Skin tissue image recognition method based on polarization imaging and related device

Abstract

The invention relates to the technical field of skin tissue image recognition, in particular to a skin tissue image recognition method based on polarization imaging and a related device. The method comprises the following steps: firstly, acquiring DoP and AoP images of an undyed skin tissue slice; then, extracting four polarization characteristic parameters, including a DoP histogram peak value, a DoP histogram slope, a high-polarization region average DoP and an AoP texture anisotropy index, of the DoP image and the AoP image; and finally, performing skin tissue image recognition according to the four polarization characteristic parameters to obtain a skin tissue image recognition result. According to the method, undyed skin tissue slices are adopted, and a dyeing step is not needed, so that the detection period is shortened; meanwhile, the mode that an existing identification means based on polarization imaging needs to depend on a Mueller matrix is broken through, the detection process is greatly simplified, and the problems that a traditional polarization image identification process is complex, cost is high and accuracy is poor are solved.

Description

Technical Field

[0001] This invention relates to the field of skin tissue image recognition technology, specifically to a skin tissue image recognition method and related apparatus based on polarization imaging. Background Technology

[0002] In the continuous development of medical research and clinical practice, the accurate detection and analysis of the microscopic structure of biological tissues remains a crucial link. This is not only related to the early detection of diseases, but also has a profound impact on accurate diagnosis, effective treatment, and prognostic assessment. Traditional medical testing techniques, such as routine imaging examinations (e.g., X-rays, CT scans, MRI) and biochemical tests, have played an important role in disease diagnosis, but each has certain limitations. Imaging examinations mainly focus on macroscopic morphological and structural changes of tissues and organs. For early, minute lesions or changes at the microscopic level, their sensitivity and resolution often fail to meet clinical needs. While biochemical tests can provide information on some biomarkers, their ability to directly observe and analyze the microscopic structure of tissues is limited, and they cannot comprehensively and intuitively reflect the subtle changes within tissues.

[0003] Against this backdrop, polarization optics detection technology has gradually emerged due to its unique advantages. Polarization optics primarily studies the polarization characteristics of light waves. Biological tissues, due to the anisotropy of their microstructure (such as cell arrangement and fiber orientation), exhibit specific changes in the polarization state of incident light. Polarization optics detection technology utilizes this characteristic to reveal the microstructural features of biological tissues by detecting and analyzing the polarization information of light after it has passed through them. Compared to traditional detection techniques, polarization optics detection technology has the unique advantage of high sensitivity to the microstructure of biological tissues. It can penetrate to the cellular and molecular levels, effectively capturing subtle structural changes in biological tissues caused by disease, which is of great significance for early diagnosis and precise treatment of diseases. Therefore, polarization optics detection technology has become a highly promising and valuable tool in clinical research and diagnosis, bringing new opportunities and hope to the development of the medical field.

[0004] Currently, polarization optics detection technology mainly focuses on Mueller matrix polarization imaging. The Mueller matrix is ​​a 4×4 matrix that comprehensively describes the changes in polarization state of light as it propagates through a medium. It contains rich polarization information, and by measuring and analyzing the Mueller matrix, we can gain a deeper understanding of the polarization characteristics of biological tissues, thereby revealing their microstructure information. Existing technologies mainly include cancerous tissue identification methods based on the full element curves of Mueller matrix imaging and methods based on the rotational invariant parameters of the Mueller matrix and the texture features of the gray-level co-occurrence matrix. The core of the cancerous tissue identification method based on the full element curves of Mueller matrix imaging is to directly extract the full element curve features of 16 elements from the full Mueller matrix data to achieve the identification of cancerous tissue from normal tissue. However, it has core drawbacks: on the one hand, it relies on a full Mueller matrix measurement system, which has a complex equipment structure and high hardware costs; on the other hand, the full element curve analysis does not transform matrix elements into parameters with clear physical meaning, and the results need to be interpreted by polarization optics professionals, which is difficult to meet the needs of rapid clinical diagnosis; in addition, due to the interference of the background structure of the optical platform's light-shielding cloth, its classification accuracy for lung cancer is only 89.59%, and the classification accuracy for breast cancer at high and low magnification is 77.5% and 87.8% respectively, indicating insufficient detection robustness; although the method of cancerous tissue identification based on Mueller matrix rotation invariant parameters and gray-level co-occurrence matrix texture features has gotten rid of the direct dependence on full Mueller matrix elements and improved the readability of the results through polarization parameters with clear physical meaning, it still has significant limitations: first, there are 23 initial polarization parameters, and although 18 effective parameters are subsequently selected and retained, there is still a problem of parameter redundancy; second, the texture analysis process is cumbersome, requiring the calculation of 10 texture features for each of the 18 polarization parameter images, and each polarization parameter needs to be adapted to specific texture features, lacking a unified standard, which not only increases the computational burden but also raises the operational threshold for clinical application.

