A traditional Chinese medicine clinical research data intelligent analysis system and method and a storage medium

Abstract

This invention discloses an intelligent analysis system, method, and storage medium for TCM clinical research data, belonging to the field of medical information processing technology. It aims to solve the technical problems of deviations in tongue coating thickness determination and impaired data comparability caused by differences in image acquisition equipment and lighting environments among different sub-centers in multi-center TCM clinical research. The system includes: a hardware processing module for enhancing tongue coating-tongue body contrast, comprising an adaptive histogram equalization unit for adaptively enhancing the contrast of the input tongue image. The adaptive histogram equalization unit enhances the contrast between the tongue coating region and the tongue body region at the moment of image acquisition; and a tongue coating thickness grading feature extraction acceleration unit. This invention has the advantage of achieving real-time hardware-level extraction of tongue coating thickness features and automatic multi-center consistency verification at the data acquisition source, thereby ensuring the consistency and comparability of clinical research data.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and more specifically, to an intelligent analysis system, method, and storage medium for clinical research data in traditional Chinese medicine. Background Technology

[0002] In traditional Chinese medicine clinical research, tongue coating thickness is an important quantitative indicator for tongue diagnosis, playing a crucial role in syndrome differentiation and efficacy evaluation. In multi-center clinical research models, different sub-centers use different brands and models of image acquisition equipment, with varying optical lens parameters and photosensitive element characteristics. Furthermore, significant differences in color temperature and illuminance exist in the lighting environments of each sub-center's examination rooms, leading to systematic deviations in the contrast and brightness distribution of the same patient's tongue image across different sub-centers. The determination of tongue coating thickness is highly dependent on the contrast characteristics between the tongue coating area and the tongue body area, as well as the texture characteristics of the tongue coating. These inconsistencies in acquisition conditions directly cause deviations in tongue coating thickness determination results across multiple centers, severely affecting the comparability of clinical research data and the reliability of pooled analysis conclusions.

[0003] Existing tongue image processing methods have significant limitations in multi-center clinical research scenarios. Firstly, image acquisition and data analysis are typically separate processes. Tongue images need to be transmitted over a network to a centralized backend for processing. In high-concurrency multi-center scenarios, this places considerable pressure on transmission bandwidth and central computing power, making it difficult to meet real-time requirements. Secondly, inconsistencies in software versions or algorithm parameter configurations across different centers further exacerbate differences in judgment criteria. Furthermore, existing processing architectures lack mechanisms for real-time consistency verification and anomaly marking of tongue coating thickness grading results at the data acquisition source. Data quality issues are often only discovered during the later cleaning stage, increasing the uncertainty of research quality control and the cost of manual verification. In addition, existing solutions lack the efficient parallel extraction capability of underlying texture features from tongue images at the hardware level, making it difficult to achieve real-time calculation and fusion of multi-directional texture features using a unified microarchitecture at the acquisition front end, thus hindering the consistency assurance of multi-center data. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent analysis system, method, and storage medium for clinical research data in traditional Chinese medicine, in order to solve the technical problems of deviation in tongue coating thickness determination and impaired data comparability caused by differences in image acquisition equipment and lighting environments among various sub-centers in multi-center clinical research of traditional Chinese medicine.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an intelligent analysis system for clinical research data in traditional Chinese medicine, comprising: The hardware processing module for enhancing the contrast between tongue coating and tongue body includes an adaptive histogram equalization unit, which is used to perform adaptive contrast enhancement processing on the input tongue image. The adaptive histogram equalization unit enhances the contrast between the tongue coating area and the tongue body area at the moment of image acquisition. The tongue coating thickness grading feature extraction acceleration unit includes a texture feature extraction acceleration module, which is connected to the tongue coating-tongue body contrast enhancement hardware processing module. It is used to extract tongue coating thickness grading features from the enhanced tongue image in real time. The texture feature extraction acceleration module calculates the texture feature parameters of the tongue coating area and maps the texture feature parameters to the tongue coating thickness grading result output. The multi-center data consistency verification module includes a feature comparison unit connected to a tongue coating thickness grading feature extraction acceleration unit. The feature comparison unit performs real-time consistency verification on the thickness grading results output by the tongue coating thickness grading feature extraction acceleration unit after processing the tongue images collected from different sub-centers. Thickness grading results that exceed the preset deviation range are marked. The tongue coating-tongue body contrast enhancement hardware processing module, the tongue coating thickness grading feature extraction acceleration unit, and the multi-center data consistency verification module are all integrated into the same intelligent analysis chip.

