阅读说明：本技术 一种基于穆勒偏振技术的生物组织结构分类系统 (Biological tissue structure classification system based on Mueller polarization technology ) 是由 李艳秋 王文爱 陈国强 于 2021-07-27 设计创作，主要内容包括：本发明提供一种基于穆勒偏振技术的生物组织结构分类系统,首先获得待分类生物组织对应的穆勒矩阵及其穆勒偏振参数组合,再根据穆勒偏振参数组合的统计量构建偏振特征矩阵,最后由支持向量机对基于穆勒偏振技术获取的且由多个偏振特征量形成的偏振特征矩阵进行联合评估,得到生物组织的所属类别；本发明联合多个标量偏振参数(标量相位延迟、标量偏振度、标量退偏、标量双向衰减等)和多个矢量偏振参数(矢量相位延迟、矢量偏振度、矢量双向衰减等)作为关键特征来对生物组织进行分类,实现对生物组织微观结构全面的定性分析及精准的定量鉴别,能够大大减小随机因素的影响,提高分类的准确率。(The invention provides a biological tissue structure classification system based on a Mueller polarization technology, which comprises the steps of firstly obtaining a Mueller matrix corresponding to a biological tissue to be classified and a Mueller polarization parameter combination thereof, then constructing a polarization characteristic matrix according to statistics of the Mueller polarization parameter combination, and finally carrying out combined evaluation on the polarization characteristic matrix which is obtained based on the Mueller polarization technology and is formed by a plurality of polarization characteristic quantities by a support vector machine to obtain the category of the biological tissue; the invention combines a plurality of scalar polarization parameters (scalar phase delay, scalar polarization degree, scalar depolarization, scalar bidirectional attenuation and the like) and a plurality of vector polarization parameters (vector phase delay, vector polarization degree, vector bidirectional attenuation and the like) as key characteristics to classify biological tissues, realizes comprehensive qualitative analysis and accurate quantitative identification of the microstructure of the biological tissues, can greatly reduce the influence of random factors and improve the accuracy of classification.)

1. A biological tissue structure classification system based on a Mueller polarization technology is characterized by comprising a Mueller matrix acquisition module, a matrix decomposition module, a polarization feature extraction module and an M classification support vector machine, wherein M is at least 2;

the Mueller matrix acquisition module is used for acquiring a Mueller matrix of the biological tissue sample;

the matrix decomposition module decomposes the Mueller matrix by adopting a polar decomposition method to obtain a Mueller polarization parameter combination, wherein the Mueller polarization parameter combination is obtainedThe Mueller polarization parameter combination comprises scalar bidirectional attenuation D and vector bidirectional attenuationHorizontal line bidirectional attenuation component D_H45 DEG line bidirectional attenuation component D₄₅Circular bidirectional attenuation component D_CHorizontal line depolarization component Δ of scalar depolarization Δ_H45 DEG line depolarization component delta₄₅Circular depolarization component Δ_CDegree of scalar polarization, degree of vector polarizationHorizontal linear polarization degree component P_H45 degree linear polarization degree component P₄₅A circular polarization component P_CScalar phase delay R, vector phase delayHorizontal line phase delay component R_H45 DEG line phase delay component R₄₅Circular phase delay component R_CAnd a fast axis azimuth;

the polarization feature extraction module is used for respectively obtaining a statistic combination corresponding to each Mueller polarization parameter in the Mueller polarization parameter combinations and forming a polarization feature matrix by the statistic combination corresponding to each Mueller polarization parameter, wherein the statistic combination comprises a mean value, a standard deviation, an entropy, a skewness, a kurtosis and a quartile difference;

the M-class support vector machine is used for receiving the polarization feature matrix and outputting the class of the biological tissue sample and the probability of the biological tissue sample belonging to the class.

2. The system for classifying biological tissue structures based on the muller polarization technique according to claim 1, further comprising a true color image obtaining module, a category judging module, and a pathological analysis result obtaining module;

the true color image acquisition module is used for carrying out bidirectional vector attenuationHorizontal line bidirectional attenuation component D_H45 DEG line bidirectional attenuation component D₄₅Circular bidirectional attenuation component D_CMapping into three channels of green, red and blue, and synthesizing a vector bidirectional attenuation true color image by adopting a true color image processing algorithm;

manually judging the category of the vector bidirectional attenuation true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector bidirectional attenuation true color image, and if so, the category of the output of the M-type support vector machine is the final category of the biological tissue sample;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue sample when the category of the M classification support vector machine output is different from the category of the biological tissue sample obtained according to the vector bidirectional attenuation true color image, and taking the pathological analysis result as the final category of the biological tissue sample.

3. The system for classifying biological tissue structures based on the muller polarization technique according to claim 1, further comprising a true color image obtaining module, a category judging module, and a pathological analysis result obtaining module;

the true color image acquisition module is used for converting the vector polarization degreeHorizontal linear polarization degree component P_H45 degree linear polarization degree component P₄₅A circular polarization component P_CMapping into three channels of green, red and blue, and synthesizing a vector polarization degree true color image by adopting a true color image processing algorithm;

manually judging the category of the vector polarization true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector polarization degree true color image, and if so, the category of the output of the M-type support vector machine is the final category of the biological tissue sample;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue sample when the category of the M classified support vector machine output is different from the category of the biological tissue sample obtained according to the vector polarization degree true color map, and taking the pathological analysis result as the final category of the biological tissue sample.

4. The system for classifying biological tissue structures based on the muller polarization technique according to claim 1, further comprising a true color image obtaining module, a category judging module, and a pathological analysis result obtaining module;

the true color image acquisition module is used for delaying the vector phaseHorizontal line phase delay component R_H45 DEG line phase delay component R₄₅Circular phase delay component R_CMapping into three channels of green, red and blue, and synthesizing a vector phase delay true color image by adopting a true color image processing algorithm;

manually judging the category of the vector phase delay true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector phase delay true color image, and if so, the category of the output of the M-type support vector machine is the final category of the biological tissue sample;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue sample when the category of the M classification support vector machine output is different from the category of the biological tissue sample obtained according to the vector phase delay true color map, and taking the pathological analysis result as the final category of the biological tissue sample.

5. The Mueller polarization technology based biological tissue structure classification system as claimed in any one of claims 2 to 4, wherein when the class to which the M-class support vector machine outputs is different from the class to which the M-class support vector machine belongs, if a pathological analysis result of the biological tissue sample is the same as the class to which the true color image belongs, the biological tissue sample is re-collected to train the M-class support vector machine;

if the pathology analysis result of the biological tissue sample is the same as the class output by the M classification support vector machine, re-mining the structural information of the true color image by combining the pathology analysis, and perfecting the artificial interpretation theory of the vector Mueller polarization parameter true color image.

6. The system according to claim 1, further comprising a training module for training an M-class support vector machine, wherein the training module comprises a dataset acquisition unit, a matrix decomposition unit, a polarization feature extraction unit, a support vector machine acquisition unit, and a support vector machine verification unit;

the data set acquisition unit is used for acquiring biological tissue samples of M categories which are classified and judged according to pathology, delineating at least one detection area for each biological tissue sample, and randomly dividing each detection area corresponding to each category into a training set and a verification set according to a set proportion;

the Mueller matrix acquisition module is used for acquiring a Mueller matrix of each detection area;

the matrix decomposition unit is used for decomposing the Mueller matrices of each detection area by adopting a polar decomposition method to obtain the Mueller polarization parameter combinations corresponding to each detection area;

the polarization characteristic extraction unit is used for respectively acquiring a statistic combination corresponding to each Mueller polarization parameter in the Mueller polarization parameter combinations corresponding to each detection area to obtain a polarization characteristic matrix corresponding to each detection area;

the support vector machine acquisition unit is used for taking the polarization characteristic matrix corresponding to each detection area in the training set as input, taking the category of each detection area in the training set as output, and training the M-class support vector machine to obtain an alternative M-class support vector machine;

the support vector machine verification unit is used for inputting the polarization characteristic matrix corresponding to each detection area in the verification set into the alternative M classification support vector machine to obtain the classification result of each detection area in the verification set and the accuracy of the classification result of each detection area in the verification set by the alternative M classification support vector machine.

7. The Mueller polarization technology based biological tissue structure classification system of claim 6, wherein the support vector machine obtaining unit trains more than two M classification support vector machines to obtain more than two alternative M classification support vector machines, and each M classification support vector machine has different model parameters;

the support vector machine verification unit verifies more than two alternative M classification support vector machines, and takes the alternative M classification support vector machine with the highest accuracy as the final M classification support vector machine.

8. The system according to claim 6, further comprising a support vector machine evaluation module and a support vector machine adjustment module, wherein the support vector machine evaluation module comprises a data set acquisition subunit, a matrix decomposition subunit, a polarization feature extraction subunit, and a support vector machine test subunit;

the data set acquisition subunit is used for acquiring biological tissue samples of M categories which are not classified and judged according to pathology, and is also used for delineating at least one detection area for each biological tissue sample to obtain a test set;

the Mueller matrix acquisition module is used for acquiring a Mueller matrix of each detection area in the test set;

the matrix decomposition subunit decomposes the Mueller matrices of each detection area in the test set by adopting a polar decomposition method to obtain a Mueller polarization parameter combination corresponding to each detection area in the test set;

the polarization characteristic extraction subunit is used for respectively acquiring a statistic combination corresponding to each Mueller polarization parameter in the Mueller polarization parameter combinations corresponding to each detection area in the test set to obtain a polarization characteristic matrix corresponding to each detection area;

the support vector machine testing subunit is used for inputting the polarization characteristic matrix corresponding to each detection area in the test set into the alternative M classification support vector machine to obtain the classification result of each detection area in the test set and the probability corresponding to each classification result of the alternative M classification support vector machine;

the support vector machine adjusting module is used for generating an evaluation report according to the classification result of each detection area in the test set and the pathological analysis, adjusting the model parameters of the alternative M classification support vector machine according to the evaluation report, and then retraining the adjusted alternative M classification support vector machine by the training module until the accuracy rate meets the set requirement.

9. The Mueller polarization based biological tissue structure classification system of any one of claims 1 to 4 and 6 to 8, wherein M is 3, and the three categories are malignant lesion tissue, benign lesion tissue and normal tissue.

10. The tissue structure classification system based on the mueller polarization technology as claimed in any one of claims 1 to 4 and 6 to 8, wherein the mueller matrix acquisition module is an X-ray transmission type mueller polarization imager, an extreme ultraviolet transmission type mueller polarization imager, a deep ultraviolet transmission type mueller polarization imager, an ultraviolet transmission type mueller polarization imager, a visible light transmission type mueller polarization imager, an infrared band transmission type mueller polarization imager, an X-ray reflection type mueller polarization imager, an extreme ultraviolet reflection type mueller polarization imager, a deep ultraviolet reflection type mueller polarization imager, an ultraviolet reflection type mueller polarization imager, a visible light reflection type mueller polarization imager, an infrared band reflection type mueller polarization imager, a confocal mueller polarization imager, or a non-linear mueller polarization imager.

Technical Field

The invention belongs to the technical field of polarization measurement, and particularly relates to a biological tissue structure classification system based on a Mueller polarization technology.

Background

The interaction of light and a medium can generate scattering, and the change of the polarization state of photons in the scattering process is closely related to the microstructure of the scattering medium. Most biological tissues are high scattering media, but light loses originally carried polarization information through multiple scattering, and the contrast and resolution of imaging are affected. The Mueller polarization technology inhibits multiple scattering and loses the contribution of 'diffusion photons' of the original polarization state to an image by properly screening the polarization state of photons, and improves the effects of 'ballistic photons' and 'snake photons' of less scattering of the original polarization, thereby improving the image quality, improving the imaging contrast of a tissue shallow surface layer and reflecting a scattering medium microstructure. The Mueller polarization imaging is used as a non-marking and non-destructive detection technology capable of comprehensively reflecting the polarization optical characteristics of biological media, and has unique advantages in the field of biomedicine.

The mueller polarization imager is designed and built based on the principle that a polarization device modulates the polarization state of incident light, and can realize transmission-type and reflection-type detection of a sample, as shown in fig. 1, a common mueller polarization detection system comprises an illumination module, a polarization detection module, a CCD imaging module and a data processing module. The lighting module includes: a light source 101; a polarization detection module: comprises a Polarization State Generator (PSG) composed of a linear polarizer 102 and a quarter-wave plate 103 which are sequentially arranged, a condenser 104, a sample stage 105, a microscope objective 106, and a Polarization State Analyzer (PSA) composed of a quarter-wave plate 107 and a linear polarizer 108 which are sequentially arranged; the CCD imaging module selects a CCD camera 109 which meets the highest resolution of the objective lens; the data processing module 110 is mainly composed of parameter setting, image acquisition, data processing, result display and storage and the like. In the Mueller polarization detection, a 16-time rotating wave plate measurement method is adopted, 16 intensity images acquired by a CCD are sent to a data processing module, a 4 x 4 Mueller matrix image of a sample is further processed to obtain polarization characterization parameters of the sample, such as scalar depolarization, phase delay and bidirectional attenuation, and the like, and the polarization characteristics and structural information of the sample are analyzed. In order to improve the measurement accuracy of the Mueller polarization imager, an Eigenvalue Calibration Method (ECM) is adopted to calibrate the measurement device, so that high-accuracy and high-resolution Mueller polarization imaging of biological tissues can be realized.

In the process of detecting biological tissue lesion by the Mueller polarization imaging technology, polarization information contained in a sample Mueller matrix is complex, and structural information of a corresponding sample is not easy to read, while a polar decomposition method can extract polarization parameters with definite physical significance from the sample Mueller matrix and can be widely used for qualitative and quantitative analysis, for example, linear phase delay can reflect the arrangement sequence of fibrous structures such as protein, and the phase delay fast axis azimuth angle can reflect the orientation distribution of the fibrous structures. However, the current research has the following problems:

firstly, conventionally, a scalar polarization image is obtained by using a polar decomposition method, such as scalar phase delay, depolarization, two-way attenuation, linear phase delay, circular delay, linear depolarization, circular depolarization image and the like, a sample structure analysis is performed from a single scalar polarization parameter image, only the trend of the polarization parameter changing along with the size of a tissue lesion and the approximate distribution of a tissue structure influencing the polarization parameter can be observed, for example, only the trend of the phase delay value changing along with the different degrees of the tissue lesion can be qualitatively observed through the scalar phase delay image, or the approximate distribution of a collagen fiber structure influencing the phase delay can be observed, but the scalar phase delay image mixes a vertical distribution fiber structure reflected by a horizontal phase delay parameter, a fiber structure distributed along a 135-degree direction reflected by a 45-degree phase delay parameter, a fiber spiral structure reflected by a circular delay parameter and the like together, the details of the arrangement of the respective fiber structures cannot be observed from the images thereof, and the structural information provided is not comprehensive enough.

Second, the current quantitative analysis method combined with the mueller polarization imaging technique includes: 1. calculating statistical parameters (mean value, median, standard deviation, skewness, kurtosis and the like) of a Mueller matrix and an intensity graph (or a spatial frequency spectrogram) of a decomposition parameter, and then carrying out statistical analysis, wherein the method can realize quantitative classification from a certain characteristic quantity under a single angle, but is greatly influenced by random factors, cannot realize multi-parameter combined characterization, and the classification accuracy is determined by the quantity of sample quantities; 2. the method can be used for carrying out combined characterization on multiple parameters, is less influenced by random factors, realizes quantitative classification to a certain degree, and also has the limitation that the accuracy is determined by the sample amount.

Disclosure of Invention

In order to solve the problems, the invention provides a biological tissue structure classification system based on the Mueller polarization technology, which can improve the contrast of a polarization image, realize more comprehensive qualitative analysis on a biological tissue microstructure and more accurate quantitative identification and classification on different types of tissues.

A biological tissue structure classification system based on a Mueller polarization technology comprises a Mueller matrix acquisition module, a matrix decomposition module, a polarization feature extraction module and an M classification support vector machine, wherein M is at least 2;

the Mueller matrix acquisition module is used for acquiring a Mueller matrix of the biological tissue sample;

the matrix decomposition module decomposes the Mueller matrix by adopting a polar decomposition method to obtain a Mueller polarization parameter combination, wherein the Mueller polarization parameter combination comprises scalar bidirectional attenuation D and vector bidirectional attenuation DHorizontal line bidirectional attenuation component D_H45 DEG line bidirectional attenuation component D₄₅Circular bidirectional attenuation component D_CHorizontal line depolarization component Δ of scalar depolarization Δ_H45 DEG line depolarization component delta₄₅Circular depolarization component Δ C, scalar polarization degree, vector polarization degreeHorizontal linear polarization degree component P_H45 degree linear polarization degree component P₄₅A circular polarization component P_CScalar phase delay R, vector phase delayHorizontal line phase delay component R_H45 DEG line phase delay component R₄₅Circular phase delay component R_CAnd a fast axis azimuth;

the polarization feature extraction module is used for respectively obtaining a statistic combination corresponding to each Mueller polarization parameter in the Mueller polarization parameter combinations and forming a polarization feature matrix by the statistic combination corresponding to each Mueller polarization parameter, wherein the statistic combination comprises a mean value, a standard deviation, an entropy, a skewness, a kurtosis and a quartile difference;

the M-class support vector machine is used for receiving the polarization feature matrix and outputting the class of the biological tissue sample and the probability of the biological tissue sample belonging to the class.

Further, the biological tissue structure classification system based on the Mueller polarization technology further comprises a true color image acquisition module, a category judgment module and a pathological analysis result acquisition module;

the true color image acquisition module is used for carrying out bidirectional vector attenuationHorizontal line bidirectional attenuation component D_H45 DEG line bidirectional attenuation component D₄₅Circular bidirectional attenuation component D_CMapping into three channels of green, red and blue, and synthesizing a vector bidirectional attenuation true color image by adopting a true color image processing algorithm;

manually judging the category of the vector bidirectional attenuation true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector bidirectional attenuation true color image, and if so, the category of the output of the M-type support vector machine is the final category of the biological tissue sample;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue sample when the category of the M classification support vector machine output is different from the category of the biological tissue sample obtained according to the vector bidirectional attenuation true color image, and taking the pathological analysis result as the final category of the biological tissue sample.

Further, the biological tissue structure classification system based on the Mueller polarization technology further comprises a true color image acquisition module, a category judgment module and a pathological analysis result acquisition module;

the true color image acquisition module is used for converting the vector polarization degreeHorizontal linear polarization degree component P_H45 degree linear polarization degree component P₄₅A circular polarization component P_CMapping into three channels of green, red and blue, and synthesizing a vector polarization degree true color image by adopting a true color image processing algorithm;

manually judging the category of the vector polarization true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector polarization degree true color image, and if so, the category of the output of the M-type support vector machine is the final category of the biological tissue sample;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue sample when the category of the M classified support vector machine output is different from the category of the biological tissue sample obtained according to the vector polarization degree true color map, and taking the pathological analysis result as the final category of the biological tissue sample.

Further, the biological tissue structure classification system based on the Mueller polarization technology further comprises a true color image acquisition module, a category judgment module and a pathological analysis result acquisition module;

the true color image acquisition module is used for delaying the vector phaseHorizontal line phase delay component R_H45 DEG line phase delay component R₄₅Circular phase delay component R_CMapping into three channels of green, red and blue, and synthesizing a vector phase delay true color image by adopting a true color image processing algorithm;

manually judging the category of the vector phase delay true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector phase delay true color image, and if so, the category of the output of the M-type support vector machine is the final category of the biological tissue sample;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue sample when the category of the M classification support vector machine output is different from the category of the biological tissue sample obtained according to the vector phase delay true color map, and taking the pathological analysis result as the final category of the biological tissue sample.

Further, when the class to which the M classification support vector machine outputs is different from the class to which the M classification support vector machine obtains according to the true color image, if the pathology analysis result of the biological tissue sample is the same as the class to which the true color image obtains, the biological tissue sample is collected again to train the M classification support vector machine;

if the pathology analysis result of the biological tissue sample is the same as the class output by the M classification support vector machine, re-mining the structural information of the true color image by combining the pathology analysis, and perfecting the artificial interpretation theory of the vector Mueller polarization parameter true color image.

Further, the biological tissue structure classification system based on the Mueller polarization technology further comprises a training module for training an M classification support vector machine, wherein the training module comprises a data set acquisition unit, a matrix decomposition unit, a polarization feature extraction unit, a support vector machine acquisition unit and a support vector machine verification unit;

the data set acquisition unit is used for acquiring biological tissue samples of M categories which are classified and judged according to pathology, delineating at least one detection area for each biological tissue sample, and randomly dividing each detection area corresponding to each category into a training set and a verification set according to a set proportion;

the Mueller matrix acquisition module is used for acquiring a Mueller matrix of each detection area;

the matrix decomposition unit is used for decomposing the Mueller matrices of each detection area by adopting a polar decomposition method to obtain the Mueller polarization parameter combinations corresponding to each detection area;

the polarization characteristic extraction unit is used for respectively acquiring a statistic combination corresponding to each Mueller polarization parameter in the Mueller polarization parameter combinations corresponding to each detection area to obtain a polarization characteristic matrix corresponding to each detection area;

the support vector machine acquisition unit is used for taking the polarization characteristic matrix corresponding to each detection area in the training set as input, taking the category of each detection area in the training set as output, and training the M-class support vector machine to obtain an alternative M-class support vector machine;

the support vector machine verification unit is used for inputting the polarization characteristic matrix corresponding to each detection area in the verification set into the alternative M classification support vector machine to obtain the classification result of each detection area in the verification set and the accuracy of the classification result of each detection area in the verification set by the alternative M classification support vector machine.

Further, the support vector machine obtaining unit trains more than two M classified support vector machines to obtain more than two alternative M classified support vector machines, and the model parameters of each M classified support vector machine are different;

the support vector machine verification unit verifies more than two alternative M classification support vector machines, and takes the alternative M classification support vector machine with the highest accuracy as the final M classification support vector machine.

Further, the system for classifying the biological tissue structure based on the Mueller polarization technology further comprises a support vector machine evaluation module and a support vector machine adjustment module, wherein the support vector machine evaluation module comprises a data set acquisition subunit, a matrix decomposition subunit, a polarization feature extraction subunit and a support vector machine test subunit;

the data set acquisition subunit is used for acquiring biological tissue samples of M categories which are not classified and judged according to pathology, and is also used for delineating at least one detection area for each biological tissue sample to obtain a test set;

the Mueller matrix acquisition module is used for acquiring a Mueller matrix of each detection area in the test set;

the matrix decomposition subunit decomposes the Mueller matrices of each detection area in the test set by adopting a polar decomposition method to obtain a Mueller polarization parameter combination corresponding to each detection area in the test set;

the polarization characteristic extraction subunit is used for respectively acquiring a statistic combination corresponding to each Mueller polarization parameter in the Mueller polarization parameter combinations corresponding to each detection area in the test set to obtain a polarization characteristic matrix corresponding to each detection area;

the support vector machine testing subunit is used for inputting the polarization characteristic matrix corresponding to each detection area in the test set into the alternative M classification support vector machine to obtain the classification result of each detection area in the test set and the probability corresponding to each classification result of the alternative M classification support vector machine;

the support vector machine adjusting module is used for generating an evaluation report according to the classification result of each detection area in the test set and the pathological analysis, adjusting the model parameters of the alternative M classification support vector machine according to the evaluation report, and then retraining the adjusted alternative M classification support vector machine by the training module until the accuracy rate meets the set requirement.

Further, the M is 3, and the three categories are malignant lesion tissue, benign lesion tissue, and normal tissue, respectively.

Further, the muller matrix obtaining module is an X-ray transmission type muller polarization imager, an extreme ultraviolet transmission type muller polarization imager, a deep ultraviolet transmission type muller polarization imager, an ultraviolet transmission type muller polarization imager, a visible light transmission type muller polarization imager, an infrared band transmission type muller polarization imager, an X-ray reflection type muller polarization imager, an extreme ultraviolet reflection type muller polarization imager, a deep ultraviolet reflection type muller polarization imager, an ultraviolet reflection type muller polarization imager, a visible light reflection type muller polarization imager, an infrared band reflection type copolymerization muller polarization imager, a focus muller polarization imager, or a non-linear muller polarization imager.

Has the advantages that:

1. the invention provides a biological tissue structure classification system based on a Mueller polarization technology, which comprises the steps of firstly obtaining a Mueller matrix corresponding to a biological tissue to be classified and a Mueller polarization parameter combination thereof, then constructing a polarization characteristic matrix according to statistics of the Mueller polarization parameter combination, and finally carrying out combined evaluation on the polarization characteristic matrix which is obtained based on the Mueller polarization technology and is formed by a plurality of polarization characteristic quantities by a support vector machine to obtain the category of the biological tissue; the invention combines a plurality of scalar polarization parameters (scalar phase delay, scalar polarization degree, scalar depolarization, scalar bidirectional attenuation and the like) and a plurality of vector polarization parameters (vector phase delay, vector polarization degree, vector bidirectional attenuation and the like) as key characteristics to classify biological tissues, realizes comprehensive qualitative analysis and accurate quantitative identification of the microstructure of the biological tissues, can greatly reduce the influence of random factors and improve the accuracy of classification.

2. The invention provides a biological tissue structure classification system based on a Mueller polarization technology, which is characterized in that a true color image corresponding to a biological tissue is obtained according to Mueller polarization parameters, and compared with a traditional scalar polarization image, the true color image can provide more microstructure details of a biological tissue sample; secondly, more comprehensive qualitative analysis can be carried out by observing the true color image, whether the result of the qualitative analysis is consistent with the prediction result of the quantitative classification of the support vector machine is judged, if not, the pathological analysis result of the biological tissue is introduced for final judgment, so that more bases are provided for machine classification of the support vector machine, and the classification accuracy is further improved.

3. The invention provides a biological tissue structure classification system based on a Mueller polarization technology, vector phase delay, polarization degree, bidirectional attenuation true color image ratio scalar phase delay, polarization degree and bidirectional attenuation image contain more sample microstructure details, so that a new idea can be provided for establishing a bridge between Mueller polarization imaging and clinical application by adopting vector polarization parameters to construct a corresponding true color image; meanwhile, the true color image is very visual, so that the problem that the polarized image is difficult to read by a layman can be solved, and a bridge between optics and other disciplines is established.

4. The invention provides a biological tissue structure classification system based on a Mueller polarization technology, the performance of a traditional statistical analysis method based on the Mueller polarization imaging technology can be theoretically ensured only when the sample amount tends to be infinite, and a support vector machine has good small sample learning capacity, so that the support vector machine can obtain the optimal solution under the existing limited sample information by combining a polarization characteristic matrix formed by statistics corresponding to a plurality of polarization parameters.

5. The invention provides a biological tissue structure classification system based on a Mueller polarization technology, a required biological tissue sample does not need artificial treatment such as dyeing and glue sealing, time and cost are saved, and classification efficiency can be effectively improved on the premise of ensuring classification accuracy.

Drawings

Fig. 1 is a schematic diagram of a conventional transmission-type mueller polarization imaging system;

101-light source, 102-linear polarizer, 103-quarter wave plate, 104-condenser, 105-sample stage, 106-microscope objective, 107-quarter wave plate, 108-linear polarizer, 109-CCD camera, 110-data processing module;

fig. 2 is a schematic block diagram of a biological tissue structure classification system based on the mueller polarization technology according to the present invention;

fig. 3 is a schematic flowchart of a method for generating a true color image with vector polarization parameters according to an embodiment of the present invention;

fig. 4 is a flowchart illustrating a method and system for tissue sample classification evaluation according to an embodiment of the present invention.

Detailed Description

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

Example one

As shown in fig. 2, the present invention provides a biological tissue structure classification system based on the mueller polarization technology, which is used for performing high-contrast display and high-accuracy classification evaluation on a biological tissue sample, and includes a mueller matrix acquisition module, a matrix decomposition module, a polarization feature extraction module, an M classification support vector machine, a true color map acquisition module, a category judgment module, and a pathological analysis result acquisition module, where M is at least 2;

the Mueller matrix acquisition module is used for acquiring the Mueller matrix of the biological tissue to be classified according to the Mueller polarization imaging technology. For example, the biological tissue is a skin tissue, and the categories of the skin tissue are malignant lesion tissue (melanoma skin tissue), benign lesion tissue (pigmented nevus skin tissue), and normal skin tissue. Optionally, the muller matrix obtaining module may be an X-ray transmission type muller polarization imager, an extreme ultraviolet transmission type muller polarization imager, a deep ultraviolet transmission type muller polarization imager, an ultraviolet transmission type muller polarization imager, a visible light transmission type muller polarization imager, an infrared band transmission type muller polarization imager, an X-ray reflection type muller polarization imager, an extreme ultraviolet reflection type muller polarization imager, a deep ultraviolet reflection type muller polarization imager, an ultraviolet reflection type muller polarization imager, a visible light reflection type muller polarization imager, an infrared band reflection type copolymerization muller polarization imager, a focus muller polarization imager, a nonlinear muller polarization imager, or other autonomously constructed optical systems capable of obtaining a muller matrix.

The matrix decomposition module decomposes the Mueller matrix by adopting a polar decomposition method to obtain a Mueller polarization parameter combination, wherein the Mueller polarization parameter combination comprises scalar bidirectional attenuation D and vector bidirectional attenuation DHorizontal line bidirectional attenuation component D_H45 DEG line bidirectional attenuation component D₄₅Circular bidirectional attenuation component D_CHorizontal line depolarization component Δ of scalar depolarization Δ_H45 DEG line depolarization component delta₄₅Circular depolarization component Δ_CDegree of scalar polarization, degree of vector polarizationHorizontal linear polarization degree component P_H45 degree linear polarization degree component P₄₅A circular polarization component P_CScalar phase delay R, vector phase delayHorizontal line phase delay component R_H45 DEG line phase delay component R₄₅Circular phase delay component R_CAnd fast axis azimuth.

The polarization feature extraction module is used for respectively obtaining a statistic combination corresponding to each polarization parameter in the Mueller polarization parameter combination, and combining the statistics corresponding to each polarization parameter to form a polarization feature matrix, wherein the statistic combination includes but is not limited to a mean value, a standard deviation, an entropy, a skewness, a kurtosis and a quartile.

And the M classification support vector machine is used for receiving the polarization characteristic matrix and outputting the class of the biological tissue to be classified and the probability of the biological tissue to belong to the class.

The true color image obtaining module, the category judging module and the pathological analysis result obtaining module are used for obtaining a true color image of the biological tissue; because true color maps can provide more microstructural details of the sample than traditional scalar polarization images, more comprehensive qualitative analysis is possible.

Further, after the M classification support vector machine outputs the category of the biological tissue to be classified, the invention adopts the true color map technology to carry out external evaluation, and the basic principle is as follows: after a region is predicted, a true color image of the region is generated, and qualitative analysis is carried out by observing the true color image, so that whether the prediction result is consistent with the just quantitative classification prediction result or not is judged. As shown in fig. 3, it is a flowchart for obtaining a true color map provided by the present invention, and specifically includes the following three implementation manners.

The first method comprises the following steps:

the true color image acquisition module is used for carrying out bidirectional vector attenuationHorizontal line bidirectional attenuation component D_H45 DEG line bidirectional attenuation component D₄₅Circular bidirectional attenuation component D_CMapping into three channels of green, red and blue, and adopting true color image processing algorithm to make the three channels synthesize vector two-way attenuation true color image;

manually judging the category of the vector bidirectional attenuation true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector bidirectional attenuation true color image, if so, the category of the output of the M-type support vector machine is the final category of the biological tissue;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue when the category of the M classification support vector machine output is different from the category of the biological tissue obtained according to the vector bidirectional attenuation true color image, and taking the pathological analysis result as the final category of the biological tissue.

And the second method comprises the following steps:

the true color image acquisition module is used for converting the vector polarization degreeHorizontal linear polarization degree component P_H45 degree linear polarization degree component P₄₅A circular polarization component P_CMapping into green, red and blue three channels respectively, and synthesizing the three channels into a vector polarization degree true color image by adopting a true color image processing algorithm;

manually judging the category of the vector polarization true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector polarization degree true color image, if so, the category of the output of the M-type support vector machine is the final category of the biological tissue;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue when the category of the M classified support vector machine output is different from the category of the biological tissue obtained according to the vector polarization degree true color map, and taking the pathological analysis result as the final category of the biological tissue.

And the third is that:

the true color image acquisition module is used for delaying the vector phaseHorizontal line phase delay component R_H45 DEG line phase delay component R₄₅Circular phase delay component R_CMapping into green, red and blue three channels, and synthesizing the three channels into vector phase delay true color image by adopting true color image processing algorithm;

manually judging the category of the vector phase delay true color image;

the category judgment module is used for judging whether the category of the output of the M-type support vector machine is the same as the category of the output of the M-type support vector machine obtained according to the vector phase delay true color image, if so, the category of the output of the M-type support vector machine is the final category of the biological tissue;

the pathological analysis result acquisition module is used for acquiring the pathological analysis result of the biological tissue when the category of the M classification support vector machine output is different from the category of the biological tissue obtained according to the vector phase delay true color map, and taking the pathological analysis result as the final category of the biological tissue.

It should be noted that, in the above three implementation manners, when the category to which the M-class support vector machine outputs is different from the category to which the M-class support vector machine belongs, if the pathological analysis result of the biological tissue is the same as the category to which the true color image belongs, the biological tissue sample is collected again to train the M-class support vector machine; if the pathology analysis result of the biological tissue is the same as the belonged category output by the M classification support vector machine, re-mining the structural information of the true color image by combining the pathology analysis, and perfecting the artificial interpretation theory of the true color image.

The detailed calculation process of the Mueller polarization parameter combination is obtained by decomposing the Mueller matrix by using a polar decomposition method based on Lu-Chipman theory by the matrix decomposition module.

M represents a Mueller matrix carrying sample information, and the specific process of Mueller matrix polar decomposition is as follows:

wherein m is₀₀～m₃₃For 16 subgraphs in a Mueller matrix, M_ΔFor a depolarization matrix, M_RIs a phase delay matrix, M_DIs a bidirectional attenuation matrix;

the scalar two-way attenuation D can be first determined by means of the mueller matrix:

bidirectional attenuation matrix M_DExpressed as:

wherein T represents transpose, T_uWhich represents the transmittance of unpolarized light,is a vector two-way attenuation.

Vector two-way attenuationThree components are respectively horizontal bidirectional attenuation D_H45 degree bidirectional attenuation D₄₅Circular bidirectional attenuation D_CSpecifically, the following are shown:

continuously solving a depolarization matrix, and firstly obtaining an intermediate matrix:

M'＝M_ΔM_R＝MM_D ^-1

where M ' is a 3X 3 matrix under M ', det (M ') determines the sign of the following formula, λ₁，λ₂，λ₃Is m '(m')^TThe characteristic root of (1):

then a de-biased Mueller matrix M can be obtained_Δ：

The scalar depolarization value Δ is:

degree of vector polarizationThree component horizontal line polarization degree P_HDegree of linear polarization P of 45 DEG₄₅Degree of circular polarization P_CThe calculation formula of (a) is as follows:

finally, the phase delay matrix M is obtained_R：

Scalar phase delay value R:

vector phase delayIs delayed by R_H45 DEG line phase delay R₄₅Circular phase delay R_CThe calculation formula of (a) is as follows:

the fast axis azimuth angle θ is calculated as follows:

wherein the content of the first and second substances,ε_ijkare the livid-odd victa symbols.