[0005] In summary, existing polarization-optical detection techniques for cancerous tissue face numerous technical challenges in practical applications, including high equipment costs, difficulties in data interpretation, low detection accuracy, parameter redundancy, and operational complexity. There is an urgent need for a polarization image recognition method that offers shorter detection cycles, more accurate results, and does not rely on complex system structures and calculations, in order to promote the clinical application of polarization-optical detection technology. Summary of the Invention

[0006] To address the problems of complex, costly, and inaccurate polarization image recognition processes in existing technologies, this invention provides a skin tissue image recognition method and related apparatus based on polarization imaging.

[0007] To achieve the above objectives, the present invention employs the following technical solution: This invention provides a skin tissue image recognition method based on polarization imaging, comprising: acquiring polarization degree (DoP) images and polarization angle (AoP) images of unstained skin tissue sections; Polarization feature parameters of the first region of interest are extracted from the DoP image; the first region of interest includes the region covering the spinous layer, granular layer and basal layer of the epidermis excluding the stratum corneum; the polarization feature parameters of the first region of interest include the peak value of the DoP histogram, the slope of the DoP histogram and the average DoP of the high polarization region; The polarization feature parameters of the second region of interest (ROI) are extracted from the AoP image. The second ROI includes a first cropping unit and a second cropping unit. The first cropping unit is located in the spinous layer and is capable of capturing the complete cell texture. The second cropping unit is located in the basal layer and is capable of capturing the complete cell texture. The polarization feature parameters of the second ROI include the AoP texture anisotropy index. Based on the polarization feature parameters of the first and second regions of interest, skin tissue image recognition is performed to obtain the skin tissue image recognition result.

[0008] Optionally, the method for obtaining DoP and AoP images of unstained skin tissue sections is as follows: Obtain linearly polarized images of unstained skin tissue sections at 0°, 45°, 90°, and 135°. Calculate the Stokes parameters based on the linear polarization images at 0°, 45°, 90°, and 135°. Based on the Stokes parameters, DoP and AoP images of unstained skin tissue sections were obtained.

[0009] Optionally, the method for obtaining the peak value of the DoP histogram is as follows: A statistical histogram is plotted for the polarization degree values ​​of all pixels in the first region of interest, and the probability density corresponding to the highest point of the plotted histogram is taken as the peak value of the DoP histogram.

[0010] Optionally, the method for obtaining the slope of the DoP histogram is as follows: A statistical histogram of polarization values ​​for all pixels within the first region of interest is plotted, and the envelope of the histogram is fitted. Extract the peak points of the envelope of the histogram, and perform linear fitting on the descending segment from the peak points to obtain a straight line. Calculate the slope of this straight line to obtain the slope of the DoP histogram.

[0011] Optionally, the method for obtaining the average DoP in the high polarization region is as follows: In the first region of interest, all pixels are sorted in descending order of DoP value, starting from the maximum value and moving downwards, selecting the top 5%. The region containing each pixel is a highly polarized region. The total number of pixels in the first region of interest is given; the average value of all pixels in the high polarization region is calculated as the average DoP of the high polarization region.

[0012] Optionally, the method for obtaining the AoP texture anisotropy index is as follows: Obtain the AoP images corresponding to the first and second cropping units; In the AoP images corresponding to the first and second cropping units, the occurrence frequency of gray value combinations of adjacent pixels with a geometric distance of 1 pixel in four spatial directions is counted and used as matrix elements to construct the gray-level co-occurrence matrix of the first and second cropping units in four spatial directions; the four spatial directions are 0°, 45°, 90° and 135° spatial directions, respectively. Texture feature parameters of the first and second cropping units are extracted from the gray-level co-occurrence matrices in four spatial directions, respectively; the texture feature parameters include the contrast, energy, entropy and homogeneity of adjacent pixels; Based on the texture feature parameters of the first and second truncation units, the coefficients of variation of the texture feature parameters of the first and second truncation units in four spatial directions are calculated accordingly. The variation coefficients of the texture feature parameters of the first and second cutting units in the four spatial directions are multiplied by their respective preset weighting coefficients and then summed to obtain the directional scores of the first and second cutting units, respectively. The directional scores of the first and second interception units are weighted and fused twice to obtain the AoP texture anisotropy index.

[0013] Optionally, the method for performing skin tissue image recognition based on the polarization feature parameters of the first and second regions of interest to obtain the skin tissue image recognition result is as follows: The polarization feature parameters of the first and second regions of interest are input into a preset logistic regression classification model, and the skin tissue image recognition result is output.

[0014] The present invention also provides a skin tissue image recognition system based on polarization imaging, comprising: DoP and AoP image acquisition unit, used to acquire DoP and AoP images of unstained skin tissue sections; The first polarization feature parameter extraction unit is used to extract polarization feature parameters of the first region of interest in the DoP image; the first region of interest includes the region covering the spinous layer, granular layer and basal layer of the epidermis excluding the stratum corneum; the polarization feature parameters of the first region of interest include the peak value of the DoP histogram, the slope of the DoP histogram and the average DoP of the high polarization region. The second polarization feature parameter extraction unit is used to extract polarization feature parameters of the second region of interest in the AoP image; the second region of interest includes a first cropping unit and a second cropping unit; the first cropping unit is located in the spinous layer and is a cropping unit that can frame the complete cell texture; the second cropping unit is located in the basal layer and is a cropping unit that can frame the complete cell texture; the polarization feature parameters of the second region of interest include the AoP texture anisotropy index; The image recognition unit is used to perform skin tissue image recognition based on the polarization feature parameters of the first region of interest and the second region of interest, and to obtain the skin tissue image recognition result.

[0015] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the steps of the method described above.

[0016] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0017] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a skin tissue image recognition method based on polarization imaging. This method acquires DoP and AoP images of unstained skin tissue sections and extracts polarization feature parameters from different regions of interest. This allows for precise capture of changes in polarization characteristics caused by cell morphology, arrangement, and changes in intercellular matrix components and microstructure. By fusing multi-dimensional information (DoP and AoP information), a more comprehensive understanding of the polarization characteristics of skin tissue can be achieved, reducing errors and limitations that may arise from single-parameter analysis and improving the accuracy and reliability of image recognition. Specifically, this invention uses unstained skin tissue sections for polarization imaging and image recognition, avoiding the staining step, simplifying the operation process, shortening the detection cycle, reducing the requirements for technical personnel, and also reducing the cost of staining reagents and environmental pollution. Compared with traditional Mueller matrix polarization imaging technology, this method does not rely on a complex full Mueller matrix measurement system. It extracts only four polarization features from the DoP and AoP images, significantly reducing computational load while maintaining classification accuracy. Furthermore, extracting representative polarization feature parameters for different regions more accurately reflects the structural characteristics of skin tissue cells, reduces noise interference, and improves the accuracy of skin tissue image recognition.

[0018] This invention also provides a skin tissue image recognition system based on polarization imaging. This system, through the high integration of a DoP and AoP image acquisition unit, a first polarization feature parameter extraction unit, a second polarization feature parameter extraction unit, and an image recognition unit, achieves the acquisition of DoP and AoP images of unstained skin tissue sections, extraction of four polarization feature parameters, and skin tissue image recognition. The DoP and AoP image acquisition unit can accurately capture the unique polarization characteristics of skin tissue in the polarization images of unstained skin tissue sections, laying a solid foundation for subsequent accurate feature extraction and image recognition. The first and second polarization feature parameter extraction units have clearly defined functions, respectively extracting features from specific regions of interest in the DoP and AoP images. This enables rapid and accurate extraction of representative polarization feature parameters from complex image data, improving the speed and accuracy of data processing. By comprehensively analyzing various polarization characteristic parameters of the first and second regions of interest, the system can comprehensively evaluate skin tissue from different angles and levels. The system has a simple structure and is relatively easy to operate and maintain. It can reflect the basic microstructural characteristics of skin tissue and is not only suitable for the diagnosis of common skin diseases (such as skin cancer and psoriasis), but can also be extended to the detection and differential diagnosis of other skin diseases by adjusting the analysis model and recognition algorithm of the characteristic parameters. It has strong adaptability and scalability, and provides more comprehensive and in-depth information support for the diagnosis and treatment of skin diseases.

[0019] This invention also provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described method. The processor can quickly execute processes such as acquiring DoP and AoP images of unstained skin tissue sections, extracting four polarization feature parameters, and recognizing skin tissue images, ensuring the accuracy and efficiency of skin tissue image recognition. The computer program in the memory can be modified and optimized according to actual needs to adapt to different requirements for skin tissue image recognition.

[0020] A computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method described above. The computer-readable storage medium (such as a solid-state drive (SSD) or flash memory) has high-speed read capabilities, enabling rapid loading of the computer program into the processor for execution, ensuring efficient skin tissue image recognition. It also features flexibility and portability, high reliability and stability, support for large-scale data storage, ease of integration and expansion, reduced development and maintenance costs, high security, energy efficiency and environmental friendliness, support for various application scenarios, and promotion of standardization and normalization. These characteristics provide strong support for the diagnosis and early detection of skin diseases and have broad application prospects. Attached Figure Description

[0021] Figure 1 This is a schematic flowchart of a skin tissue image recognition method based on polarization imaging according to the present invention.

[0022] Figure 2 This is a schematic diagram of the selection of the first region of interest and the second region of interest in this invention; wherein, a is a schematic diagram of the selection of the first region of interest, and b is a schematic diagram of the selection of the second region of interest.

[0023] Figure 3 This is a schematic diagram showing the numerical comparison of key image features extracted in this embodiment of the invention in normal tissue and squamous cell carcinoma tissue of the skin.

[0024] Figure 4 This is a flowchart illustrating the training process of the logistic regression model in an embodiment of the present invention.

[0025] Figure 5 This is a graph showing the performance index variation trend of an embodiment of the present invention under different training set sizes.

[0026] Figure 6 This is a structural diagram of a skin tissue image recognition system based on polarization imaging according to the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0029] The present invention will be further described in detail below with reference to specific embodiments. These descriptions are for explanation purposes only and are not intended to limit the scope of the invention.

[0030] See Figure 1 This invention provides a skin tissue image recognition method based on polarization imaging, comprising: S1: Obtain the DoP and AoP images of unstained skin tissue sections, specifically: To obtain linearly polarized images of unstained skin tissue sections at 0°, 45°, 90°, and 135°, the following steps were taken: the unstained skin tissue section was fixed on the stage and positioned aligned with the epidermal layer; images were acquired using a polarizing camera, observed through a 20x microscope objective, and focused until the epidermal structure was clear; the automatic exposure function was turned off, and images were captured; the acquired images were segmented to obtain four linearly polarized images corresponding to the 0°, 45°, 90°, and 135° angles.

[0031] The 0°, 45°, 90° and 135° linear polarization images of the unstained skin tissue sections are preprocessed, including non-uniformity correction and brightness equalization, and the Stokes parameters are calculated based on the 0°, 45°, 90° and 135° linear polarization images. Based on the Stokes parameters, DoP and AoP images of unstained skin tissue sections were obtained.

[0032] In the preparation of the unstained skin tissue sections, the sample is directly collected, the tissue is fixed, embedded, sectioned and dewaxed, and then directly capped, without the need for staining and related steps.