[0006] Preferably, the adaptive histogram equalization unit includes a block histogram statistics subunit, a cumulative distribution function calculation subunit, and a pixel mapping subunit; The block histogram statistics subunit is used to perform block division processing on the tongue image and to statistically analyze the pixel gray values ​​in each sub-block to generate a gray distribution histogram corresponding to each sub-block. The cumulative distribution function calculation subunit is connected to the block histogram statistics subunit, and is used to receive the gray-level distribution histogram of each sub-block and calculate the cumulative distribution function corresponding to each sub-block based on the gray-level distribution histogram; The pixel mapping subunit is connected to the cumulative distribution function calculation subunit and is used to perform grayscale remapping operation on the pixel grayscale values ​​in each sub-block according to the calculated cumulative distribution function. For any sub-block, let the gray level within that sub-block be... The probability of its occurrence is Then the cumulative distribution function corresponding to that sub-block The calculation formula is expressed as: ; The pixel mapping subunit is based on the cumulative distribution function. The original grayscale values ​​within each sub-block Perform a grayscale remapping operation; the remapped grayscale value The calculation formula is expressed as: ; in, This represents the total number of gray levels. For the grayscale index variable in the summation operation, grayscale Normalized occurrence probability within a sub-block grayscale The corresponding cumulative distribution function value, This represents the original grayscale value of a pixel within a sub-block of the original tongue image. This indicates that the original grayscale value will be... The function value obtained after substituting into the cumulative distribution function, This is the new grayscale value of the pixel after the grayscale remapping operation. This represents the maximum value of the grayscale level.

[0007] Preferably, the adaptive histogram equalization unit further includes a contrast limiting subunit, which is connected to the block histogram statistics subunit and the cumulative distribution function calculation subunit. The contrast limiting subunit is used to perform a clipping operation on the gray-level distribution histogram of each sub-block and redistribute the number of pixels exceeding the preset clipping threshold to other gray-level intervals of the gray-level distribution histogram. Let the preset shear threshold be gray levels in the original histogram The number of pixels is The histogram after clipping The calculation formula is expressed as: ; To evenly redistribute the total number of pixels exceeding the clipping threshold to the incremental values ​​corresponding to each gray level, the calculation formula is as follows: ; in, gray levels in the original gray-level distribution histogram The corresponding number of pixels, The gray levels in the grayscale distribution histogram after cropping are shown. The corresponding number of pixels; For all gray levels that exceed the clipping threshold, calculate the sum of the number of pixels exceeding the threshold.

[0008] Preferably, the texture feature extraction acceleration module includes a gray-level co-occurrence matrix hardware computing array, which is used to perform parallel computing operations on the gray-level co-occurrence matrix feature parameters of the tongue coating region in multiple preset directions within a single hardware clock cycle. The feature parameters include contrast parameters, energy parameters, and entropy parameters. For a given image of a tongue coating region, let its gray-level co-occurrence matrix be... ,in and These represent the grayscale index variables of the two pixels. The spatial distance between pixel pairs The directional angle between pixel pairs; Calculate contrast parameters based on the gray-level co-occurrence matrix. Energy parameters and entropy parameter The calculation formulas are expressed as follows: ; ; ; in, To be at a given spatial distance and direction angle Under the condition of gray level With gray level Simultaneous normalized probability values This is the natural logarithm operator.

[0009] Preferably, the texture feature extraction acceleration module further includes a feature fusion and hierarchical decision logic unit, which is connected to the gray-level co-occurrence matrix hardware computing array. The feature fusion and hierarchical decision logic unit is used to perform weighted fusion processing on the feature parameters in multiple directions output by the gray-level co-occurrence matrix hardware computing array to generate comprehensive texture feature parameters, and output the tongue coating thickness grading result according to the comprehensive texture feature parameters. The tongue coating thickness grading result includes three grading categories: thin coating grading, slightly thick coating grading, and thick coating grading. For the There are several preset directions, and their corresponding contrast parameters are set as follows: Energy parameters are The entropy parameter is First, the feature parameters are normalized to obtain the normalized feature parameters. , , The normalization calculation formulas are expressed as follows: ; ; ; Then the directional comprehensive eigenvalue in this direction The calculation formula is expressed as: ; All Directional comprehensive feature value in each preset direction Weighted fusion is performed to obtain comprehensive texture feature parameters. The calculation formula is expressed as: ; The feature fusion and hierarchical decision-making logic unit will integrate the texture feature parameters. The results are compared with preset grading threshold ranges, and the corresponding tongue coating thickness grading results are mapped and output based on the comparison results. in, For all The maximum value among the original values ​​of the contrast parameter in each preset direction For all The maximum value among the original values ​​of the entropy parameter in each preset direction. , , These are the preset weighting coefficients corresponding to the normalized contrast parameter, energy parameter, and normalized entropy parameter, respectively. For the first Directional comprehensive feature values ​​in a preset direction, For the first The direction weight coefficients corresponding to each preset direction.

