阅读说明：本技术 肺结节属性特征信息的提取方法及系统 (The extracting method and system of Lung neoplasm attributive character information ) 是由 马杰超 崔星 田希 陈宽 王少康 于 2019-08-30 设计创作，主要内容包括：本发明公开了一种肺结节属性特征信息的提取方法及系统,获取目标信息；采用预设特征提取深度学习模型对目标信息进行特征提取,得到第一特征信息；采用预设数据挖掘算法对目标信息进行特征提取,得到第二特征信息；根据第一特征信息和第二特征信息进行融合并生成维度矢量；将维度矢量输入至目标分类器中,得到目标信息的肺结节属性特征信息。在本发明中可以基于深度学习模型获得目标信息中的低维度像素特征,基于数据挖掘算法获得高维度的语义信息特征,这样使得获得的特征信息更加完整和准确,然后基于对上述特征信息的融合和分类,使得获得的肺结节属性特征信息更加准确,从而可以提升肺结节润侵性的分级的准确性。(The invention discloses the extracting methods and system of a kind of Lung neoplasm attributive character information, obtain target information；Feature extraction is carried out to target information using default feature extraction deep learning model, obtains fisrt feature information；Feature extraction is carried out to target information using preset data mining algorithm, obtains second feature information；It is merged according to fisrt feature information and second feature information and generates dimension vector；Dimension vector is input in object classifiers, the Lung neoplasm attributive character information of target information is obtained.The low dimensional pixel characteristic in target information can be obtained based on deep learning model in the present invention, high-dimensional semantic information feature is obtained based on data mining algorithm, make the characteristic information obtained more complete and accurate in this way, it is then based on the fusion and classification to features described above information, so that the Lung neoplasm attributive character information obtained is more accurate, so as to promote the accuracy that Lung neoplasm moistens the classification of invading property.)

1. a kind of extracting method of Lung neoplasm attributive character information, which is characterized in that this method comprises:

Target information is obtained, the target information is generated according to Lung neoplasm CT figure to be analyzed；

Feature extraction is carried out to the target information using default feature extraction deep learning model, obtains fisrt feature information, Wherein, the parameter of the default feature extraction deep learning model is by being trained to obtain to multiple Lung neoplasm CT figures , the fisrt feature information representation low dimensional pixel characteristic；

Feature extraction is carried out to the target information using preset data mining algorithm, obtains second feature information, wherein described Preset data mining algorithm according to the Lung neoplasm CT figure be associated analysis obtain, the second feature information representation High-dimensional semantic information feature；

Fusion treatment is carried out according to the fisrt feature information and the second feature information, and is generated according to fused feature Dimension vector；

The dimension vector is input in object classifiers, the Lung neoplasm attributive character information of the target information, institute are obtained State the rating information of invading property of Lung neoplasm attributive character information representation Lung neoplasm profit.

2. the method according to claim 1, wherein the acquisition target information, comprising:

Obtain lung's CT scan CT image of target patient；

Lung neoplasm image is extracted in lung's CT scan CT image of the target patient, obtains target letter Breath.

3. the method according to claim 1, wherein the default feature extraction deep learning model includes convolution Layer, pond layer and full articulamentum, wherein the convolutional layer is used to learn the Feature Mapping of the Lung neoplasm CT figure of input, institute The computation complexity that pond layer is used to reduce the deep learning model is stated, the full articulamentum is used to the feature obtained to study Mapping formats, and the characteristic information obtained is identified by the object classifiers.

4. the method according to claim 1, wherein using preset data mining algorithm to the target information into Row feature extraction obtains second feature information, comprising:

Feature extraction is carried out to the target information using preset data mining algorithm, obtains initial characteristics information, it is described initial Characteristic information includes shape feature, strength characteristic, textural characteristics and high-order feature；

Feature selecting is carried out to the initial characteristics, and signature analysis is carried out to the initial characteristics after selection, obtains second feature Information.