[0033] S2: Extract polarization feature parameters of the first region of interest from the DoP image; the polarization feature parameters of the first region of interest include the peak value of the DoP histogram, the slope of the DoP histogram, and the average DoP in the high polarization region, specifically: See Figure 2 a. On the original image of the unstained skin tissue section, manually select a region covering the spinous layer, granular layer and basal layer of the epidermis except for the stratum corneum. Record the total number of pixels in this region as N, and take it as the first region of interest. The method for obtaining the peak value of the DoP histogram is as follows: A statistical histogram is plotted for the polarization values ​​of all pixels in the first region of interest. The probability density corresponding to the highest point of the plotted histogram is taken as the peak value of the DoP histogram, which reflects the most representative polarization level in the image.

[0034] The method for obtaining the slope of the DoP histogram is as follows: A statistical histogram of polarization values ​​for all pixels within the first region of interest is plotted, and the envelope of the histogram is fitted. Extract the peak points of the envelope of the histogram, and perform linear fitting on the descending segment from the peak points to obtain a straight line. Calculate the slope of this straight line to obtain the slope of the DoP histogram.

[0035] The method for obtaining the average DoP in the high polarization region is as follows: In the first region of interest, all pixels are sorted in descending order of DoP value, starting from the maximum value and moving downwards, selecting the top 5%. The region containing each pixel is a highly polarized region. The total number of pixels in the first region of interest is denoted as 1. The average value of all pixels in the high polarization region is calculated as the average DoP of the high polarization region. The average DoP of the high polarization region quantifies the typical polarization degree of the region with the strongest polarization characteristics in the first region of interest.

[0036] S3: Extract polarization feature parameters of the second region of interest from the AoP image; the polarization feature parameters of the second region of interest include the AoP texture anisotropy index, specifically: See Figure 2 b. In the original image of the unstained skin tissue section, manually select an area of ​​appropriate size that can frame the complete cell texture in the spinous layer and basal layer of the skin tissue, respectively, as the first and second cropping units. Obtain the AoP images corresponding to the first and second cropping units, defining four spatial directions: 0° (horizontal), 90° (vertical), 45° (diagonal direction from top right to bottom left), and 135° (diagonal direction from top left to bottom right). Count the frequency of grayscale value combinations between adjacent pixels with a geometric distance of one pixel in each of the four spatial directions, and construct a corresponding gray-level co-occurrence matrix. The matrix elements are denoted as... ,in, This represents the grayscale value of the previous pixel. It represents the grayscale value of the next pixel that is one pixel away in a certain spatial direction; Texture feature parameters of the first and second truncation units are extracted from the gray-level co-occurrence matrices in four spatial directions, respectively. These texture feature parameters include the contrast, energy, entropy, and homogeneity of adjacent pixels, specifically: Contrast: ; energy: ; entropy: ; Homogeneity: .

[0037] Based on the texture feature parameters of the first and second cutting units, the variation coefficients of the texture feature parameters of the first and second cutting units in four spatial directions are calculated accordingly. At this time, the first and second cutting units respectively obtain the variation coefficients in four spatial directions based on their respective texture feature parameters in four directions, and each variation coefficient corresponds to a preset weighting coefficient. The variation coefficients of the texture feature parameters of the first and second cutting units in the four spatial directions are multiplied by their respective preset weighting coefficients and then summed to obtain the directional scores of the first and second cutting units, respectively. The directional scores of the first and second extraction units are weighted and fused twice to obtain the AoP texture anisotropy index. The AoP texture anisotropy index represents the orderly arrangement and directional regularity of the microstructures of different dermal layers in the skin tissue.

[0038] See Figure 3To verify that the ability of the four extracted polarization feature parameters (in the figure, feature 1 corresponds to the average DoP in the high polarization region, feature 2 corresponds to the peak value of the DoP histogram, feature 3 corresponds to the slope of the DoP histogram, and feature 4 corresponds to the AoP texture anisotropy index) to distinguish between normal and cancerous skin tissues is not accidental, statistical significance tests were performed. Using the collected sample dataset, independent samples t-tests and Mann-Whitney U tests were conducted on the four feature values ​​of the normal tissue group and the squamous cell carcinoma tissue group, respectively. The test results showed that the differences of these four features between the two groups were all highly statistically significant (the p-values ​​were all less than 0.05).

[0039] S4: Based on the polarization feature parameters of the first and second regions of interest, skin tissue image recognition is performed to obtain the skin tissue image recognition result, specifically: The polarization feature parameters of the first and second regions of interest are input into a preset logistic regression classification model, and the skin tissue image recognition result is output. The logistic regression classification model is constructed in the following way: Step 1: Construct a dataset comprising several independent biological samples, half of which are pathologically confirmed skin cancer cell tissues, and the other half are normal skin cell tissues. Multiple unstained tissue sections are prepared for each biological sample, and multiple polarization images are acquired for each section to calculate DoP and AoP images, forming a set of image data. Polarization feature parameters are extracted from each set of images to form the initial dataset for model training and evaluation.

[0040] Step Two: Leave-one-out cross-validation. The initial dataset is stratified according to each biological source, divided into several mutually exclusive subsets. All data corresponding to each biological source is used as the test set, and the remaining data as the training set, for several rounds of training and validation. In each training round, based on the current training set data, stratified cross-validation is used to calculate the hyperparameter λ. The average accuracy of all layers is used as the metric to determine the hyperparameter for that round, which is then used as the initial λ. A search is then conducted in the vicinity of this λ, combining accuracy and coefficient balance to finally determine the optimal λ that balances performance and coefficient balance. Then, the weight coefficients for polarization feature fusion are calculated on all training set data in the current round. Finally, the trained model is applied to the validation set data, and the classification results for that round are recorded. After all rounds of training, the judgment results of all data are summarized to obtain the overall classification performance, ensuring the model's stability in distinguishing different samples.

[0041] Step 3: Systematically analyze the impact of training set size on performance. Under the K-fold cross-validation framework described in Step 2, the training set size is successively restricted to the amount of data corresponding to different numbers of biological sources, and the mean accuracy, sensitivity, and specificity of the model are evaluated at each size.

[0042] Step 4: Confirm the performance of the final model.

[0043] See Figure 4 Taking the training process of a logistic regression classifier with eight independent samples as an example, this method ensures the model has high discrimination accuracy and systematically determines the minimum effective training set size required for the model performance to reach a stable plateau, thereby achieving high-efficiency and high-precision model training. The specific operation process is as follows: Step 1: Building the dataset To establish the training and testing datasets, eight independent biological samples were first selected: four were pathologically confirmed squamous cell carcinoma tissues of the skin (from four different patients), and the other four were normal skin tissues (from four different anatomical sites). Multiple tissue sections were prepared for each biological sample, and multiple polarization images were acquired for each section to calculate DoP and AoP images, forming a set of image data. Four-dimensional feature vectors were extracted from each set of images. Ultimately, a total of 160 sets of feature vectors were obtained, constituting the initial dataset for model training and evaluation. All feature vectors are accompanied by the pathological classification label of their source biological sample.

[0044] Step 2: Leave-one-out cross-validation The dataset was stratified according to eight biological sources, dividing it into eight mutually exclusive subsets. The entire dataset from each single source was used as the test set, while the data from the remaining seven sources were used as the training set, for eight rounds of training and validation.

[0045] In each training round, based on the current 140 training datasets, a grid search is performed on the hyperparameter λ of the logistic regression model using three-fold cross-validation. The average accuracy of all layers is used as the metric to determine the hyperparameter for that round. Using the determined hyperparameter λ as the initial λ, a search is then conducted in its vicinity. Combining accuracy and coefficient balance as comprehensive metrics, the optimal λ that balances performance and coefficient balance is finally determined. Then, weight coefficients are calculated on all training datasets for the current round. Finally, the trained model is applied to 20 validation datasets, and the classification results for that round are recorded. After eight rounds of training, the judgment results for all data are summarized to obtain the overall classification performance, ensuring the model's stability in distinguishing different samples.

[0046] Step 3: System Analysis of the Impact of Training Set Size on Performance To quantify the relationship between model performance and training data size, and to verify its high data efficiency, a series of controlled experiments were conducted. Within the K-fold cross-validation framework described in step two, the training set size was sequentially limited to data amounts corresponding to 2 to 7 biological sources, and the mean accuracy, sensitivity, and specificity of the model were evaluated at each size. Figure 5 As shown, when the training set reaches the amount of data corresponding to approximately 5 biological sources, the model performance enters a stable plateau, and subsequent increases in training data do not significantly contribute to performance improvement.

[0047] Step 4: Confirm the performance of the final model Based on the analysis findings in step three, the model trained using all seven biological sources (a total of 140 sets of training data) represents the upper limit of stability achievable by the method of this invention. See also Figure 5 The model achieved excellent classification performance under strict leave-one-out cross-validation.

[0048] See Figure 6 The present invention also provides a skin tissue image recognition system based on polarization imaging, comprising: DoP and AoP image acquisition unit, used to acquire DoP and AoP images of unstained skin tissue sections; The first polarization feature parameter extraction unit is used to extract polarization feature parameters of the first region of interest in the DoP image; the first region of interest includes the region covering the spinous layer, granular layer and basal layer of the epidermis excluding the stratum corneum; the polarization feature parameters of the first region of interest include the peak value of the DoP histogram, the slope of the DoP histogram and the average DoP of the high polarization region. The second polarization feature parameter extraction unit is used to extract polarization feature parameters of the second region of interest in the AoP image; the second region of interest includes a first cropping unit and a second cropping unit; the first cropping unit is located in the spinous layer and is a cropping unit that can frame the complete cell texture; the second cropping unit is located in the basal layer and is a cropping unit that can frame the complete cell texture; the polarization feature parameters of the second region of interest include the AoP texture anisotropy index; The image recognition unit is used to perform skin tissue image recognition based on the polarization feature parameters of the first region of interest and the second region of interest, and to obtain the skin tissue image recognition result.

[0049] This system, through the high integration of DoP and AoP image acquisition units, a first polarization feature parameter extraction unit, a second polarization feature parameter extraction unit, and an image recognition unit, achieves the acquisition of DoP and AoP images of unstained skin tissue sections, the extraction of four polarization feature parameters, and the recognition of skin tissue images. The system has a simple structure, enabling comprehensive evaluation of skin tissue from different angles and levels. It is relatively easy to operate and maintain, and can reflect the basic microstructural characteristics of skin tissue. It is not only suitable for the diagnosis of common skin diseases (such as skin cancer and psoriasis), but can also be extended to the detection and differential diagnosis of other skin diseases by adjusting the feature parameter analysis model and recognition algorithm. It has strong adaptability and scalability, providing more comprehensive and in-depth information support for the diagnosis and treatment of skin diseases.

[0050] This invention provides a terminal device comprising: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in the various method embodiments described above. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0051] The computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention.

[0052] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0053] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0054] The memory can be used to store the computer program and / or module. The processor implements various functions of the terminal device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0055] If the modules / units integrated into the terminal device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the technical solution of the present invention in any way. Those skilled in the art should understand that, without departing from the spirit and principles of the present invention, the technical solution can be modified and replaced in several simple ways, and these modifications and replacements are all within the scope of protection covered by the claims.

Claims

1. A skin tissue image recognition method based on polarization imaging, characterized in that, include: Obtain DoP and AoP images of unstained skin tissue sections; Polarization feature parameters of the first region of interest are extracted from the DoP image; The first region of interest includes the region covering the spinous layer, granular layer and basal layer of the epidermis excluding the stratum corneum; the polarization characteristic parameters of the first region of interest include the peak value of the DoP histogram, the slope of the DoP histogram and the average DoP in the high polarization region; The polarization feature parameters of the second region of interest (ROI) are extracted from the AoP image. The second ROI includes a first cropping unit and a second cropping unit. The first cropping unit is located in the spinous layer and is capable of capturing the complete cell texture. The second cropping unit is located in the basal layer and is capable of capturing the complete cell texture. The polarization feature parameters of the second ROI include the AoP texture anisotropy index. Based on the polarization feature parameters of the first and second regions of interest, skin tissue image recognition is performed to obtain the skin tissue image recognition result.

2. The skin tissue image recognition method based on polarization imaging according to claim 1, characterized in that, The method for obtaining DoP and AoP images of unstained skin tissue sections is as follows: Obtain linearly polarized images of unstained skin tissue sections at 0°, 45°, 90°, and 135°. Calculate the Stokes parameters based on the linear polarization images at 0°, 45°, 90°, and 135°. Based on the Stokes parameters, DoP and AoP images of unstained skin tissue sections were obtained.

3. The skin tissue image recognition method based on polarization imaging according to claim 1, characterized in that, The method for obtaining the peak value of the DoP histogram is as follows: A statistical histogram is plotted for the polarization degree values ​​of all pixels in the first region of interest, and the probability density corresponding to the highest point of the plotted histogram is taken as the peak value of the DoP histogram.

4. The skin tissue image recognition method based on polarization imaging according to claim 1, characterized in that, The method for obtaining the slope of the DoP histogram is as follows: A statistical histogram of polarization values ​​for all pixels within the first region of interest is plotted, and the envelope of the histogram is fitted. Extract the peak points of the envelope of the histogram, and perform linear fitting on the descending segment from the peak points to obtain a straight line. Calculate the slope of this straight line to obtain the slope of the DoP histogram.

5. The skin tissue image recognition method based on polarization imaging according to claim 1, characterized in that, The method for obtaining the average DoP in the highly polarized region is as follows: In the first region of interest, all pixels are sorted in descending order of DoP value, starting from the maximum value and moving downwards, selecting the top 5%. The region containing each pixel is a highly polarized region. The total number of pixels in the first region of interest is given; the average value of all pixels in the high polarization region is calculated as the average DoP of the high polarization region.

6. The skin tissue image recognition method based on polarization imaging according to claim 1, characterized in that, The method for obtaining the AoP texture anisotropy index is as follows: Obtain the AoP images corresponding to the first and second cropping units; In the AoP images corresponding to the first and second cropping units, the occurrence frequency of gray value combinations of adjacent pixels with a geometric distance of 1 pixel in four spatial directions is counted and used as matrix elements to construct the gray-level co-occurrence matrix of the first and second cropping units in four spatial directions; the four spatial directions are 0°, 45°, 90° and 135° spatial directions, respectively. Texture feature parameters of the first and second cropping units are extracted from the gray-level co-occurrence matrices in four spatial directions, respectively; the texture feature parameters include the contrast, energy, entropy and homogeneity of adjacent pixels; Based on the texture feature parameters of the first and second truncation units, the coefficients of variation of the texture feature parameters of the first and second truncation units in four spatial directions are calculated accordingly. The variation coefficients of the texture feature parameters of the first and second cutting units in the four spatial directions are multiplied by their respective preset weighting coefficients and then summed to obtain the directional scores of the first and second cutting units, respectively. The directional scores of the first and second interception units are weighted and fused twice to obtain the AoP texture anisotropy index.

7. The skin tissue image recognition method based on polarization imaging according to claim 1, characterized in that, The method for performing skin tissue image recognition based on the polarization feature parameters of the first and second regions of interest to obtain the skin tissue image recognition result is as follows: The polarization feature parameters of the first and second regions of interest are input into a preset logistic regression classification model, and the skin tissue image recognition result is output.

8. A skin tissue image recognition system based on polarization imaging, characterized in that, include: DoP and AoP image acquisition unit, used to acquire DoP and AoP images of unstained skin tissue sections; The first polarization feature parameter extraction unit is used to extract polarization feature parameters of the first region of interest in the DoP image; the first region of interest includes the region covering the spinous layer, granular layer and basal layer of the epidermis excluding the stratum corneum; the polarization feature parameters of the first region of interest include the peak value of the DoP histogram, the slope of the DoP histogram and the average DoP of the high polarization region. The second polarization feature parameter extraction unit is used to extract polarization feature parameters of the second region of interest in the AoP image; the second region of interest includes a first cropping unit and a second cropping unit; the first cropping unit is located in the spinous layer and is a cropping unit that can frame the complete cell texture; the second cropping unit is located in the basal layer and is a cropping unit that can frame the complete cell texture; the polarization feature parameters of the second region of interest include the AoP texture anisotropy index; The image recognition unit is used to perform skin tissue image recognition based on the polarization feature parameters of the first region of interest and the second region of interest, and to obtain the skin tissue image recognition result.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images