WO2022006917A1

WO2022006917A1 - Artificial intelligence-based lung magnetic resonance image recognition apparatus and method

Info

Publication number: WO2022006917A1
Application number: PCT/CN2020/101519
Authority: WO
Inventors: 高伟明; 葛新科; 姚育东; 钱唯; 郑斌; 齐守良; 张红治; 周亮; 陈琦
Original assignee: 深圳市安测健康信息技术有限公司
Priority date: 2019-07-05
Filing date: 2020-07-11
Publication date: 2022-01-13
Also published as: CN112184613A

Abstract

The present invention provides an artificial intelligence-based lung magnetic resonance image recognition apparatus and method. The method comprises the steps of: acquiring lung magnetic resonance images of target objects using a magnetic resonance imaging device; performing lung region segmentation on the lung magnetic resonance images, and respectively registering preselected lung lobe atlases to the lung magnetic resonance images; fusing segmentation boundaries formed by the lung lobe atlases to generate segmentation results of the lung magnetic resonance images; registering the lung magnetic resonance images to labeled and positioned structural images, and screening feature lung regions corresponding to the structural images of sequences; extracting, for each feature lung region, a lung lobe texture feature most relevant to physiological properties of a lung region pathology; recognizing alveoli and vascular contours of lung lobes from the lung magnetic resonance images; and using the lung lobe texture feature, alveolus information, and vascular contour information as multi-dimensional pathological features of the lung lobes. The present invention can improve the sensitivity, specificity and examination accuracy of the detection of lung abnormalities caused by lung diseases.

Description

基于人工智能的磁共振肺部影像识别装置及方法Magnetic resonance lung image recognition device and method based on artificial intelligence

技术领域technical field

本发明涉及基于人工智能的影像处理技术领域，尤其涉及一种基于人工智能的磁共振肺部影像识别装置及方法。The present invention relates to the technical field of image processing based on artificial intelligence, in particular to an artificial intelligence-based magnetic resonance lung image recognition device and method.

背景技术Background technique

随着经济的快速发展和大气环境污染的日益严重，肺癌已经成为发病率和死亡率增长最快，对人群健康和生命威胁最大的恶性肿瘤。磁共振成像(Magnetic Resonance Imaging，简称MRI)，是利用核磁共振(Nuclear Magnetic Resonance，简称NMR)原理，依据所释放的能量在物质内部不同结构环境中不同的衰减，进而通过外加梯度磁场检测所发射出的电磁波，了解构成该物质原子核的位置和种类，据此呈现物体内部结构影像的技术。MRI因为其无放射、非侵入等诸多无法比拟的优越性在医学诊断和研究中占据着重要位置，并已经在人类健康和公共卫生事业中发挥了巨大的作用。With the rapid economic development and the increasingly serious air pollution, lung cancer has become the malignant tumor with the fastest growing morbidity and mortality, and the greatest threat to human health and life. Magnetic Resonance Imaging Resonance Imaging, referred to as MRI), uses the principle of nuclear magnetic resonance (Nuclear Magnetic Resonance, referred to as NMR), according to the different attenuation of the released energy in different structural environments inside the material, and then detects the emitted electromagnetic waves by applying a gradient magnetic field to understand The position and type of the nuclei that make up the substance, and the technology that presents an image of the internal structure of the object. MRI occupies an important position in medical diagnosis and research because of its incomparable advantages such as non-radiation and non-invasiveness, and has played a huge role in human health and public health.

MRI是一种常用的医学断层成像方法，它利用磁共振现象从人体中获得电磁信号，并重建出人体信息。这种技术利用核磁共振原理，依据所释放的能量在物质内部不同结构环境中不同的衰减，通过外加梯度磁场检测所发射出的电磁波，即可得知构成这一物体原子核的位置和种类，据此可以绘制成物体内部的结构影像。人体三分之二的重量为水分，而且人体内器官和组织中的水分并不相同，很多疾病的病理过程会导致水分形态的变化，即可由磁共振影像反应出来，但目前医生对磁共振肺部影像的应用主要是根据影像所反映的肺部结构的病变进行主观判别与分类。MRI is a commonly used medical tomography method, which uses the magnetic resonance phenomenon to obtain electromagnetic signals from the human body and reconstruct the human body information. This technology uses the principle of nuclear magnetic resonance, according to the different attenuation of the released energy in different structural environments inside the material, and detects the emitted electromagnetic waves by applying a gradient magnetic field to know the position and type of the nuclei that constitute the object. This can be drawn as an image of the structure inside the object. Two-thirds of the weight of the human body is water, and the water in the organs and tissues of the human body is not the same. The pathological process of many diseases will lead to changes in the form of water, which can be reflected by magnetic resonance imaging. The application of external imaging is mainly based on the subjective judgment and classification of the lesions of the lung structure reflected by the imaging.

现有磁共振肺部影像的阅片及分类完全依赖于医生视觉观察，在经验的基础上进行主观的判断，对病灶的发现过度依赖经验，缺乏可量化的标准。此外，医生在读写磁共振肺部影像报告时的主观性也比较强，缺乏统一的量化标准和话术体系，对电子病历标准化和大数据挖掘造成了瓶颈。The reading and classification of the existing magnetic resonance lung images completely rely on the doctor's visual observation, and make subjective judgments based on experience. The discovery of lesions relies too much on experience and lacks quantifiable standards. In addition, doctors are also highly subjective when reading and writing MRI lung image reports, and lack of a unified quantitative standard and language system, which has created a bottleneck for electronic medical record standardization and big data mining.

技术问题technical problem

本发明的主要目的在于提供一种基于人工智能的磁共振肺部影像识别装置及方法，能够同时以肺叶纹理特征、肺泡及血管轮廓等信息为指标识别磁共振肺部影像，提高对肺部疾病所引起的肺部异常检测的敏感性、特异性和检查准确性。The main purpose of the present invention is to provide a magnetic resonance lung image recognition device and method based on artificial intelligence, which can simultaneously use information such as lung lobe texture features, alveoli and blood vessel contours as indicators to identify magnetic resonance lung images, so as to improve the detection of lung diseases. Sensitivity, specificity, and testing accuracy for the detection of induced pulmonary abnormalities.

技术解决方案technical solutions

为实现上述目的，本发明提供一种基于人工智能的磁共振肺部影像识别装置，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：选取若干健康人和不同肺结节疾病患者作为目标对象，利用磁共振成像设备获取目标对象的磁共振肺部影像；从肺叶图谱库筛选出与磁共振肺部影像匹配度最高的若干套肺叶图谱；对目标对象的磁共振肺部影像进行肺区分割，并将预选的肺叶图谱分别配准到磁共振肺部影像；对多个肺叶图谱所形成的分割边界进行融合生成磁共振肺部影像的分割结果；将磁共振肺部影像配准到已标注和定位的结构影像，并结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区；对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征；从磁共振肺部影像识别肺叶的肺泡及血管轮廓；将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征。In order to achieve the above purpose, the present invention provides a magnetic resonance lung image recognition device based on artificial intelligence, comprising a processor suitable for implementing various computer program instructions and a memory suitable for storing a plurality of computer program instructions. The instructions are loaded by the processor and the following steps are performed: selecting a number of healthy people and patients with different pulmonary nodule diseases as target objects, and using magnetic resonance imaging equipment to obtain magnetic resonance lung images of the target objects; several sets of lung lobe atlases with the highest matching degree of external images; lung area segmentation is performed on the magnetic resonance lung images of the target object, and the preselected lung lobe atlases are respectively registered to the magnetic resonance lung images; the segmentation formed by multiple lung lobe atlases The boundary is fused to generate the segmentation results of the magnetic resonance lung images; the magnetic resonance lung images are registered to the marked and localized structural images, and the corresponding features of each sequence are selected by combining the corresponding changes between the structural images of different sequences. Lung area: Mining the multi-dimensional information of each characteristic lung area, and extracting the lobe texture features most relevant to the physiological characteristics of the lung area’s diseased state; Identifying the alveoli and blood vessel contours of the lung lobe from the magnetic resonance lung image; The features, alveolar information and blood vessel contour information are converted into the corresponding k-space data, and the large signal data in the k-space data is screened as the multi-dimensional pathological features of the lung lobes.