Example two

Based on the foregoing embodiments, this embodiment provides a subsystem for training an M-class support vector machine, where the subsystem includes a training module for training the M-class support vector machine, a support vector machine evaluation module, and a support vector machine adjustment module, where the training module includes a data set obtaining unit, a matrix decomposition unit, a polarization feature extraction unit, a support vector machine obtaining unit, and a support vector machine verification unit; the support vector machine evaluation module comprises a data set acquisition subunit, a matrix decomposition subunit, a polarization characteristic extraction subunit and a support vector machine testing subunit. Meanwhile, the work flow of the subsystem is shown in fig. 4, and the specific functions of each module are defined as follows:

the data set acquisition unit is used for acquiring M types of biological tissue samples which are classified and judged according to pathology, wherein the number of the biological tissue samples of each type can be set to be the same for preliminary control variables and research convenience, and the biological tissue samples are marked as N; the data set acquisition unit is also used for delineating at least one detection area for each biological tissue sample, setting each category to delineate K detection areas for research convenience, and randomly dividing the K detection areas corresponding to each category into a training set and a verification set according to a set proportion; for example, the data set acquiring unit may acquire N cases of skin melanoma, pigmented nevus and normal skin tissue samples classified and determined according to pathology in the pathology department of hospitals, classify and label each tissue sample to define a specific detection area, and generate K detection cases for each type. And randomly dividing K detection cases of malignant melanoma, pigmented nevus and normal skin into a training set and a verification set according to the quantity ratio I: J.

The Mueller matrix acquisition module is used for acquiring the Mueller matrix of each detection area.

And the matrix decomposition unit is used for decomposing the Mueller matrix of each detection area by adopting a polar decomposition method to obtain the Mueller polarization parameter combination corresponding to each detection area.

The polarization feature extraction unit is used for respectively obtaining a statistic combination corresponding to each polarization parameter in the Mueller polarization parameter combinations corresponding to each detection area, and obtaining a polarization feature matrix corresponding to each detection area.

The support vector machine obtaining unit is used for taking the polarization characteristic matrix corresponding to each detection area in the training set as input, taking the category of each detection area in the training set as output, and training the M-class support vector machine to obtain the alternative M-class support vector machine.

The support vector machine verification unit is used for inputting the polarization characteristic matrix corresponding to each detection area in the verification set into the alternative M classification support vector machine to obtain the classification result of each detection area in the verification set, the classification result of each detection area in the verification set and the accuracy between the classification category of each detection area in the verification set by the alternative M classification support vector machine.

The data set acquisition subunit is used for acquiring M types of biological tissue samples which are not classified and judged according to pathology, and the biological tissue samples of all types are the same in number and are marked as L; the data set acquisition unit is also used for delineating at least one detection area for each biological tissue sample to obtain a test set; for example, the data set acquiring subunit may acquire L cases of skin melanoma, pigmented nevus and normal skin tissue samples, which are not classified according to pathology in the pathology department of hospitals, and divide the L cases into test sets.

The Mueller matrix acquisition module is used for acquiring the Mueller matrix of each detection area in the test set.

And the matrix decomposition subunit decomposes the Mueller matrices of the detection areas in the test set respectively by adopting a polar decomposition method to obtain the Mueller polarization parameter combinations corresponding to the detection areas in the test set.

And the polarization characteristic extraction subunit is used for respectively acquiring a statistic combination corresponding to each polarization parameter in the Mueller polarization parameter combinations corresponding to each detection area in the test set to obtain a polarization characteristic matrix corresponding to each detection area.

And the support vector machine testing subunit is used for inputting the polarization characteristic matrix corresponding to each detection area in the test set into the alternative M classification support vector machine to obtain the classification result of the alternative M classification support vector machine on each detection area in the test set.

The support vector machine adjusting module is used for generating an evaluation report according to the classification result of each detection area in the test set and the pathological analysis, adjusting model parameters of the alternative M classification support vector machine, such as kernel function type and initial network parameters, according to the evaluation report, and then retraining the adjusted alternative M classification support vector machine by the training module until the accuracy rate meets the set requirement.

It should be noted that the support vector machine obtaining unit trains more than two M-class support vector machines at the same time, so as to obtain more than two alternative M-class support vector machines, and model parameters of each M-class support vector machine, such as kernel function types, are different from initial network parameters; the support vector machine verification unit verifies more than two alternative M classification support vector machines at the same time, and the alternative M classification support vector machine with the highest accuracy is used as the final M classification support vector machine.

Therefore, the collected biological tissue samples are divided into two categories of known pathological types and unknown pathological types according to the preset requirements, the samples with the known pathological types are randomly divided according to the preset requirements to form a training set and a verification set, and the samples with the unknown pathological types can be divided into a test set. Then, performing Mueller polarization detection on each sample of the training set and the verification set to obtain a Mueller matrix image, and solving three components of vector phase delay, vector polarization degree and vector bidirectional attenuation and other scalar polarization parameters by using a polar decomposition method; calculating the common statistical parameters (mean value, standard deviation, entropy, skewness, kurtosis, quartering difference and the like) of each polarization parameter to generate a polarization characteristic matrix; and finally, inputting the polarization characteristic quantity of the training set samples into a support vector machine for training, constructing a classification model for distinguishing different types of biological tissue samples, and verifying by using the sample data of the verification set to show the high-accuracy classification capability of the model.

The present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof, and it will be understood by those skilled in the art that various changes and modifications may be made herein without departing from the spirit and scope of the invention as defined in the appended claims.