5. according to the method described in claim 3, it is characterized in that, described according to the fisrt feature information and described second special Reference breath carries out fusion treatment, and generates dimension vector according to fused feature, comprising:

The fisrt feature information is handled using the full articulamentum, obtains the one-dimensional characteristic vector comprising N number of value；

The second feature information is encoded using the preset data mining algorithm, generating is worth one-dimensional arrow including M Amount；

Group is carried out including a M worth n dimensional vector ns to described in the one-dimensional characteristic vector sum comprising N number of value according to channel rank Superposition is closed, dimension vector is generated, the dimension vector is a n dimensional vector n of M+N value size, wherein N and M is that numerical value is different Positive integer.

6. according to the method described in claim 4, it is characterized in that, the signature analysis includes correlation analysis, clustering With one of principal component analysis or a variety of.

7. a kind of extraction system of Lung neoplasm attributive character information, which is characterized in that the system includes:

Information acquisition unit, for obtaining target information, the target information is generated according to Lung neoplasm CT figure to be analyzed；

First extraction unit, for carrying out feature extraction to the target information using default feature extraction deep learning model, Obtain fisrt feature information, wherein the parameter of the default feature extraction deep learning model is by multiple lung knots Section CT figure is trained, the fisrt feature information representation low dimensional pixel characteristic；

Second extraction unit obtains second for carrying out feature extraction to the target information using preset data mining algorithm Characteristic information, wherein the preset data mining algorithm according to the Lung neoplasm CT figure be associated analysis obtain, institute State the high-dimensional semantic information feature of second feature information representation；

Vector generation unit, for carrying out fusion treatment, and root according to the fisrt feature information and the second feature information Dimension vector is generated according to fused feature；

Taxon obtains the Lung neoplasm category of the target information for the dimension vector to be input in object classifiers Property characteristic information, the rating information of invading property of Lung neoplasm attributive character information representation Lung neoplasm profit.

8. system according to claim 7, which is characterized in that the acquisition information acquisition unit, comprising:

Image obtains subelement, for obtaining lung's CT scan CT image of target patient；

Image zooming-out subelement, for extracting lung knot in lung's CT scan CT image of the target patient Image is saved, target information is obtained.

9. system according to claim 7, which is characterized in that the default feature extraction deep learning model includes convolution Layer, pond layer and full articulamentum, wherein the convolutional layer is used to learn the Feature Mapping of the Lung neoplasm CT figure of input, institute The computation complexity that pond layer is used to reduce the deep learning model is stated, the full articulamentum is used to the feature obtained to study Mapping formats, and the characteristic information obtained is identified by the object classifiers；

Second extraction unit, comprising:

Feature extraction subelement obtains just for carrying out feature extraction to the target information using preset data mining algorithm Beginning characteristic information, the initial characteristics information include shape feature, strength characteristic, textural characteristics and high-order feature；

Subelement is analyzed, for carrying out feature selecting to the initial characteristics, and feature point is carried out to the initial characteristics after selection Analysis obtains second feature information；The signature analysis include one of correlation analysis, clustering and principal component analysis or It is a variety of.

10. system according to claim 9, which is characterized in that the vector generation unit includes:

Information processing subelement is obtained for being handled using the full articulamentum the fisrt feature information comprising N number of The one-dimensional characteristic vector of value；

Coded sub-units generate packet for encoding using the preset data mining algorithm to the second feature information Include a M worth n dimensional vector ns；

Subelement is generated, for being worth to described in the one-dimensional characteristic vector sum comprising N number of value including M according to channel rank One n dimensional vector n is combined superposition, generates dimension vector, and the dimension vector is a n dimensional vector n of M+N value size, wherein N It is the different positive integer of numerical value with M.

Technical field

The present invention relates to technical field of information processing, more particularly to a kind of extracting method of Lung neoplasm attributive character information And system.

Background technique

The main reason for lung cancer is global cancer related mortality, with multi-layer helical computed tomography (CT) technology Development is universal with low-dose spiral CT screening, and getting up early may find that more and more cases of lung cancer, to greatly reduce Death toll.Cause lung cancer in high-resolution ct (HRCT), tubercle shows that reality, part-solid or non-real venereal disease become, and big Part part-solid and non-real venereal disease change are accredited as pernicious.In order to solve the problems, such as the Accurate classification of Lung neoplasm, international lung cancer is ground Study carefully association (IASLC), American Thoracic Society (ATS) and European pneumatology meeting (ERS) and proposes a kind of new adenocarcinoma of lung classification side Method.According to the presence of reality ingredient in the size and analysis of cases of lesion, ground glass tubercle (ggsn) is divided into atypia gland cancer Property hyperplasia (AAH), in situ adenocarcinoma (AIS), minimally invasive cancer gland (MIA) or invasion sexual gland cancer (IAC).It is pre- due to different case hypotypes After differ greatly, the therapeutic choice of new patient class and follow-up have a major impact.

The profit of existing Lung neoplasm, which invades degree classification all, to be analyzed by medical worker, so that the work of medical worker It measures huge, and is influenced meeting so that having differences property of classification results by supervisor's factor of different doctors.Although in the prior art Also the profit for solving Lung neoplasm through computer technology invades the classification problem of degree, but can there is special to Lung neoplasm attribute The extraction inaccuracy of reference breath, and key message can be lost, finally to be realized according to the Lung neoplasm attribute information extracted Classification results inaccuracy.

Summary of the invention

It is directed to the above problem, the present invention provides the extracting method and system of a kind of Lung neoplasm attributive character information, realizes Accurate extraction to Lung neoplasm attributive character information, to promote the Accurate classification for invading Lung neoplasm profit degree.

To achieve the goals above, the present invention provides the following technical scheme that

A kind of extracting method of Lung neoplasm attributive character information, this method comprises:

Target information is obtained, the target information is generated according to Lung neoplasm CT figure to be analyzed；

Feature extraction is carried out to the target information using default feature extraction deep learning model, obtains fisrt feature letter Breath, wherein the parameter of the default feature extraction deep learning model is by being trained to multiple Lung neoplasm CT figures It obtains, the fisrt feature information representation low dimensional pixel characteristic；

Feature extraction is carried out to the target information using preset data mining algorithm, obtains second feature information, wherein The preset data mining algorithm according to the Lung neoplasm CT figure be associated analysis obtain, the second feature information Characterize high-dimensional semantic information feature；

Fusion treatment is carried out according to the fisrt feature information and the second feature information, and according to fused feature Generate dimension vector；

The dimension vector is input in object classifiers, the Lung neoplasm attributive character letter of the target information is obtained Breath, the rating information of the invading property of Lung neoplasm attributive character information representation Lung neoplasm profit.

Optionally, the acquisition target information, comprising:

Obtain lung's CT scan CT image of target patient；

Lung neoplasm image is extracted in lung's CT scan CT image of the target patient, obtains target Information.

Optionally, the default feature extraction deep learning model includes convolutional layer, pond layer and full articulamentum, wherein The convolutional layer is used to learn the Feature Mapping of the Lung neoplasm CT figure of input, and the pond layer is used to reduce the depth The computation complexity of model is practised, the full articulamentum is used to format the Feature Mapping that study obtains, so that obtaining Characteristic information can be identified by the object classifiers.

Optionally, feature extraction is carried out to the target information using preset data mining algorithm, obtains second feature letter Breath, comprising:

Feature extraction is carried out to the target information using preset data mining algorithm, obtains initial characteristics information, it is described Initial characteristics information includes shape feature, strength characteristic, textural characteristics and high-order feature；

Feature selecting is carried out to the initial characteristics, and signature analysis is carried out to the initial characteristics after selection, obtains second Characteristic information.

Optionally, described according to the fisrt feature information and second feature information progress fusion treatment, and according to Fused feature generates dimension vector, comprising:

The fisrt feature information is handled using the full articulamentum, the one-dimensional characteristic comprising N number of value is obtained and swears Amount；

The second feature information is encoded using the preset data mining algorithm, generating includes the one of M value N dimensional vector n；

According to channel rank to described in the one-dimensional characteristic vector sum comprising N number of value include M be worth a n dimensional vector n into Row combination superposition, generate dimension vector, the dimension vector be M+N value size a n dimensional vector n, wherein N and M for numerical value not Same positive integer.

Optionally, the signature analysis includes one of correlation analysis, clustering and principal component analysis or a variety of.

A kind of extraction system of Lung neoplasm attributive character information, the system include:

Information acquisition unit, for obtaining target information, the target information is according to Lung neoplasm CT figure life to be analyzed At；

First extraction unit is mentioned for carrying out feature to the target information using default feature extraction deep learning model It takes, obtains fisrt feature information, wherein the parameter of the default feature extraction deep learning model is by multiple lungs Tubercle CT figure is trained, the fisrt feature information representation low dimensional pixel characteristic；

Second extraction unit is obtained for carrying out feature extraction to the target information using preset data mining algorithm Second feature information, wherein the preset data mining algorithm is obtained according to being associated analysis to the Lung neoplasm CT figure , the high-dimensional semantic information feature of the second feature information representation；

Vector generation unit, for carrying out fusion treatment according to the fisrt feature information and the second feature information, And dimension vector is generated according to fused feature；

Taxon obtains the lung knot of the target information for the dimension vector to be input in object classifiers Section attribute characteristic information, the rating information of the invading property of Lung neoplasm attributive character information representation Lung neoplasm profit.

Optionally, the acquisition information acquisition unit, comprising:

Image obtains subelement, for obtaining lung's CT scan CT image of target patient；

Image zooming-out subelement, for being extracted in lung's CT scan CT image of the target patient Lung neoplasm image, obtains target information.

Optionally, the default feature extraction deep learning model includes convolutional layer, pond layer and full articulamentum, wherein The convolutional layer is used to learn the Feature Mapping of the Lung neoplasm CT figure of input, and the pond layer is used to reduce the depth The computation complexity of model is practised, the full articulamentum is used to format the Feature Mapping that study obtains, so that obtaining Characteristic information can be identified by the object classifiers；

Second extraction unit, comprising:

Feature extraction subelement is obtained for carrying out feature extraction to the target information using preset data mining algorithm Initial characteristics information is obtained, the initial characteristics information includes shape feature, strength characteristic, textural characteristics and high-order feature；

Subelement is analyzed, for carrying out feature selecting to the initial characteristics, and the initial characteristics after selection are carried out special Sign analysis, obtains second feature information；The signature analysis includes one in correlation analysis, clustering and principal component analysis Kind is a variety of.

Optionally, the vector generation unit includes:

Information processing subelement is wrapped for being handled using the full articulamentum the fisrt feature information One-dimensional characteristic vector containing N number of value；

Coded sub-units, it is raw for being encoded using the preset data mining algorithm to the second feature information At the n dimensional vector n for including M value；

Generate subelement, for according to channel rank to including M a described in the one-dimensional characteristic vector sum comprising N number of value One n dimensional vector n of value is combined superposition, generates dimension vector, and the dimension vector is a n dimensional vector n of M+N value size, In, N and M is the different positive integer of numerical value.

Compared to the prior art, it the present invention provides the extracting method and system of a kind of Lung neoplasm attributive character information, obtains Take target information；Feature extraction is carried out to the target information using default feature extraction deep learning model, obtains the first spy Reference breath；Feature extraction is carried out to the target information using preset data mining algorithm, obtains second feature information；According to institute Dimension vector is generated after stating fisrt feature information and second feature information fusion；The dimension vector is input to target point In class device, the Lung neoplasm attributive character information of the target information is obtained.Deep learning model sum number is used in the present invention According to mining algorithm simultaneously to target information processing, the low dimensional pixel that can be obtained in target information based on deep learning model is special Sign, high-dimensional semantic information feature is obtained based on data mining algorithm, so that the characteristic information of acquisition it is more complete and Accurately, it is then based on the fusion and classification to features described above information, so that the Lung neoplasm attributive character information obtained is more accurate, In this way to realize that the classification of invading property of Lung neoplasm profit provides accurate reference information, so as to promote point of invading property of Lung neoplasm profit The accuracy of grade.

Detailed description of the invention

In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis The attached drawing of offer obtains other attached drawings.

Fig. 1 is a kind of flow diagram of the extracting method of Lung neoplasm attributive character information provided in an embodiment of the present invention；

Fig. 2 is a kind of example architecture figure of deep learning model provided in an embodiment of the present invention；

Fig. 3 is a kind of schematic diagram that attribute information is extracted using data mining algorithm provided in an embodiment of the present invention；

Fig. 4 is the schematic diagram of the processing mode provided in an embodiment of the present invention for combining data mining and deep learning；

Fig. 5 is a kind of structural schematic diagram of the extraction system of Lung neoplasm attributive character information provided in an embodiment of the present invention.

Specific embodiment

Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall within the protection scope of the present invention.