进一步地，所述计算机程序指令由处理器加载还执行如下步骤：针对目标对象的磁共振肺部影像选定一个标准肺叶模板，对每一个目标对象相同层面的多序列磁共振肺部影像进行配准。Further, the computer program instructions are loaded by the processor and also perform the following steps: selecting a standard lung lobe template for the magnetic resonance lung image of the target object, and matching the multi-sequence magnetic resonance lung images of the same level of each target object. allow.

进一步地，所述对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征的步骤包括：根据肺叶纹理的构成建立相似性测度算法，针对目标对象的磁共振肺部影像执行肺叶结构标注和肺叶纹理提取，再与肺叶图谱库进行比对，以相似度最高的肺叶纹理类别作为对该目标对象的纹理识别结果。Further, the step of mining the multi-dimensional information of each characteristic lung region and extracting the lung lobe texture features most relevant to the physiological characteristics of the diseased state of the lung region includes: establishing a similarity measurement algorithm according to the composition of the lung lobe texture, aiming at the following steps: The MRI lung image of the target object is subjected to lung lobe structure labeling and lung lobe texture extraction, and then compared with the lung lobe atlas library, and the lobe texture category with the highest similarity is used as the texture recognition result of the target object.

进一步地，所述从磁共振肺部影像识别肺叶的肺泡及血管轮廓的步骤包括：通过中值滤波对磁共振肺部影像进行滤波降噪；通过阈值分割与纹理分割相结合的方式提取肺区的肺泡信息及血管轮廓信息。Further, the step of recognizing the alveoli and blood vessel contours of the lung lobe from the magnetic resonance lung image includes: filtering and denoising the magnetic resonance lung image by median filtering; extracting the lung region by a combination of threshold segmentation and texture segmentation. alveolar information and vessel contour information.

进一步地，所述将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征的步骤包括如下步骤：将去除非肺部影像信息但包含纹理特征、肺泡信息及血管轮廓信息的影像进行灰度反转，通过傅里叶变换将影像数据转化为对应的k空间数据；筛选并识别k空间数据中大信号的复数数据a+bi，按照

的大小筛选大信号，将获得的k空间数据中大信号数据作为肺叶的多维病理特征。 Further, the step of converting lung lobe texture features, alveolar information and blood vessel contour information into corresponding k-space data, and screening large signal data in the k-space data as the multi-dimensional pathological features of the lung lobes includes the following steps: removing non-lungs. Grayscale inversion is performed on images with partial image information but contains texture features, alveolar information and blood vessel contour information, and the image data is converted into corresponding k-space data through Fourier transform; the complex data of large signals in the k-space data are screened and identified. a+bi, according to

The size of the large signal was screened, and the large signal data in the obtained k-space data was used as the multi-dimensional pathological feature of the lung lobe.

另一方面，本发明还提供一种基于人工智能的磁共振肺部影像识别方法，应用于计算机装置中，该计算机装置连接有磁共振成像设备，该方法包括如下步骤：选取若干健康人和不同肺结节疾病患者作为目标对象，利用磁共振成像设备获取目标对象的磁共振肺部影像；从肺叶图谱库筛选出与磁共振肺部影像匹配度最高的若干套肺叶图谱；对目标对象的磁共振肺部影像进行肺区分割，并将预选的肺叶图谱分别配准到磁共振肺部影像；对多个肺叶图谱所形成的分割边界进行融合生成磁共振肺部影像的分割结果；将磁共振肺部影像配准到已标注和定位的结构影像，并结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区；对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征；从磁共振肺部影像识别肺叶的肺泡及血管轮廓；将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征。On the other hand, the present invention also provides an artificial intelligence-based magnetic resonance lung image recognition method, which is applied to a computer device, the computer device is connected with magnetic resonance imaging equipment, and the method includes the following steps: selecting a number of healthy people and different people Patients with pulmonary nodule disease are used as the target object, and the magnetic resonance imaging equipment of the target object is used to obtain the magnetic resonance lung image of the target object; several sets of lung lobe atlases with the highest matching degree with the magnetic resonance lung image are selected from the lung lobe atlas library; The lung area is segmented from the resonant lung image, and the preselected lung lobe atlases are registered to the magnetic resonance lung image respectively; the segmentation boundaries formed by multiple lung lobe atlases are fused to generate the segmentation result of the magnetic resonance lung image; The lung images are registered to the marked and localized structural images, and the corresponding changes between the structural images of different sequences are combined to screen out the characteristic lung areas corresponding to each sequence; the multi-dimensional information of each characteristic lung area is mined, and the Extract the lung lobe texture features that are most relevant to the physiological characteristics of the lung disease state; identify the alveoli and blood vessel contours of the lung lobe from the magnetic resonance lung image; convert the lung lobe texture features, alveolar information and blood vessel contour information into the corresponding k-space data, And screen the large signal data in the k-space data as the multi-dimensional pathological features of lung lobes.

进一步地，所述的基于人工智能的磁共振肺部影像识别方法还包括如下步骤：针对目标对象的磁共振肺部影像选定一个标准肺叶模板，对每一个目标对象相同层面的多序列磁共振肺部影像进行配准。Further, the artificial intelligence-based magnetic resonance lung image recognition method further includes the following steps: selecting a standard lung lobe template for the magnetic resonance lung image of the target object, and performing multi-sequence magnetic resonance imaging on the same level of each target object. Lung image registration.

进一步地，所述将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征的步骤包括：将去除非肺部影像信息但包含纹理特征、肺泡信息及血管轮廓信息的影像进行灰度反转，通过傅里叶变换将影像数据转化为对应的k空间数据；筛选并识别k空间数据中大信号的复数数据a+bi，按照

的大小筛选大信号，将获得的k空间数据中大信号数据作为肺叶的多维病理特征。 Further, the step of converting lung lobe texture features, alveolar information and blood vessel contour information into corresponding k-space data, and screening large signal data in the k-space data as the multi-dimensional pathological features of the lung lobe includes: removing non-pulmonary images. Grayscale inversion is performed on images that contain texture features, alveolar information and blood vessel contour information, and the image data is converted into corresponding k-space data through Fourier transform; complex data a+ of large signals in k-space data is screened and identified. bi, according to

有益效果beneficial effect

相较于现有技术，本发明基于磁共振影像的肺叶病理特征的识别及提取技术，该技术允许对肺部组织的多种重要性质进行非侵入的同步量化检测，并且为肺叶各区域复杂的功能、生理及物理变化和形态变化提供了智能化的检测与识别方法。与人工阅片相比，改善了人体对肺部生理异常描述的话术体系，提高了以磁共振影像为基础的对人体肺部生理、物理和功能特性的检查精度，同时以肺叶纹理特征、肺泡及血管轮廓等信息为指标，提高了对肺部疾病所引起的肺部异常检测的敏感性、特异性和检查准确性，为医生对肺癌的诊断治疗提供医学指导。Compared with the prior art, the present invention is based on the identification and extraction technology of pulmonary lobe pathological features based on magnetic resonance images. This technology allows non-invasive simultaneous quantitative detection of various important properties of lung tissue, and is complex Functional, physiological and physical changes and morphological changes provide intelligent detection and identification methods. Compared with manual reading, it improves the human body's description of lung physiological abnormalities, and improves the inspection accuracy of human lung physiological, physical and functional characteristics based on magnetic resonance imaging. And blood vessel contour and other information as indicators, improve the sensitivity, specificity and inspection accuracy of lung abnormality detection caused by lung diseases, and provide medical guidance for doctors to diagnose and treat lung cancer.

附图说明Description of drawings

图1是本发明基于人工智能的磁共振肺部影像识别装置的较佳实施例的结构方框示意图。FIG. 1 is a schematic structural block diagram of a preferred embodiment of a magnetic resonance lung image recognition device based on artificial intelligence of the present invention.

图2是本发明基于人工智能的磁共振肺部影像识别方法较佳实施例的方法流程图。FIG. 2 is a method flow chart of a preferred embodiment of the artificial intelligence-based magnetic resonance lung image recognition method of the present invention.

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。The realization, functional characteristics and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments.

本发明的实施方式Embodiments of the present invention

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。In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific embodiments, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

参照图1所示，图1是本发明基于人工智能的磁共振肺部影像识别装置的较佳实施例的结构示意图。在本实施例中，所述基于人工智能的磁共振肺部影像识别装置1包括，但不仅限于，适于存储各种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以及显示器13。所述存储器11和显示器13均通过电连接线与所述处理器12进行电气连接，并通过数据总线与处理器12进行数据传输连接。所述处理器12能够调用存储在所述存储器11中的基于人工智能的磁共振肺部影像识别程序10，并执行该磁共振肺部影像识别程序10从磁共振成像设备2输入的磁共振肺部影像数据，并对磁共振肺部影像进行识别。所述磁共振肺部影像识别装置1可以为安装有本发明所述基于人工智能的磁共振肺部影像识别程序10的个人计算机、笔记本电脑、服务器等计算机装置。Referring to FIG. 1 , FIG. 1 is a schematic structural diagram of a preferred embodiment of an artificial intelligence-based magnetic resonance lung image recognition device of the present invention. In this embodiment, the artificial intelligence-based magnetic resonance lung image recognition device 1 includes, but is not limited to, a memory 11 suitable for storing various computer program instructions, a processor 12 for executing various computer program instructions, and a display 13. Both the memory 11 and the display 13 are electrically connected to the processor 12 through electrical connecting lines, and are connected to the processor 12 for data transmission through a data bus. The processor 12 can call the artificial intelligence-based magnetic resonance lung image recognition program 10 stored in the memory 11, and execute the magnetic resonance lung image recognition program 10 input from the magnetic resonance imaging device 2. Image data and identification of magnetic resonance lung images. The magnetic resonance lung image recognition device 1 may be a computer device such as a personal computer, a notebook computer, and a server installed with the artificial intelligence-based magnetic resonance lung image recognition program 10 of the present invention.

在本实施例中，所述磁共振肺部影像识别装置1连接有磁共振成像设备2，该磁共振成像设备2能够扫描目标对象的人体肺部得到不同序列（例如T1、T2或DTI序列）的影像磁共振肺部影像。所述磁共振肺部影像识别装置1能够从磁共振成像设备2获取不同序列的磁共振肺部影像，并通过处理器12执行磁共振肺部影像识别程序10对不同序列的磁共振肺部影像进行处理，利用人工智能方法对不同序列的磁共振肺部影像快速准确地识别肺叶纹理特征、肺泡信息及血管轮廓信息等多维病理特征。In this embodiment, the magnetic resonance lung image recognition device 1 is connected with a magnetic resonance imaging device 2, which can scan the human lungs of the target object to obtain different sequences (eg T1, T2 or DTI sequences) Imaging Magnetic Resonance Lung Imaging. The magnetic resonance lung image recognition device 1 can acquire magnetic resonance lung images of different sequences from the magnetic resonance imaging device 2, and execute the magnetic resonance lung image recognition program 10 through the processor 12 to perform magnetic resonance lung images of different sequences. Processing, using artificial intelligence methods to quickly and accurately identify multi-dimensional pathological features such as lung lobe texture features, alveolar information and blood vessel contour information from different sequences of magnetic resonance lung images.

在本实施例中，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）、磁性存储器、磁盘、光盘等。所述存储器11在一些实施例中可以是所述磁共振肺部影像识别装置1的内部存储单元，例如该基于人工智能的磁共振肺部影像识别装置1的硬盘、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或光盘等。所述存储器11在另一些实施例中也可以是磁共振肺部影像识别装置1的外部存储设备，例如该基于人工智能的磁共振肺部影像识别装置1上配备的插接式硬盘，智能存储卡（Smart Media Card, SMC），安全数字（Secure Digital, SD）卡，闪存卡（Flash Card）等。进一步地，所述存储器11还可以既包括磁共振肺部影像识别装置1的内部存储单元也包括外部存储设备。所述存储器11不仅可以用于存储安装于磁共振肺部影像识别装置1的应用软件及各类数据，例如存储磁共振肺部影像识别程序10的程序代码等，还可以用于暂时地存储已经输出或者将要输出的数据。In this embodiment, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (eg, SD or DX memory, etc.), magnetic memory, etc. , disk, CD, etc. The memory 11 may be an internal storage unit of the magnetic resonance lung image recognition device 1 in some embodiments, such as a hard disk, a read-only memory ROM, a random access memory of the artificial intelligence-based magnetic resonance lung image recognition device 1. RAM, electrically erasable memory EEPROM, flash memory FLASH or optical disk, etc. The memory 11 may also be an external storage device of the magnetic resonance lung image recognition device 1 in other embodiments, for example, a plug-in hard disk equipped on the artificial intelligence-based magnetic resonance lung image recognition device 1, intelligent storage Card (Smart Media Card, SMC), Secure Digital, SD) card, flash card (Flash Card), etc. Further, the memory 11 may also include both an internal storage unit of the magnetic resonance lung image recognition apparatus 1 and an external storage device. The memory 11 can not only be used to store the application software and various data installed in the magnetic resonance lung image recognition device 1, such as the program code of the magnetic resonance lung image recognition program 10, etc., but also can be used to temporarily store Output or data to be output.

在本实施例中，所述处理器12在一些实施例中可以是一中央处理器（Central Processing Unit, CPU）、控制器、微控制器、微处理器或其他数据处理芯片，用于调用并运行存储器11中存储的程序代码或处理数据，例如执行基于人工智能的磁共振肺部影像识别程序10等。所述显示器13可以为触摸显示屏也可以为通用的LED显示屏，能够显示识别出的肺叶纹理特征、肺泡信息及血管轮廓信息等多维病理特征。In this embodiment, the processor 12 may be a central processing unit (Central Processing Unit) in some embodiments. Processing Unit, CPU), controller, microcontroller, microprocessor or other data processing chip for calling and running the program code or processing data stored in the memory 11, such as performing artificial intelligence-based magnetic resonance lung image recognition Procedure 10 et al. The display 13 can be a touch display screen or a general LED display screen, and can display the identified multi-dimensional pathological features such as texture features of lung lobes, alveolar information, and blood vessel contour information.

可选地，在其他实施例中，所述基于人工智能的磁共振肺部影像识别程序10还可以被分割为一个或者多个模块，一个或者多个模块被存储于存储器11中，并由一个或多个处理器（本实施例为处理器12）所执行以完成本发明，本发明所称的模块是指能够完成特定功能的一系列计算机程序指令段，用于描述基于人工智能的磁共振肺部影像识别程序10在所述磁共振肺部影像识别装置1中的执行过程。Optionally, in other embodiments, the artificial intelligence-based magnetic resonance lung image recognition program 10 can also be divided into one or more modules, one or more modules are stored in the memory 11, and are composed of one or more modules. or multiple processors (the processor 12 in this embodiment) are executed to complete the present invention. The module referred to in the present invention refers to a series of computer program instruction segments that can perform specific functions, and is used to describe artificial intelligence-based magnetic resonance imaging. The execution process of the lung image recognition program 10 in the magnetic resonance lung image recognition apparatus 1 .

在本实施例中，所述基于人工智能的磁共振肺部影像识别程序10由多条计算机程序指令组成的程序模块组成，包括但不局限于，磁共振肺部影像获取模块101、磁共振肺部影像处理模块102、特征肺区筛选模块103、肺叶特征识别模块104以及肺叶特征输出模块105。本发明所称的模块是指一种能够被所述磁共振肺部影像识别装置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，其存储在所述磁共振肺部影像识别装置1的存储器11中。In this embodiment, the artificial intelligence-based magnetic resonance lung image recognition program 10 is composed of program modules composed of a plurality of computer program instructions, including but not limited to a magnetic resonance lung image acquisition module 101, a magnetic resonance lung image The external image processing module 102 , the characteristic lung region screening module 103 , the lung lobe feature identification module 104 and the lung lobe feature output module 105 . 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 magnetic resonance lung image recognition device 1 and can perform fixed functions, which are stored in the magnetic resonance lung image in the memory 11 of the identification device 1 .

所述磁共振肺部影像获取模块101用于选取若干健康人和不同肺结节疾病患者作为目标对象，利用磁共振成像设备2获取目标对象的磁共振肺部影像；在目标对象呼吸空气的情况下，利用磁共振成像设备对目标对象肺部区域进行磁共振成像获取肺部的氢质子影像。质子成像过程中由于肺部氢质子含量非常少，采用自旋回波（Spin Echo，简称SE）类型的序列与其他类型序列相比有利于提高信噪比，得到更好的影像。在本实施例中，所述磁共振肺部影像包括所有磁共振的序列所能产生的影像以及这些序列产生影像的任意组合，在所指的任意序列影像的组合中，单序列T1、T2或DTI影像是所允许的最小影像集合。The magnetic resonance lung image acquisition module 101 is used to select a number of healthy people and patients with different pulmonary nodule diseases as target objects, and use the magnetic resonance imaging device 2 to acquire the magnetic resonance lung images of the target object; in the case of the target object breathing air Next, use magnetic resonance imaging equipment to perform magnetic resonance imaging on the lung area of the target object to obtain hydrogen proton images of the lungs. During proton imaging, spin echo (Spin echo) was used because the hydrogen proton content in the lungs was very low. Compared with other types of sequences, the sequence of Echo (SE) type is beneficial to improve the signal-to-noise ratio and obtain better images. In this embodiment, the magnetic resonance lung images include images that can be generated by all sequences of magnetic resonance and any combination of images generated by these sequences. DTI images are the smallest set of images allowed.

所述磁共振肺部影像处理模块102用于从肺叶图谱库筛选出与磁共振肺部影像匹配度最高的若干套肺叶图谱；在本实施例中，在已有的肺叶图谱库中进行图谱预选，通过图谱匹配算法筛选与磁共振肺部影像匹配度最高的10～20套肺叶图谱，作为分析磁共振肺部影像的参考标准。所述肺叶图谱库建立的步骤：基于大量的肺部影像样本及其病例信息，利用人工智能的方法对不同性别、年龄以及肺结节疾病所对应的肺叶纹理进行分类，再进行人工的鉴别与确认，最后形成不同人群的肺叶图谱数据，该肺叶图谱数据将随着样本量的增加而不断扩张，建立对未知肺结节疾病的肺叶纹理识别的标准库基础。The magnetic resonance lung image processing module 102 is used to screen out several sets of lung lobe atlases with the highest matching degree with the magnetic resonance lung images from the lung lobe atlas library; in this embodiment, atlas preselection is performed in the existing lung lobe atlas library , 10 to 20 sets of lung lobe atlases with the highest matching degree with MRI lung images were screened by the atlas matching algorithm, which was used as the reference standard for analyzing MRI lung images. The steps of establishing the lung lobe atlas library: based on a large number of lung image samples and their case information, use artificial intelligence to classify the lung lobe textures corresponding to different genders, ages and pulmonary nodule diseases, and then perform manual identification and identification. Confirmed, and finally formed the lung lobe atlas data of different populations, the lung lobe atlas data will continue to expand with the increase of the sample size, and establish a standard library basis for the identification of lung lobe texture for unknown pulmonary nodular diseases.

所述特征肺区筛选模块103用于对磁共振肺部影像进行肺区分割，并将预选的肺叶图谱分别配准到磁共振肺部影像；在本实施例中，针对T1、T2或DTI序列的影像进行基于多肺叶图谱对磁共振肺部影像的分割，将上述预选出来的肺叶图谱分别配准到磁共振肺部影像。针对目标对象的磁共振肺部影像选定一个标准肺叶模板，对每一个目标对象的磁共振肺部影像相同层面的多序列扫描影像进行配准，配准同一个体相同层面的多序列扫描影像以消除头动效应，之后对其进行影像标准化操作以实现不同个体间的影像横向对比。The characteristic lung area screening module 103 is used to perform lung area segmentation on the magnetic resonance lung image, and register the preselected lung lobe atlases to the magnetic resonance lung image respectively; in this embodiment, for T1, T2 or DTI sequences The MRI lung images are segmented based on the multi-lobe atlas, and the pre-selected lung lobe atlases are registered to the MRI lung images respectively. A standard lung lobe template is selected for the magnetic resonance lung image of the target object, and the multi-sequence scan images of the same level of the magnetic resonance lung image of each target object are registered. The head motion effect was eliminated, and then image normalization was performed on it to achieve horizontal comparison of images between different individuals.

所述特征肺区筛选模块103还用于对多个肺叶图谱所形成的分割边界进行融合生成磁共振肺部影像的肺区分割结果。在本实施例中，在对多个肺叶图谱所形成的分割边界进行融合，最后生成磁共振肺部影像的肺区分割结果，该肺区分割结果最大的特色就是对每一个肺区进行了标注和定位。The characteristic lung region screening module 103 is further configured to fuse the segmentation boundaries formed by multiple lung lobe atlases to generate a lung region segmentation result of the magnetic resonance lung image. In this embodiment, the segmentation boundaries formed by multiple lung lobe atlases are fused, and finally the lung area segmentation result of the magnetic resonance lung image is generated. The biggest feature of the lung area segmentation result is that each lung area is labeled and positioning.

所述特征肺区筛选模块103还用于将不同序列的磁共振肺部影像配准到已标注和定位的结构影像，同时结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区。在本实施例中，已完成标注和定位的结构影像是肺叶纹理构建的基础，针对每一个结构标注的区域进行深度的信息挖掘便是肺叶纹理提取；磁共振不同序列的成像结果承载了肺叶纹理的不同特性，不同序列所产生影像的对比度对应了不同的肺叶纹理形状，有病变性敏感度的成像序列会失去解剖结构的对比度，只有将结构影像的标注和定位配准到这些序列的影像上才能完成对每个肺区的信息挖掘和纹理提取。将不同序列的磁共振肺部影像配准到已标注和定位的结构影像，将不同序列的磁共振肺部影像配准到已标注和定位的结构影像，同时结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区。The characteristic lung area screening module 103 is also used for registering the magnetic resonance lung images of different sequences to the marked and localized structural images, and at the same time, combining the corresponding changes between the structural images of different sequences to screen out the corresponding changes of each sequence. Characteristic lung area. In this embodiment, the marked and localized structural images are the basis for the construction of the lung lobe texture, and the depth information mining for each structurally marked area is the lung lobe texture extraction; the imaging results of different sequences of magnetic resonance carry the lung lobe texture The contrast of images generated by different sequences corresponds to different texture shapes of lung lobes. Imaging sequences with lesion sensitivity will lose the contrast of anatomical structures. Only the annotation and positioning of structural images can be registered to the images of these sequences. Only then can the information mining and texture extraction of each lung area be completed. Register different sequences of magnetic resonance lung images to the marked and localized structural images, register different sequences of magnetic resonance lung images to the marked and localized structural images, and combine the differences between the different sequences of structural images. The corresponding change relationship was used to screen out the characteristic lung regions corresponding to each sequence.

所述肺叶特征识别模块104用于对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征。具体包括如下步骤：根据肺叶纹理的构成建立相似性测度算法，针对目标对象的磁共振肺部影像首先执行肺叶结构标注步骤和肺叶纹理提取步骤，从而提取肺叶纹理，再与肺叶图谱库进行比对，以相似度最高的肺叶纹理类别作为对该目标对象的纹理识别结果。在本实施例中，分别对每个特征肺区的多维信息(例如肺叶体积、纹理粗细、规则程度等)进行挖掘，并提取出与特定肺病变(例如，肺结节)的生理特性最相关的多维特征，从而完成肺叶纹理的信息提取。The lung lobe feature identification module 104 is used to mine the multi-dimensional information of each characteristic lung region, and extract the lung lobe texture feature most relevant to the physiological characteristics of the lung region's diseased state. Specifically, it includes the following steps: establishing a similarity measurement algorithm according to the composition of lung lobe textures, first performing the lung lobe structure labeling step and the lung lobe texture extraction step for the magnetic resonance lung image of the target object, so as to extract the lung lobe texture, and then compare it with the lung lobe atlas library , and use the lung lobe texture category with the highest similarity as the texture recognition result of the target object. In this embodiment, the multi-dimensional information (such as lung lobe volume, texture thickness, regularity, etc.) of each characteristic lung region is mined separately, and the most relevant physiological characteristics of specific lung lesions (such as lung nodules) are extracted. The multi-dimensional features of the lung lobe texture can be extracted.

所述肺叶特征识别模块104还用于从磁共振肺部影像中识别肺叶的肺泡及血管轮廓。具体地，所述肺叶特征识别模块104通过中值滤波对磁共振肺部影像进行滤波降噪；所述中值滤波为非线性滤波方法，和其他线性滤波方法相比，中值滤波在去除噪声的同时能够很好的保留影像中肺部区域和血管轮廓；所述肺叶特征识别模块104通过阈值分割、纹理分割相结合的方式提取肺区的肺泡信息及血管轮廓信息。在本实施例中，同时使用阈值分割或纹理分割可以更加准确的将肺部区域从影像中准确的分割出来，有效降低分割的错误率。The lung lobe feature identification module 104 is also used to identify the alveoli and blood vessel contours of the lung lobe from the magnetic resonance lung image. Specifically, the lung lobe feature identification module 104 performs filtering and noise reduction on the magnetic resonance lung image through median filtering; the median filtering is a nonlinear filtering method. Compared with other linear filtering methods, median filtering is used to remove noise. At the same time, the lung area and blood vessel contour in the image can be well preserved; the lung lobe feature recognition module 104 extracts the alveolar information and blood vessel contour information of the lung area by a combination of threshold segmentation and texture segmentation. In this embodiment, the use of threshold segmentation or texture segmentation at the same time can more accurately segment the lung region from the image, thereby effectively reducing the error rate of segmentation.

所述肺叶特征识别模块104还用于将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选获取k空间数据中大信号数据作为肺叶的多维病理特征。具体地，所述肺叶特征识别模块104将去除非肺部影像信息但包含肺部肺泡及血管轮廓信息的影像进行灰度反转，通过傅里叶变换将影像数据转化为其对应的k空间数据。在质子影像中肺泡由于含水量少，表现为黑色，而肺部组织含水量多表现为白色。在超极化气体成像过程中这一情况将正好相反，肺部组织表现为黑色，肺泡表现为白色。在这一步中通过对质子影像的灰度反转预测了超极化气体的成像结果，同时通过傅里叶变换得到了可能的k空间数据分布。所述肺叶特征识别模块104筛选并识别k空间数据中大信号数据（k空间数据采用a+bi复数形式表示的数据），按照

的大小筛选大信号，将获得的k空间数据中大信号数据作为肺叶的多维病理特征。筛选的过程即按照

作为标准对k空间所有的信号值进行排序。其中，

表征了成像过程中所针对的原子核自旋密度的大小，将获得的k空间数据中大信号的复数数据a+bi作为肺叶的多维病理特征。 The lobe feature identification module 104 is further configured to convert the texture features, alveolar information and blood vessel contour information of the lobe into corresponding k-space data, and screen and obtain large signal data in the k-space data as the multi-dimensional pathological features of the lobe. Specifically, the pulmonary lobe feature identification module 104 performs grayscale inversion of the image that removes non-pulmonary image information but contains pulmonary alveoli and blood vessel contour information, and converts the image data into its corresponding k-space data through Fourier transform . In proton imaging, alveoli appear black due to low water content, while lung tissue appears white with more water content. During hyperpolarized gas imaging, the situation is reversed, with lung tissue appearing black and alveoli appearing white. In this step, the imaging result of hyperpolarized gas is predicted by grayscale inversion of the proton image, and the possible k-space data distribution is obtained by Fourier transform. The lung lobe feature identification module 104 screens and identifies large signal data in k-space data (k-space data is represented by a+bi complex number form), according to

The size of the large signal was screened, and the large signal data in the obtained k-space data was used as the multi-dimensional pathological feature of the lung lobe. The screening process is based on

Sorts all signal values in k-space as a criterion. in,

The size of the nuclear spin density targeted in the imaging process was characterized, and the complex data a+bi of the large signal in the obtained k-space data was used as the multi-dimensional pathological feature of the lung lobe.

所述肺叶特征输出模块105用于将识别出的肺叶纹理特征、肺泡信息及血管轮廓信息等多维病理特征输出显示在显示器13上，或者输出存在存储器11中，以供医生在肺结节诊断治疗方面提供更加全面的参考。The pulmonary lobe feature output module 105 is used to output the identified multi-dimensional pathological features such as pulmonary lobe texture features, alveolar information, and blood vessel contour information on the display 13, or store the output in the memory 11 for doctors to diagnose and treat pulmonary nodules. provide a more comprehensive reference.

参考图2所示，是本发明基于人工智能的磁共振肺部影像识别方法较佳实施例的流程图。在本实施例中，所述基于人工智能的磁共振肺部影像识别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（例如本实施例的存储器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（例如本实施例的处理器12）加载并执行如下步骤。Referring to FIG. 2 , it is a flowchart of a preferred embodiment of the artificial intelligence-based magnetic resonance lung image recognition method of the present invention. In this embodiment, various method steps of the artificial intelligence-based magnetic resonance lung image recognition method are implemented by a computer software program, and the computer software program is stored in a computer-readable storage medium (such as a computer-readable storage medium in the form of computer program instructions). In the memory 11) of this embodiment, the computer-readable storage medium may include: read-only memory, random access memory, magnetic disk or optical disk, etc., and the computer program instructions can be loaded by a processor (for example, the processor 12 of this embodiment) and stored Perform the following steps.

步骤S21，选取若干健康人和不同肺结节疾病患者作为目标对象，利用磁共振成像设备2获取目标对象的磁共振肺部影像；在目标对象呼吸空气的情况下，利用磁共振成像设备对目标对象肺部区域进行磁共振成像获取肺部的氢质子影像。质子成像过程中由于肺部氢质子含量非常少，采用自旋回波（Spin Echo，简称SE）类型的序列与其他类型序列相比有利于提高信噪比，得到更好的影像。在本实施例中，所述磁共振肺部影像包括所有磁共振的序列所能产生的影像以及这些序列产生影像的任意组合，在所指的任意序列影像的组合中，单序列T1、T2或DTI影像是所允许的最小影像集合。Step S21, select a number of healthy people and patients with different pulmonary nodule diseases as the target objects, and use the magnetic resonance imaging equipment 2 to obtain the magnetic resonance lung images of the target objects; Magnetic resonance imaging of the subject's lung area to obtain hydrogen proton images of the lungs. During proton imaging, spin echo (Spin echo) was used because the hydrogen proton content in the lungs was very low. Compared with other types of sequences, the sequence of Echo (SE) type is beneficial to improve the signal-to-noise ratio and obtain better images. In this embodiment, the magnetic resonance lung images include images that can be generated by all sequences of magnetic resonance and any combination of images generated by these sequences. DTI images are the smallest set of images allowed.

步骤S22，从肺叶图谱库筛选出与磁共振肺部影像匹配度最高的若干套肺叶图谱；在本实施例中，在已有的肺叶图谱库中进行图谱预选，通过图谱匹配算法筛选与磁共振肺部影像匹配度最高的10～20套肺叶图谱，作为分析磁共振肺部影像的参考标准。所述肺叶图谱库建立的步骤：基于大量的肺部影像样本及其病例信息，利用人工智能的方法对不同性别、年龄以及肺结节疾病所对应的肺叶纹理进行分类，再进行人工的鉴别与确认，最后形成不同人群的肺叶图谱数据，该肺叶图谱数据将随着样本量的增加而不断扩张，建立对未知肺结节疾病的肺叶纹理识别的标准库基础。In step S22, several sets of lung lobe atlases with the highest matching degree with the magnetic resonance lung image are selected from the lung lobe atlas library; in this embodiment, atlas pre-selection is performed in the existing lung lobe atlas library, and the atlas matching algorithm is used to screen and MRI. The 10-20 sets of lung lobe atlases with the highest matching degree of lung images are used as the reference standard for analyzing MRI lung images. The steps of establishing the lung lobe atlas library: based on a large number of lung image samples and their case information, use artificial intelligence to classify the lung lobe textures corresponding to different genders, ages and pulmonary nodule diseases, and then perform manual identification and identification. Confirmed, and finally formed the lung lobe atlas data of different populations, the lung lobe atlas data will continue to expand with the increase of the sample size, and establish a standard library basis for the identification of lung lobe texture for unknown pulmonary nodular diseases.

步骤S23，对磁共振肺部影像进行肺区分割，并将预选的肺叶图谱分别配准到磁共振肺部影像；在本实施例中，针对T1、T2或DTI序列的影像进行基于多肺叶图谱对磁共振肺部影像的分割，将上述预选出来的肺叶图谱分别配准到磁共振肺部影像。针对目标对象的磁共振肺部影像选定一个标准肺叶模板，对每一个目标对象的磁共振肺部影像相同层面的多序列扫描影像进行配准，配准同一个体相同层面的多序列扫描影像以消除头动效应，之后对其进行影像标准化操作以实现不同个体间的影像横向对比。In step S23, lung region segmentation is performed on the magnetic resonance lung image, and the preselected lung lobe atlases are respectively registered to the magnetic resonance lung image; For the segmentation of the magnetic resonance lung image, the above pre-selected lung lobe atlases are respectively registered to the magnetic resonance lung image. A standard lung lobe template is selected for the magnetic resonance lung image of the target object, and the multi-sequence scan images of the same level of the magnetic resonance lung image of each target object are registered. The head motion effect was eliminated, and then image normalization was performed on it to achieve horizontal comparison of images between different individuals.

步骤S24，对多个肺叶图谱所形成的分割边界进行融合生成磁共振肺部影像的肺区分割结果。在本实施例中，在对多个肺叶图谱所形成的分割边界进行融合，最后生成磁共振肺部影像的肺区分割结果，该肺区分割结果最大的特色就是对每一个肺区进行了标注和定位。Step S24 , fuse the segmentation boundaries formed by the multiple lung lobe atlases to generate a lung region segmentation result of the magnetic resonance lung image. In this embodiment, the segmentation boundaries formed by multiple lung lobe atlases are fused, and finally the lung area segmentation result of the magnetic resonance lung image is generated. The biggest feature of the lung area segmentation result is that each lung area is labeled and positioning.

步骤S25，将不同序列的磁共振肺部影像配准到已标注和定位的结构影像，同时结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区。在本实施例中，已完成标注和定位的结构影像是肺叶纹理构建的基础，针对每一个结构标注的区域进行深度的信息挖掘便是肺叶纹理提取；磁共振不同序列的成像结果承载了肺叶纹理的不同特性，不同序列所产生影像的对比度对应了不同的肺叶纹理形状，有病变性敏感度的成像序列会失去解剖结构的对比度，只有将结构影像的标注和定位配准到这些序列的影像上才能完成对每个肺区的信息挖掘和纹理提取。将不同序列的磁共振肺部影像配准到已标注和定位的结构影像，将不同序列的磁共振肺部影像配准到已标注和定位的结构影像，同时结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区。Step S25 , register the magnetic resonance lung images of different sequences to the marked and localized structural images, and screen out the characteristic lung regions corresponding to each sequence in combination with the corresponding changes between the structural images of different sequences. In this embodiment, the marked and localized structural images are the basis for the construction of the lung lobe texture, and the depth information mining for each structurally marked area is the lung lobe texture extraction; the imaging results of different sequences of magnetic resonance carry the lung lobe texture The contrast of images generated by different sequences corresponds to different texture shapes of lung lobes. Imaging sequences with lesion sensitivity will lose the contrast of anatomical structures. Only the annotation and positioning of structural images can be registered to the images of these sequences. Only then can the information mining and texture extraction of each lung area be completed. Register different sequences of magnetic resonance lung images to the marked and localized structural images, register different sequences of magnetic resonance lung images to the marked and localized structural images, and combine the differences between the different sequences of structural images. The corresponding change relationship was used to screen out the characteristic lung regions corresponding to each sequence.

步骤S26，对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征。具体包括如下步骤：根据肺叶纹理的构成建立相似性测度算法，针对目标对象的磁共振肺部影像首先执行肺叶结构标注步骤和肺叶纹理提取步骤，从而提取肺叶纹理，再与肺叶图谱库进行比对，以相似度最高的肺叶纹理类别作为对该目标对象的纹理识别结果。在本实施例中，分别对每个特征肺区的多维信息(例如肺叶体积、纹理粗细、规则程度等)进行挖掘，并提取出与特定肺病变(例如，肺结节)的生理特性最相关的多维特征，从而完成肺叶纹理的信息提取。In step S26, the multi-dimensional information of each characteristic lung region is mined, and the lung lobe texture feature most relevant to the physiological characteristics of the diseased state of the lung region is extracted. Specifically, it includes the following steps: establishing a similarity measurement algorithm according to the composition of lung lobe textures, first performing the lung lobe structure labeling step and the lung lobe texture extraction step for the magnetic resonance lung image of the target object, so as to extract the lung lobe texture, and then compare it with the lung lobe atlas library , and use the lung lobe texture category with the highest similarity as the texture recognition result of the target object. In this embodiment, the multi-dimensional information (such as lung lobe volume, texture thickness, regularity, etc.) of each characteristic lung region is mined separately, and the most relevant physiological characteristics of specific lung lesions (such as lung nodules) are extracted. The multi-dimensional features of the lung lobe texture can be extracted.

步骤S27，从磁共振肺部影像中识别肺叶的肺泡及血管轮廓；该步骤S27具体包括如下步骤：通过中值滤波对磁共振肺部影像进行滤波降噪；所述中值滤波为非线性滤波方法，和其他线性滤波方法相比，中值滤波在去除噪声的同时能够很好的保留影像中肺部区域和血管轮廓；通过阈值分割、纹理分割相结合的方式提取肺区的肺泡信息及血管轮廓信息。在本实施例中，同时使用阈值分割或纹理分割可以更加准确的将肺部区域从影像中准确的分割出来，有效降低分割的错误率。Step S27, identifying the alveoli and blood vessel contours of the lung lobe from the magnetic resonance lung image; the step S27 specifically includes the following steps: filtering and denoising the magnetic resonance lung image by median filtering; the median filtering is nonlinear filtering Compared with other linear filtering methods, median filtering can well preserve the contours of the lung area and blood vessels in the image while removing noise; the alveolar information and blood vessels of the lung area are extracted by a combination of threshold segmentation and texture segmentation. profile information. In this embodiment, the use of threshold segmentation or texture segmentation at the same time can more accurately segment the lung region from the image, thereby effectively reducing the error rate of segmentation.

步骤S28，将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选获取k空间数据中大信号数据作为肺叶的多维病理特征。该步骤S28具体包括如下步骤：将去除非肺部影像信息但包含肺部肺泡及血管轮廓信息的影像进行灰度反转，通过傅里叶变换将影像数据转化为其对应的k空间数据。在质子影像中肺泡由于含水量少，表现为黑色，而肺部组织含水量多表现为白色。在超极化气体成像过程中这一情况将正好相反，肺部组织表现为黑色，肺泡表现为白色。在这一步中通过对质子影像的灰度反转预测了超极化气体的成像结果，同时通过傅里叶变换得到了可能的k空间数据分布；筛选并识别k空间数据中大信号数据（k空间数据采用a+bi复数形式表示的数据），按照

作为标准对k空间所有的信号值进行排序。其中，

表征了成像过程中所针对的原子核自旋密度的大小，将这个步骤所获得的k空间数据中大信号的复数数据a+bi作为肺叶的多维病理特征。 In step S28, the texture features of the lung lobes, the alveolar information and the blood vessel contour information are converted into corresponding k-space data, and the large signal data in the k-space data is screened and obtained as the multi-dimensional pathological features of the lung lobes. The step S28 specifically includes the following steps: performing grayscale inversion of the image from which the non-pulmonary image information is removed but including the pulmonary alveolar and blood vessel contour information, and converting the image data into its corresponding k-space data through Fourier transform. In proton imaging, alveoli appear black due to low water content, while lung tissue appears white with more water content. During hyperpolarized gas imaging, the situation is reversed, with lung tissue appearing black and alveoli appearing white. In this step, the imaging result of hyperpolarized gas is predicted by the grayscale inversion of the proton image, and the possible k-space data distribution is obtained through Fourier transform; Spatial data is represented by a+bi complex number), according to

Sorts all signal values in k-space as a criterion. in,

The size of the nuclear spin density targeted in the imaging process was characterized, and the complex data a+bi of the large signal in the k-space data obtained in this step was used as the multi-dimensional pathological feature of the lung lobe.

在本实施例中，所述基于人工智能的磁共振肺部影像识别方法还包括如下步骤：将识别出的肺叶纹理特征、肺泡信息及血管轮廓信息等多维病理特征输出显示在显示器13上，或者输出存在存储器11中，以供医生在肺结节诊断治疗方面提供更加全面的参考。In this embodiment, the artificial intelligence-based magnetic resonance lung image identification method further includes the following steps: outputting the identified multi-dimensional pathological features such as lung lobe texture features, alveolar information, and blood vessel contour information on the display 13, or The output is stored in the memory 11 for doctors to provide a more comprehensive reference in the diagnosis and treatment of pulmonary nodules.

本发明还提供一种计算机可读存储介质，该计算机可读存储介质存储多条计算机程序指令，所述计算机程序指令由计算机装置的处理器加载并执行本发明所述基于人工智能的磁共振肺部影像识别方法的各个步骤。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。The present invention also provides a computer-readable storage medium, which stores a plurality of computer program instructions, and the computer program instructions are loaded by the processor of the computer device and execute the artificial intelligence-based magnetic resonance lung of the present invention. Each step of the image recognition method. Those skilled in the art can understand that all or part of the steps of the various methods in the above-mentioned embodiments can be completed by relevant program instructions, and the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, Disk or CD etc.

本发明所述基于人工智能的磁共振肺部影像识别装置及方法，基于磁共振影像的肺叶病理特征的识别及提取技术，该技术允许对肺部组织的多种重要性质进行非侵入的同步量化检测，并且为肺叶各区域复杂的功能、生理及物理变化和形态变化提供了智能化的检测与识别方法。与人工阅片相比，改善了人体对肺部生理异常描述的话术体系，提高了以磁共振影像为基础的对人体肺部生理、物理和功能特性的检查精度，同时以肺叶纹理特征、肺泡及血管轮廓等信息为指标，提高了对肺部疾病所引起的肺部异常检测的敏感性、特异性和检查准确性，为医生对肺癌的诊断治疗提供医学指导。The artificial intelligence-based magnetic resonance lung image identification device and method of the present invention, the identification and extraction technology of lung lobe pathological features based on magnetic resonance images, allows non-invasive simultaneous quantification of various important properties of lung tissue It also provides intelligent detection and identification methods for complex functional, physiological and physical changes and morphological changes in various areas of the lung lobe. Compared with manual reading, it improves the human body's description of lung physiological abnormalities, and improves the inspection accuracy of human lung physiological, physical and functional characteristics based on magnetic resonance imaging. And blood vessel contour and other information as indicators, improve the sensitivity, specificity and inspection accuracy of lung abnormality detection caused by lung diseases, and provide medical guidance for doctors to diagnose and treat lung cancer.

以上仅为本发明的较佳实施例，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。The above are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the description and drawings of the present invention, or directly or indirectly applied to other related technologies Fields are similarly included in the scope of patent protection of the present invention.

工业实用性Industrial Applicability

Claims

一种基于人工智能的磁共振肺部影像识别装置，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，其特征在于，所述计算机程序指令由处理器加载并执行如下步骤：A magnetic resonance lung image recognition device based on artificial intelligence, comprising a processor suitable for implementing various computer program instructions and a memory suitable for storing a plurality of computer program instructions, wherein the computer program instructions are executed by the processor. Load and perform the following steps:

选取若干健康人和不同肺结节疾病患者作为目标对象，利用磁共振成像设备获取目标对象的磁共振肺部影像；Select a number of healthy people and patients with different pulmonary nodule diseases as target objects, and use magnetic resonance imaging equipment to obtain magnetic resonance lung images of the target objects;

从肺叶图谱库筛选出与磁共振肺部影像匹配度最高的若干套肺叶图谱；Several sets of lobe atlases with the highest matching degree with MRI lung images were screened from the lung lobe atlas library;

对目标对象的磁共振肺部影像进行肺区分割，并将预选的肺叶图谱分别配准到磁共振肺部影像；Perform lung region segmentation on the magnetic resonance lung image of the target object, and register the preselected lung lobe maps to the magnetic resonance lung image respectively;

对多个肺叶图谱所形成的分割边界进行融合生成磁共振肺部影像的分割结果；Fusion of segmentation boundaries formed by multiple lung lobe atlases generates segmentation results of magnetic resonance lung images;

将磁共振肺部影像配准到已标注和定位的结构影像，并结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区；The magnetic resonance lung images are registered to the marked and localized structural images, and the characteristic lung regions corresponding to each sequence are screened by combining the corresponding changes between the structural images of different sequences;

对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征；Mining the multi-dimensional information of each characteristic lung area, and extracting the lung lobe texture features most relevant to the physiological characteristics of the lung area's diseased state;

从磁共振肺部影像识别肺叶的肺泡及血管轮廓；Identify alveolar and vascular contours of lung lobes from magnetic resonance lung images;

将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征。The texture features of lung lobes, alveolar information and blood vessel contour information are converted into corresponding k-space data, and the large signal data in the k-space data are screened as multi-dimensional pathological features of lung lobes.
如权利要求1所述的基于人工智能的磁共振肺部影像识别装置，其特征在于，所述计算机程序指令由处理器加载还执行如下步骤：The artificial intelligence-based magnetic resonance lung image recognition device of claim 1, wherein the computer program instructions are loaded by the processor and further perform the following steps:

针对目标对象的磁共振肺部影像选定一个标准肺叶模板，对每一个目标对象相同层面的多序列磁共振肺部影像进行配准。A standard lung lobe template is selected for the magnetic resonance lung images of the target object, and the multi-sequence magnetic resonance lung images of the same level of each target object are registered.
如权利要求1所述的基于人工智能的磁共振肺部影像识别装置，其特征在于，所述对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征的步骤包括：The artificial intelligence-based magnetic resonance lung image recognition device according to claim 1, wherein the multi-dimensional information of each characteristic lung region is mined, and the most relevant physiological characteristics of the diseased state of the lung region are extracted. The steps of the lobe texture feature include:

根据肺叶纹理的构成建立相似性测度算法，针对目标对象的磁共振肺部影像执行肺叶结构标注和肺叶纹理提取，再与肺叶图谱库进行比对，以相似度最高的肺叶纹理类别作为对该目标对象的纹理识别结果。According to the composition of lung lobe texture, a similarity measurement algorithm is established, and lung lobe structure annotation and lung lobe texture extraction are performed for the magnetic resonance lung image of the target object, and then compared with the lung lobe atlas library, and the lung lobe texture category with the highest similarity is used as the target object. The texture recognition result of the object.
如权利要求1所述的基于人工智能的磁共振肺部影像识别装置，其特征在于，所述从磁共振肺部影像识别肺叶的肺泡及血管轮廓的步骤包括：The artificial intelligence-based magnetic resonance lung image recognition device according to claim 1, wherein the step of identifying the alveoli and blood vessel contours of the lung lobe from the magnetic resonance lung image comprises:

通过中值滤波对磁共振肺部影像进行滤波降噪；Filter and denoise MRI lung images by median filtering;

通过阈值分割与纹理分割相结合的方式提取肺区的肺泡信息及血管轮廓信息。The alveolar information and blood vessel contour information of the lung area are extracted by the combination of threshold segmentation and texture segmentation.
如权利要求4所述的基于人工智能的磁共振肺部影像识别装置，其特征在于，所述将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征的步骤包括如下步骤：The artificial intelligence-based magnetic resonance lung image recognition device according to claim 4, wherein the lung lobe texture features, alveolar information and blood vessel contour information are converted into corresponding k-space data, and the k-space data are filtered out. The steps of using the large signal data as multidimensional pathological features of lung lobes include the following steps:

将去除非肺部影像信息但包含纹理特征、肺泡信息及血管轮廓信息的影像进行灰度反转，通过傅里叶变换将影像数据转化为对应的k空间数据；Grayscale inversion is performed on the image that removes non-pulmonary image information but contains texture features, alveolar information and blood vessel contour information, and transforms the image data into corresponding k-space data through Fourier transform;

筛选并识别k空间数据中大信号的复数数据a+bi，按照
的大小筛选大信号，将获得的k空间数据中大信号数据作为肺叶的多维病理特征。 Screen and identify complex data a+bi of large signals in k-space data, according to
The size of the large signal was screened, and the large signal data in the obtained k-space data was used as the multi-dimensional pathological feature of the lung lobe.
一种基于人工智能的磁共振肺部影像识别方法，应用于计算机装置中，该计算机装置连接有磁共振成像设备，其特征在于，该方法包括如下步骤：A magnetic resonance lung image recognition method based on artificial intelligence, which is applied to a computer device, and the computer device is connected with magnetic resonance imaging equipment, and is characterized in that, the method comprises the following steps:

选取若干健康人和不同肺结节疾病患者作为目标对象，利用磁共振成像设备获取目标对象的磁共振肺部影像；Select a number of healthy people and patients with different pulmonary nodule diseases as target objects, and use magnetic resonance imaging equipment to obtain magnetic resonance lung images of the target objects;

从肺叶图谱库筛选出与磁共振肺部影像匹配度最高的若干套肺叶图谱；Several sets of lobe atlases with the highest matching degree with MRI lung images were screened from the lung lobe atlas library;

对目标对象的磁共振肺部影像进行肺区分割，并将预选的肺叶图谱分别配准到磁共振肺部影像；Perform lung region segmentation on the magnetic resonance lung image of the target object, and register the preselected lung lobe maps to the magnetic resonance lung image respectively;

对多个肺叶图谱所形成的分割边界进行融合生成磁共振肺部影像的分割结果；Fusion of segmentation boundaries formed by multiple lung lobe atlases generates segmentation results of magnetic resonance lung images;

将磁共振肺部影像配准到已标注和定位的结构影像，并结合不同序列的结构影像之间的对应变化关系筛选出各序列对应的特征肺区；The magnetic resonance lung images are registered to the marked and localized structural images, and the characteristic lung regions corresponding to each sequence are screened by combining the corresponding changes between the structural images of different sequences;

对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征；Mining the multi-dimensional information of each characteristic lung area, and extracting the lung lobe texture features most relevant to the physiological characteristics of the lung area's diseased state;

从磁共振肺部影像识别肺叶的肺泡及血管轮廓；Identify alveolar and vascular contours of lung lobes from magnetic resonance lung images;

将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征。The texture features of lung lobes, alveolar information and blood vessel contour information are converted into corresponding k-space data, and the large signal data in the k-space data are screened as multi-dimensional pathological features of lung lobes.
如权利要求6所述的基于人工智能的磁共振肺部影像识别方法，其特征在于，该方法还包括如下步骤：The artificial intelligence-based magnetic resonance lung image recognition method according to claim 6, wherein the method further comprises the following steps:

针对目标对象的磁共振肺部影像选定一个标准肺叶模板，对每一个目标对象相同层面的多序列磁共振肺部影像进行配准。A standard lung lobe template is selected for the magnetic resonance lung images of the target object, and the multi-sequence magnetic resonance lung images of the same level of each target object are registered.
如权利要求6所述的基于人工智能的磁共振肺部影像识别方法，其特征在于，所述对每个特征肺区的多维信息进行挖掘，并提取出与肺区病变状态的生理特性最相关的肺叶纹理特征的步骤包括：The artificial intelligence-based magnetic resonance lung image recognition method according to claim 6, wherein the multi-dimensional information of each characteristic lung region is mined, and the most relevant physiological characteristics of the lung region's diseased state are extracted. The steps of the lobe texture feature include:

根据肺叶纹理的构成建立相似性测度算法，针对目标对象的磁共振肺部影像执行肺叶结构标注和肺叶纹理提取，再与肺叶图谱库进行比对，以相似度最高的肺叶纹理类别作为对该目标对象的纹理识别结果。According to the composition of lung lobe texture, a similarity measurement algorithm is established, and lung lobe structure annotation and lung lobe texture extraction are performed for the magnetic resonance lung image of the target object, and then compared with the lung lobe atlas library, and the lung lobe texture category with the highest similarity is used as the target object. The texture recognition result of the object.
如权利要求6所述的基于人工智能的磁共振肺部影像识别方法，其特征在于，所述从磁共振肺部影像识别肺叶的肺泡及血管轮廓的步骤包括：The artificial intelligence-based magnetic resonance lung image recognition method according to claim 6, wherein the step of identifying the alveoli and blood vessel contours of the lung lobe from the magnetic resonance lung image comprises:

通过中值滤波对磁共振肺部影像进行滤波降噪；Filter and denoise MRI lung images by median filtering;

通过阈值分割与纹理分割相结合的方式提取肺区的肺泡信息及血管轮廓信息。The alveolar information and blood vessel contour information of the lung area are extracted by the combination of threshold segmentation and texture segmentation.
如权利要求9所述的基于人工智能的磁共振肺部影像识别方法，其特征在于，所述将肺叶纹理特征、肺泡信息及血管轮廓信息转换为对应的k空间数据，并筛选k空间数据中的大信号数据作为肺叶的多维病理特征的步骤包括如下步骤：The artificial intelligence-based magnetic resonance lung image recognition method according to claim 9, wherein the lung lobe texture feature, alveolar information and blood vessel contour information are converted into corresponding k-space data, and the k-space data are screened. The steps of using the large signal data as multidimensional pathological features of lung lobes include the following steps:

将去除非肺部影像信息但包含纹理特征、肺泡信息及血管轮廓信息的影像进行灰度反转，通过傅里叶变换将影像数据转化为对应的k空间数据；Grayscale inversion is performed on the image that removes non-pulmonary image information but contains texture features, alveolar information and blood vessel contour information, and transforms the image data into corresponding k-space data through Fourier transform;

筛选并识别k空间数据中大信号的复数数据a+bi，按照
的大小筛选大信号，将获得的k空间数据中大信号数据作为肺叶的多维病理特征。 Screen and identify complex data a+bi of large signals in k-space data, according to
The size of the large signal was screened, and the large signal data in the obtained k-space data was used as the multi-dimensional pathological feature of the lung lobe.