[0010] Preferably, the feature comparison unit includes a reference feature value storage array and a comparison logic module. The reference feature value storage array is used to store a pre-set reference feature vector for tongue coating thickness grading. The comparison logic module is connected to the reference feature value storage array and is used to perform a parallel comparison operation between the thickness grading features obtained after the tongue image collected by the current sub-center is processed by the tongue coating thickness grading feature extraction acceleration unit and the reference feature vector, and output a consistency verification conclusion based on the comparison result.

[0011] Preferably, the multi-center data consistency verification module further includes a deviation marker register, which is connected to the comparison logic module and is used to add abnormal marker information to the thickness grading result when the comparison logic module detects that the deviation value between the thickness grading feature of the current sub-center and the reference feature vector exceeds a preset deviation range.

[0012] Preferably, the intelligent analysis chip further includes a multi-center data synchronization interface, which is connected to a multi-center data consistency verification module. This interface is used to receive tongue image data from different clinical research sub-centers and to package the thickness grading results after verification by the multi-center data consistency verification module with the sub-center identification information corresponding to each sub-center before outputting them to the outside.

[0013] A method for intelligent analysis of clinical research data in traditional Chinese medicine includes the following steps: S1. Through the tongue coating-tongue body contrast enhancement hardware processing module, the input multicenter clinical study tongue image is subjected to contrast adaptive enhancement processing, and the contrast between the tongue coating area and the tongue body area is enhanced at the moment of image acquisition. S2. Through the tongue coating thickness grading feature extraction acceleration unit, the tongue coating thickness grading features are extracted in real time from the enhanced tongue image. Based on the texture feature extraction acceleration module, the texture feature parameters of the tongue coating area are calculated, and the texture feature parameters are mapped to the tongue coating thickness grading results and then output. S3. Through the multi-center data consistency verification module, real-time consistency verification processing is performed on the thickness grading results corresponding to the tongue images collected from different sub-centers, and anomaly marking is performed on the thickness grading results that exceed the preset deviation range.

[0014] A computer-readable storage medium storing a computer program, wherein when the computer program is loaded into an intelligent analysis chip in a traditional Chinese medicine clinical research data intelligent analysis system, the intelligent analysis chip is configured to execute a traditional Chinese medicine clinical research data intelligent analysis method.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention integrates a hardware processing module for enhancing tongue coating-tongue body contrast and an adaptive histogram equalization unit into an intelligent analysis chip, performing hardware-level enhancement of the contrast between the tongue coating and tongue body regions at the moment of image acquisition. This module embeds the adaptive histogram equalization algorithm into the chip's internal logic circuitry, achieving local adaptive contrast enhancement through a hardware pipeline architecture of block histogram statistics, cumulative distribution function calculation, and pixel grayscale remapping. Because the processing is completed at the chip hardware level, it is unaffected by external processor load and software environment differences, effectively eliminating deviations in tongue coating thickness determination caused by differences in equipment models and lighting conditions between different centers, ensuring the consistency of tongue image data in multi-center clinical studies from the data source.

[0016] 2. This invention also integrates a tongue coating thickness grading feature extraction acceleration unit into the intelligent analysis chip. This unit has a built-in hardware-embedded texture feature extraction acceleration module that performs real-time calculations on the texture features of the tongue coating region at the chip hardware layer. The module's internal gray-level co-occurrence matrix hardware calculation array synchronously extracts contrast, energy, and entropy parameters through a multi-directional parallel computing unit. After normalization and weighted fusion by the feature fusion and grading decision logic unit, a standardized tongue coating thickness grading result is output. The complete processing chain is completed in a closed loop within the chip, eliminating the need to transmit image data to an external processor. This reduces transmission bandwidth requirements and processing latency. Furthermore, the unified algorithm parameters embedded in the chip hardware layer ensure that each sub-center uses a consistent standard to perform thickness grading, fundamentally solving the problem of inconsistent judgment standards caused by version or parameter differences in traditional software solutions.

[0017] 3. This invention also integrates a multi-center data consistency verification module into the intelligent analysis chip. This module includes a feature comparison unit and a deviation marker register, enabling real-time quality verification and anomaly marking of tongue coating thickness grading results at each sub-center. The feature comparison unit has a built-in reference feature value storage array, storing reference feature vectors calibrated by the central laboratory based on standard tongue image samples. After the sub-center completes the thickness grading, the comparison logic module instantly calculates the deviation between the measured feature vector and the reference feature vector, and the deviation marker register automatically determines and outputs anomaly markers based on a preset threshold. This hardware-level verification mechanism performs quality screening at the data source. Anomaly-marked thickness grading results can be automatically identified and given priority attention in subsequent analyses, avoiding distortion of research conclusions due to data anomalies from individual sub-centers and significantly reducing the manual cost of subsequent data verification. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system framework of the present invention; Figure 2 This is a schematic diagram of the hardware processing module for enhancing the contrast between tongue coating and tongue body in this invention. Figure 3 This is a schematic diagram of the framework of the tongue coating thickness grading feature extraction acceleration unit of the present invention; Figure 4 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0019] To facilitate understanding of the technical solution of the present invention by those skilled in the art, the technical solution of the present invention will now be further described in conjunction with the accompanying drawings.

[0020] Example 1, such as Figures 1-3 As shown, this invention provides an intelligent analysis system for clinical research data in traditional Chinese medicine, comprising: The hardware processing module for enhancing the contrast between tongue coating and tongue body includes an adaptive histogram equalization unit, which is used to perform adaptive contrast enhancement processing on the input tongue image. The adaptive histogram equalization unit enhances the contrast between the tongue coating area and the tongue body area at the moment of image acquisition, so as to eliminate the deviation in tongue coating thickness determination caused by the difference in optical characteristics between different acquisition devices. The adaptive histogram equalization unit includes a block histogram statistics subunit, a cumulative distribution function calculation subunit, and a pixel mapping subunit; The block histogram statistics subunit is used to perform block division processing on the tongue image and to statistically analyze the pixel gray values ​​in each sub-block to generate a gray distribution histogram corresponding to each sub-block. The cumulative distribution function calculation subunit is connected to the block histogram statistics subunit, which is used to receive the gray-level distribution histogram of each sub-block and calculate the cumulative distribution function corresponding to each sub-block based on the gray-level distribution histogram; The pixel mapping subunit is connected to the cumulative distribution function calculation subunit, and is used to perform grayscale remapping operation on the pixel grayscale values ​​in each sub-block according to the calculated cumulative distribution function, thereby achieving adaptive contrast enhancement processing in the local area. For any sub-block, let the gray level within that sub-block be... The probability of its occurrence is Then the cumulative distribution function corresponding to that sub-block The calculation formula is expressed as: ; Pixel mapping subunits based on cumulative distribution function The original grayscale values ​​within each sub-block Perform a grayscale remapping operation; the remapped grayscale value The calculation formula is expressed as: ; in, This represents a grayscale variable, and its value range is... ; This represents the total number of gray levels. The grayscale index variable in the summation operation is accumulated from zero to the current grayscale level. ; grayscale The normalized probability of occurrence within a sub-block is the ratio of the number of pixels at that gray level to the total number of pixels within the sub-block. grayscale The corresponding cumulative distribution function value is defined as the gray level from The sum of cumulative probabilities, whose range is: ; This represents the original grayscale value of a pixel within a sub-block of the original tongue image; This indicates that the original grayscale value will be... The function value obtained after substituting into the cumulative distribution function; This is the new grayscale value of the pixel after the grayscale remapping operation; This represents the maximum value of the grayscale level. The block histogram statistical subunit statistically analyzes the pixel grayscale distribution within each sub-block of the tongue image to obtain the probability of occurrence of each grayscale level. Subsequently, the cumulative distribution function calculation sub-units, arranged in ascending order of gray levels, progressively sum the probabilities of occurrence for each gray level to obtain the cumulative distribution function. The cumulative distribution function exhibits a monotonically increasing property, and its value reflects the cumulative proportion of a certain gray level within a sub-block. