明提高了肺结节图像分类检测的准确性，具有广泛的应用范围。

Description

实现肺结节图像分类检测的计算机装置及方法 技术领域
本发明涉及肺结节图像处理的技术领域，尤其涉及一种实现肺结节图像分类检测的计算机装置及方法。
 背景技术
近年来，肺结节临床诊断过程中肺部CT图像的应用越来越广泛，通过肺部CT图像分析肺结节的危险程度具有非常大的意义，尤其对早期肺结节的研究，研制一种高精度、临床意义大、鲁棒性强的肺结节图像分类检测系统变得越来越重要。由于肺结节的复杂性，目前基于肺部CT图像处理的肺结节分类检测技术缺乏完备性。现有的技术对肺结节的分类不具备实用性，影响对肺结节分类检测精度不高，在实际应用中缺乏实用性。
 技术问题
本发明的主要目的在于提供一种实现肺结节图像分类检测的计算机装置及方法，旨在解决现有基于肺部CT图像处理的肺结节分类检测技术缺乏完备性而影响对肺结节分类检测精度不高的问题。
 技术解决方案
为实现上述目的，本发明提供一种实现肺结节图像分类检测的计算机装置，一种实现肺结节图像分类检测的计算机装置，该计算机装置通过通信网络远程连接有CT扫描仪，并通过数据库连接与图像数据库连接，该计算机装置包括适于实现各种计算机程序指令的处理器以及适于存储多条计算机程序指令的存储器，所述计算机程序指令由处理器加载并执行如下步骤：通过CT扫描仪扫描患者的肺部CT图像，对肺部CT图像进行自适应形态分割得到肺结节图像；按照危险程度将肺结节图像标记成不同的类别，并将标记类别的肺结节图像存储至图像数据库中；基于图像数据库中肺结节图像建立肺结节图像单元库；计算肺结节图像单元库中两两图像单元之间的距离得到距离矩阵；对距离矩阵进行聚类并确定聚类数量来获得每个聚类的肺结节图像单元；计算每个肺结节图像单元的肺结节CT值密度分布特征；基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类;根据每个肺结节图像的肺结节CT值密度分布剔除假阳性的肺结节图像保留疑似肺结节图像。
进一步地，所述对肺部CT图像进行自适应形态分割得到肺结节图像的步骤包括如下步骤：对肺部CT图像进行预处理确定双肺的边界进行肺实质分割；根据肺部CT图像的影像学和解剖学特征对肺部CT图像进行粗分割；应用FCM方法对粗分割后的肺部CT图像中每个候选结节进行精细分割，获得候选肺结节的基本图像；判断候选肺结节是否与胸膜和血管相连；分别与胸膜和血管相连的候选结节应用圆切法进行分割得到肺结节图像。
进一步地，所述基于图像数据库中肺结节图像建立肺结节图像单元库的步骤包括如下步骤：步骤A，判断肺结节图像的各图像单元像素点的CT值是否都大于预定的最小灰度值，如果是，则保存至暂存数据集；否则，遍历整个肺结节图像；步骤B，重复步骤A直到所有图像单元像素点都筛选完毕；步骤C，随机挑选所述的暂存数据集中的若干图像单元，并保存至肺结节图像单元库。
进一步地，所述计算每个聚类的肺结节图像单元的肺结节CT值密度分布特征的步骤包括如下步骤：逐个计算每个肺结节图像中CT值非零像素的密度分布级别，以该CT值非零像素为中心提取大小为b×b的小方块图像，其中b为小方块图像的边长；匹配所述小块图像与肺结节图像单元库中每一个肺结节图像单元，匹配到的肺结节图像单元的类别将作为测试像素的密度分布级别；遍历地匹配整个肺结节图像，得到其CT值密度分布图像；对密度分布图像中进行类别点数统计并归一化得到一个多维CT值密度分布特征向量，所述的CT值密度分布特征向量的维度数等于聚类数量。
进一步地，所述基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类的步骤包括如下步骤：计算图像数据库中的所有肺结节图像得到CT值密度分布特征向量生成特征集，并将该特征集分割为训练集和测试集；从特征集中随机提取不少于预定比例的特征数据作为训练的特征集输入有监督机器学习模型进行训练，并进行参数优化；将特征集中除训练集外的其余特征数据作为测试特征集，将测试特征向量输入所述有监督机器学习模型进行肺结节危险程度分类识别。
另一方面，本发明还提供一种实现肺结节图像分类检测的方法，应用于计算机装置中，该计算机装置通过通信网络远程连接有CT扫描仪，并通过数据库连接与图像数据库连接，该方法包括如下步骤：通过CT扫描仪扫描患者的肺部CT图像，对肺部CT图像进行自适应形态分割得到肺结节图像；按照危险程度将肺结节图像标记成不同的类别，并将标记类别的肺结节图像存储至图像数据库中；基于图像数据库中肺结节图像建立肺结节图像单元库；计算肺结节图像单元库中两两图像单元之间的距离得到距离矩阵；对距离矩阵进行聚类并确定聚类数量来获得每个聚类的肺结节图像单元；计算每个肺结节图像单元的肺结节CT值密度分布特征；基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类;根据每个肺结节图像的肺结节CT值密度分布剔除假阳性的肺结节图像保留疑似肺结节图像。
进一步地，所述对肺部CT图像进行自适应形态分割得到肺结节图像的步骤包括如下步骤：对肺部CT图像进行预处理确定双肺的边界进行肺实质分割；根据肺部CT图像的影像学和解剖学特征对肺部CT图像进行粗分割；应用FCM方法对粗分割后的肺部CT图像中每个候选结节进行精细分割，获得候选肺结节的基本图像；判断候选肺结节是否与胸膜和血管相连；分别与胸膜和血管相连的候选结节应用圆切法进行分割得到肺结节图像。
进一步地，所述基于图像数据库中肺结节图像建立肺结节图像单元库的步骤包括如下步骤：步骤A，判断肺结节图像的各图像单元像素点的CT值是否都大于预定的最小灰度值，如果是，则保存至暂存数据集；否则，遍历整个肺结节图像；步骤B，重复步骤A直到所有图像单元像素点都筛选完毕；步骤C，随机挑选所述的暂存数据集中的若干图像单元，并保存至肺结节图像单元库。
进一步地，进一步地，所述计算每个聚类的肺结节图像单元的肺结节CT值密度分布特征的步骤包括如下步骤：逐个计算每个肺结节图像中CT值非零像素的密度分布级别，以该CT值非零像素为中心提取大小为b×b的小方块图像，其中b为小方块图像的边长；匹配所述小块图像与肺结节图像单元库中每一个肺结节图像单元，匹配到的肺结节图像单元的类别将作为测试像素的密度分布级别；遍历地匹配整个肺结节图像，得到其CT值密度分布图像；对密度分布图像中进行类别点数统计并归一化得到一个多维CT值密度分布特征向量，所述的CT值密度分布特征向量的维度数等于聚类数量。
进一步地，所述基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类的步骤包括如下步骤：计算图像数据库中的所有肺结节图像得到CT值密度分布特征向量生成特征集，并将该特征集分割为训练集和测试集；从特征集中随机提取不少于预定比例的特征数据作为训练的特征集输入有监督机器学习模型进行训练，并进行参数优化；将特征集中除训练集外的其余特征数据作为测试特征集，将测试特征向量输入所述有监督机器学习模型进行肺结节危险程度分类识别。
 有益效果
相较于现有技术，本发明所述实现肺结节图像分类检测的计算机装置及方法能够基于无监督聚类计算肺结节图像的CT值密度分布并提取该图像的密度分布特征，使用常用的聚类方法对肺结节数据进行分析，并利用轮廓指数的聚类评价参数评估聚类效果，提升处理效率、自适应程度以及方法的鲁棒性。本发明使用CT值密度分布计算方式计算肺结节图像的CT值密度分布特征，该特征用于表征肺结节CT值的稠密度，从而将肺结节的生理性转换为信息领域中的图像密度分布特征；本发明还能够基于有监督机器学习模型对不同肺结节类别密度分布特征的差异进行学习、分类，从而提高肺结节图像分类检测的准确性，具有广泛的应用范围。
 附图说明
图1是本发明实现肺结节图像分类检测的计算机装置的优选实施例的方框示意图；
图2是本发明实现肺结节图像分类检测的方法优选实施例的流程图。
本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
 本发明的实施方式
为更进一步阐述本发明为达成预定发明目的所采取的技术手段及功效，以下结合附图及较佳实施例，对本发明的具体实施方式、结构、特征及其功效，详细说明如下。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
参照图1所示，图1是本发明实现肺结节图像分类检测的计算机装置的优选实施例的方框示意图。在本实施例中，所述计算机装置1安装有肺结节图像分类检测系统10，该计算机装置1通过通信网络3与CT扫描仪2进行远程信息通讯，例如计算机装置1从CT扫描仪2获取患者的肺部CT图像。所述计算机装置1可以为个人计算机、大型计算机、工作站计算机、服务器、云平台服务器等具有数据处理和通信功能的计算装置。
所述CT扫描仪2设置在健康检查机构、大型医院等医疗机构中，能够扫描患者的肺部CT图像。所述通信网络3可以为无线网路（例如GPRS、WIFI、Bluetooth等通信网路）或互联网际网络（例如Internet等网络）。所述计算机装置1还通过数据库连接5与图像数据库4连接，所述图像数据库4用于存储每一个患者的肺结节图像，所述数据库连接5可以为开放数据库连接（Open Database Connectivity，ODBC）以及Java数据库连接（Java Data Base Connectivity，JDBC）。
在本实例中，所述实现肺结节图像分类检测的计算机装置1包括，但不仅限于，肺结节图像分类检测系统10、适于存储多条计算机程序指令的存储器11、执行各种计算机程序指令的处理器12以及通信单元13。所述存储器11可以为一种只读存储器ROM，随机存储器RAM、电可擦写存储器EEPROM、快闪存储器FLASH、磁盘或光盘等。所述处理器12为一种中央处理器（CPU）、微控制器（MCU）、数据处理芯片、或者具有数据处理功能的信息处理单元。所述通信单元13为一种具有远程通讯功能的有线或无线通讯接口，例如，支持GSM、GPRS、WCDMA、CDMA、WIFI、蓝牙（Bluetooth）等通讯技术的通讯接口。
在本实施例中，所述肺结节图像分类检测系统10由各种计算机程序指令组成的程序模块组成，包括但不局限于，肺结节图像获取模块101、肺结节图元聚类模块102、肺结节特征提取模块103、以及肺结节危险程度分类模块104。本发明所称的模块是指一种能够被计算机装置1的处理器12执行并且能够完成固定功能的一系列计算机程序指令段，其存储在计算机装置1的存储器11中，以下结合图2具体说明每一个模块的具体功能。
参考图2所示，是本发明实现肺结节图像分类检测的方法优选实施例的流程图。在本实施例中，所述实现肺结节图像分类检测的方法的各种方法步骤通过计算机软件程序来实现，该计算机软件程序以计算机程序指令的形式存储于计算机可读存储介质（例如存储器11）中，计算机可读存储介质可以包括：只读存储器、随机存储器、磁盘或光盘等，所述计算机程序指令能够被处理器（例如处理器12）加载并执行如下步骤S21至步骤S28。
步骤S21，通过CT扫描仪扫描患者的肺部CT图像，对肺部CT图像进行自适应形态分割得到肺结节图像；在本实施例中，CT扫描仪2扫描患者的肺部区域获取肺部CT图像，肺结节图像获取模块101通过通信单元13从CT扫描仪2获取肺部CT图像，并对肺部CT图像进行自适应形态分割得到肺结节图像。在本实施例中，肺结节图像获取模块101对肺部CT图像进行自适应形态分割之前还包括如下步骤：步骤1，利用肺部CAD（Computer Aided Diagnosis）方法对肺部CT图像进行预处理，确定双肺的边界进行肺实质分割（Lung Segmentation）；在肺实质分割的过程中，由于左、右肺前内侧之间的纵隔区比较窄，当存在部分容积效应现象时，这个区域与肺区的对比度往往很低，造成左右肺区不能被成功分割。步骤2，根据肺部CT图像的影像学和解剖学特征对肺部CT图像进行粗分割，针对分割后出现左右肺未完全分离的情况，采用自适应形态分割得到肺结节图像。本发明采用自适应形态分割肺结节的主要分割依据是：肺结节中心具有较高CT值，而肺结节的边界是不规则的闭合或半闭合曲线；步骤3，应用FCM（Fuzzy C-Means）方法对粗分割后的肺部CT图像中每个候选结节进行精细分割，获得候选肺结节的基本图像；步骤4，判断候选结节是否与胸膜和血管相连；步骤5，分别与胸膜和血管相连的候选结节应用圆切法进行分割得到肺结节图像，实现结节图像的自适应形态的优质分割。
步骤S22，按照危险程度将肺结节图像标记成不同的类别，并将标记类别的肺结节图像存储至图像数据库4中；在本实施例中，肺结节图像获取模块101按照危险程度将肺结节图像标记成不同的类别，并将标记类别的肺结节图像存储至图像数据库4中，所述标记类别包括肺实质标记和疑似肺结节标记两种标记。
步骤S23，基于图像数据库中肺结节图像建立肺结节图像单元库；在本实施例中，肺结节图元聚类模块102基于图像数据库中肺结节图像建立肺结节图像单元库，包括如下步骤：步骤A，判断肺结节图像的各图像单元像素点的CT值是否都大于预定的最小灰度值，如果是，则保存至暂存数据集，否则，遍历整个肺结节图像；步骤B，重复执行步骤A直到所有图像单元像素点都筛选完毕；步骤C，随机挑选所述的暂存数据集中的若干图像单元，并保存至肺结节图像单元库。
步骤S24，计算肺结节图像单元库中两两图像单元之间的距离得到距离矩阵；在本实施例中，肺结节图元聚类模块102计算肺结节图像单元库中两两图像单元之间的距离得到距离矩阵。在较佳的实施方式中，所述的步骤S24包括以下步骤：设u、v为肺结节图像单元库中任意两个图像单元，依次将u、v由大小为a×a转换为1×2a的一维向量，分别为u＝{u i|i＝0,1,2,…,2a}、和v＝{v i|i＝0,1,2,…,2a}，其中a为图像单元的边长；计算5组两个图像单元之间的距离为d得到五个距离矩阵D j，j＝1、2、3、4、5，设max，min分别为D j中的最大值和最小值，目标选择的距离矩阵为D i；其中：D i＝D j，i＝max-min的最大值。
步骤S25，对距离矩阵进行聚类并确定聚类数量来获得每个聚类的肺结节图像单元；在本实施例中，肺结节图元聚类模块102使用无监督聚类算法对距离矩阵进行聚类并确定聚类数量来获得每个聚类的肺结节图像单元。所述无监督聚类算法是现有机器学习技术中的一种聚类算法，包括K-Means 聚类、分层聚类、 t-SNE聚类、DBSCAN聚类等聚类算法。
步骤S26，计算每个聚类的肺结节图像单元的肺结节CT值密度分布特征。在本实施例中，所述CT值密度分布特征包括肺结节的危险程度，不同的肺结节类别具有不同的图像特征。肺结节特征提取模块103逐个计算每个肺结节图像中CT值非零像素的密度分布级别，以该CT值非零像素为中心提取大小为b×b的小方块图像，其中b为小方块图像的边长；匹配所述小块图像与肺结节图像单元库中每一个肺结节图像单元，匹配到的肺结节图像单元的类别将作为测试像素的密度分布级别；遍历地匹配整个肺结节图像，得到其CT值密度分布图像；对密度分布图像中进行类别点数统计并归一化得到一个多维CT值密度分布特征向量，所述CT值密度分布特征向量的维度数等于聚类数量。
步骤S27，基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类；具体地，肺结节危险程度分类模块104基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类，包括如下步骤：将图像数据库4中的所有肺结节图像都得到CT值密度分布特征向量，生成特征集，并将该特征集分割为训练集和测试集；从特征集中随机提取不少于预定比例（例如70％、80％或其他合适百分比）的特征数据作为训练的特征集输入有监督机器学习模型进行训练，并进行参数优化；将特征集中除训练集外的其余特征数据作为测试特征集，将测试特征向量输入上述的机器学习模型进行肺结节危险程度分类识别。
步骤S28，根据每个肺结节图像的肺结节CT值密度分布剔除假阳性的肺结节图像保留疑似肺结节图像；本实施例中，肺结节危险程度分类模块104根据每个肺结节图像的肺结节CT值密度分布剔除假阳性的肺结节图像保留疑似肺结节图像，以供医生对患者的肺部是否发生病变作为辅助参考依据，从而提高肺癌筛查、检测、诊断的准确率。
本发明实现肺结节图像分类检测的计算机装置及方法能够基于无监督聚类计算肺结节图像的CT值密度分布并提取该图像的密度分布特征，使用常用的聚类方法对肺结节数据进行分析，并利用轮廓指数的聚类评价参数评估聚类效果，提升处理效率、自适应程度以及方法的鲁棒性。本发明使用CT值密度分布计算方式计算肺结节图像的CT值密度分布特征，该特征用于表征肺结节CT值的稠密度，从而将肺结节的生理性转换为信息领域中的图像密度分布特征；本发明还基于有监督机器学习模型对不同肺结节类别密度分布特征的差异进行学习、分类，从而提高肺结节图像分类检测的准确性，具有广泛的应用范围。
本领域技术人员可以理解，上述实施方式中各种方法的全部或部分步骤可以通过相关程序指令完成，该程序可以存储于计算机可读存储介质中，存储介质可以包括：只读存储器、随机存储器、磁盘或光盘等。
以上仅为本发明的优选实施例，并非因此限制本发明的专利范围，凡是利用本发明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，均同理包括在本发明的专利保护范围内。
 工业实用性
相较于现有技术，本发明所述实现肺结节图像分类检测的计算机装置及方法能够基于无监督聚类计算肺结节图像的CT值密度分布并提取该图像的密度分布特征，使用常用的聚类方法对肺结节数据进行分析，并利用轮廓指数的聚类评价参数评估聚类效果，提升处理效率、自适应程度以及方法的鲁棒性。本发明使用CT值密度分布计算方式计算肺结节图像的CT值密度分布特征，该特征用于表征肺结节CT值的稠密度，从而将肺结节的生理性转换为信息领域中的图像密度分布特征；本发明还能够基于有监督机器学习模型对不同肺结节类别密度分布特征的差异进行学习、分类，从而提高肺结节图像分类检测的准确性，具有广泛的应用范围。

Claims (10)

  1. 一种实现肺结节图像分类检测的计算机装置，该计算机装置通过通信网络远程连接有CT扫描仪，并通过数据库连接与图像数据库连接，其特征在于，该计算机装置包括适于实现各种计算机程序指令的处理器以及适于存储多条计算机程序指令的存储器，所述计算机程序指令由处理器加载并执行如下步骤：
    通过CT扫描仪扫描患者的肺部CT图像，对肺部CT图像进行自适应形态分割得到肺结节图像；
    按照危险程度将肺结节图像标记成不同的类别，并将标记类别的肺结节图像存储至图像数据库中；
    基于图像数据库中肺结节图像建立肺结节图像单元库；
    计算肺结节图像单元库中两两图像单元之间的距离得到距离矩阵；
    对距离矩阵进行聚类并确定聚类数量来获得每个聚类的肺结节图像单元；
    计算每个聚类的肺结节图像单元的肺结节CT值密度分布特征；
    基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类;
    根据每个肺结节图像的肺结节CT值密度分布剔除假阳性的肺结节图像保留疑似肺结节图像。
  2. 如权利要求1所述的实现肺结节图像分类检测的计算机装置，其特征在于，所述对肺部CT图像进行自适应形态分割得到肺结节图像的步骤包括如下步骤：
    对肺部CT图像进行预处理确定双肺的边界进行肺实质分割；
    根据肺部CT图像的影像学和解剖学特征对肺部CT图像进行粗分割；
    应用FCM方法对粗分割后的肺部CT图像中每个候选结节进行精细分割，获得候选肺结节的基本图像；
    判断候选肺结节是否与胸膜和血管相连；
    分别与胸膜和血管相连的候选结节应用圆切法进行分割得到肺结节图像。
  3. 如权利要求1所述的实现肺结节图像分类检测的计算机装置，其特征在于，所述基于图像数据库中肺结节图像建立肺结节图像单元库的步骤包括如下步骤：
    步骤A，判断肺结节图像的各图像单元像素点的CT值是否都大于预定的最小灰度值，如果是，则保存至暂存数据集；否则，遍历整个肺结节图像；
    步骤B，重复步骤A直到所有图像单元像素点都筛选完毕；
    步骤C，随机挑选所述的暂存数据集中的若干图像单元，并保存至肺结节图像单元库。
  4. 如权利要求1所述的实现肺结节图像分类检测的计算机装置，其特征在于，所述计算每个聚类的肺结节图像单元的肺结节CT值密度分布特征的步骤包括如下步骤：
    逐个计算每个肺结节图像中CT值非零像素的密度分布级别，以该CT值非零像素为中心提取大小为b×b的小方块图像，其中b为小方块图像的边长；
    匹配所述小块图像与肺结节图像单元库中每一个肺结节图像单元，匹配到的肺结节图像单元的类别将作为测试像素的密度分布级别；
    遍历地匹配整个肺结节图像，得到其CT值密度分布图像；
    对密度分布图像中进行类别点数统计并归一化得到一个多维CT值密度分布特征向量，所述的CT值密度分布特征向量的维度数等于聚类数量。
  5. 如权利要求1所述的实现肺结节图像分类检测的计算机装置，其特征在于，所述基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类的步骤包括如下步骤：
    计算图像数据库中的所有肺结节图像得到CT值密度分布特征向量生成特征集，并将该特征集分割为训练集和测试集；
    从特征集中随机提取不少于预定比例的特征数据作为训练的特征集输入有监督机器学习模型进行训练，并进行参数优化；
    将特征集中除训练集外的其余特征数据作为测试特征集，将测试特征向量输入所述有监督机器学习模型进行肺结节危险程度分类识别。
  6. 一种实现肺结节图像分类检测的方法，应用于计算机装置中，该计算机装置通过通信网络远程连接有CT扫描仪，并通过数据库连接与图像数据库连接，其特征在于，该方法包括如下步骤：
    通过CT扫描仪扫描患者的肺部CT图像，对肺部CT图像进行自适应形态分割得到肺结节图像；
    按照危险程度将肺结节图像标记成不同的类别，并将标记类别的肺结节图像存储至图像数据库中；
    基于图像数据库中肺结节图像建立肺结节图像单元库；
    计算肺结节图像单元库中两两图像单元之间的距离得到距离矩阵；
    对距离矩阵进行聚类并确定聚类数量来获得每个聚类的肺结节图像单元；
    计算每个聚类的肺结节图像单元的肺结节CT值密度分布特征；
    基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类;
    根据每个肺结节图像的肺结节CT值密度分布剔除假阳性的肺结节图像保留疑似肺结节图像。
  7. 如权利要求6所述的实现肺结节图像分类检测的方法，其特征在于，所述对肺部CT图像进行自适应形态分割得到肺结节图像的步骤包括如下步骤：
    对肺部CT图像进行预处理确定双肺的边界进行肺实质分割；
    根据肺部CT图像的影像学和解剖学特征对肺部CT图像进行粗分割；
    应用FCM方法对粗分割后的肺部CT图像中每个候选结节进行精细分割，获得候选肺结节的基本图像；
    判断候选肺结节是否与胸膜和血管相连；
    分别与胸膜和血管相连的候选结节应用圆切法进行分割得到肺结节图像。
  8. 如权利要求6所述的实现肺结节图像分类检测的方法，其特征在于，所述基于图像数据库中肺结节图像建立肺结节图像单元库的步骤包括如下步骤：
    步骤A，判断肺结节图像的各图像单元像素点的CT值是否都大于预定的最小灰度值，如果是，则保存至暂存数据集；否则，遍历整个肺结节图像；
    步骤B，重复步骤A直到所有图像单元像素点都筛选完毕；
    步骤C，随机挑选所述的暂存数据集中的若干图像单元，并保存至肺结节图像单元库。
  9. 如权利要求6所述的实现肺结节图像分类检测的方法，其特征在于，所述计算每个聚类的肺结节图像单元的肺结节CT值密度分布特征的步骤包括如下步骤：
    逐个计算每个肺结节图像中CT值非零像素的密度分布级别，以该CT值非零像素为中心提取大小为b×b的小方块图像，其中b为小方块图像的边长；
    匹配所述小块图像与肺结节图像单元库中每一个肺结节图像单元，匹配到的肺结节图像单元的类别将作为测试像素的密度分布级别；
    遍历地匹配整个肺结节图像，得到其CT值密度分布图像；
    对密度分布图像中进行类别点数统计并归一化得到一个多维CT值密度分布特征向量，所述的CT值密度分布特征向量的维度数等于聚类数量。
  10. 如权利要求6所述的实现肺结节图像分类检测的方法，其特征在于，所述基于有监督机器学习模型使用肺结节CT值密度分布特征实现肺结节危险程度的训练和分类的步骤包括如下步骤：
    计算图像数据库中的所有肺结节图像得到CT值密度分布特征向量生成特征集，并将该特征集分割为训练集和测试集；
    从特征集中随机提取不少于预定比例的特征数据作为训练的特征集输入有监督机器学习模型进行训练，并进行参数优化；
    将特征集中除训练集外的其余特征数据作为测试特征集，将测试特征向量输入所述有监督机器学习模型进行肺结节危险程度分类识别。
PCT/CN2019/091190 2018-06-16 2019-06-14 实现肺结节图像分类检测的计算机装置及方法 WO2019238104A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810624285.4A CN110610472A (zh) 2018-06-16 2018-06-16 实现肺结节图像分类检测的计算机装置及方法
CN201810624285.4 2018-06-16

Publications (1)

Publication Number Publication Date
WO2019238104A1 true WO2019238104A1 (zh) 2019-12-19

Family

ID=68842718

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/091190 WO2019238104A1 (zh) 2018-06-16 2019-06-14 实现肺结节图像分类检测的计算机装置及方法

Country Status (2)

Country Link
CN (1) CN110610472A (zh)
WO (1) WO2019238104A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111815591A (zh) * 2020-06-29 2020-10-23 郑州大学 基于ct图像的肺结节检测方法
CN112116603A (zh) * 2020-09-14 2020-12-22 中国科学院大学宁波华美医院 一种基于多任务学习的肺结节假阳性筛查方法
CN112686898A (zh) * 2021-03-15 2021-04-20 四川大学 一种基于自监督学习的放疗靶区自动分割方法
CN116883996A (zh) * 2023-09-01 2023-10-13 武汉互创联合科技有限公司 基于旋转等变网络的胚胎发育阶段预测与质量评估系统

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113808090A (zh) * 2021-09-08 2021-12-17 中国科学院深圳先进技术研究院 一种磁共振图像处理方法、终端设备及计算机存储介质
CN115294401B (zh) * 2022-08-23 2023-04-14 北京医准智能科技有限公司 一种基于ct图像的分类方法、装置、设备及存储介质
CN116934757B (zh) * 2023-09-18 2023-11-21 电子科技大学 一种用于肺结节假阳性删减的方法、设备及存储介质

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106250701A (zh) * 2016-08-10 2016-12-21 华东理工大学 计算机软件系统中实现肺结节危险程度分类的系统和方法
CN108010013A (zh) * 2017-11-03 2018-05-08 天津大学 一种肺ct图像肺结节检测方法

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106250701A (zh) * 2016-08-10 2016-12-21 华东理工大学 计算机软件系统中实现肺结节危险程度分类的系统和方法
CN108010013A (zh) * 2017-11-03 2018-05-08 天津大学 一种肺ct图像肺结节检测方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
CAO, LEI: "Analysis and Identification of Pulmonary Nodule Images", INFORMATION & TECHNOLOGY, CHINA DOCTORAL DISSERTATIONS FULL-TEXT DATABASE, 15 January 2010 (2010-01-15), pages 30 - 42 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111815591A (zh) * 2020-06-29 2020-10-23 郑州大学 基于ct图像的肺结节检测方法
CN111815591B (zh) * 2020-06-29 2024-02-27 郑州大学 基于ct图像的肺结节检测方法
CN112116603A (zh) * 2020-09-14 2020-12-22 中国科学院大学宁波华美医院 一种基于多任务学习的肺结节假阳性筛查方法
CN112686898A (zh) * 2021-03-15 2021-04-20 四川大学 一种基于自监督学习的放疗靶区自动分割方法
CN116883996A (zh) * 2023-09-01 2023-10-13 武汉互创联合科技有限公司 基于旋转等变网络的胚胎发育阶段预测与质量评估系统
CN116883996B (zh) * 2023-09-01 2023-12-08 武汉互创联合科技有限公司 基于旋转等变网络的胚胎发育阶段预测与质量评估系统

Also Published As

Publication number Publication date
CN110610472A (zh) 2019-12-24

Similar Documents

Publication Publication Date Title
WO2019238104A1 (zh) 实现肺结节图像分类检测的计算机装置及方法
Shi et al. Automatic road crack detection using random structured forests
Alsmadi et al. Fish recognition based on robust features extraction from size and shape measurements using neural network
WO2018108129A1 (zh) 用于识别物体类别的方法及装置、电子设备
EP3171332A1 (en) Methods and systems for inspecting goods
CN111931811B (zh) 一种基于超像素图像相似度的计算方法
Pan et al. Cell detection in pathology and microscopy images with multi-scale fully convolutional neural networks
Song et al. Using HOG-LBP features and MMP learning to recognize imaging signs of lung lesions
CN108062749B (zh) 肛提肌裂孔的识别方法、装置和电子设备
CN110188763B (zh) 一种基于改进图模型的图像显著性检测方法
Liu et al. A fully automatic segmentation algorithm for CT lung images based on random forest
Wen et al. Virus image classification using multi-scale completed local binary pattern features extracted from filtered images by multi-scale principal component analysis
CN112183237A (zh) 基于颜色空间自适应阈值分割的白细胞自动分类方法
Hao et al. Vp-detector: A 3d multi-scale dense convolutional neural network for macromolecule localization and classification in cryo-electron tomograms
Fang et al. Images crack detection technology based on improved K-means algorithm
CN111488811B (zh) 人脸识别方法、装置、终端设备及计算机可读介质
Rathore et al. CBISC: a novel approach for colon biopsy image segmentation and classification
Xie et al. Automatic neuron tracing in volumetric microscopy images with anisotropic path searching
CN110853030A (zh) 生物反应器病毒感染细胞质量评价方法
Zhou et al. Automatic segmentation of liver from CT scans with CCP–TSPM algorithm
CN109191452A (zh) 一种基于主动学习的腹腔ct图像腹膜转移自动标记方法
He et al. FRFCM clustering segmentation method for medical MR image feature diagnosis
CN111753723B (zh) 一种基于密度校准的指纹识别方法及装置
Wang et al. Current situation and review of image segmentation
Liu et al. Automatic Lung Parenchyma Segmentation of CT Images Based on Matrix Grey Incidence.

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: 19818844

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: 19818844

Country of ref document: EP

Kind code of ref document: A1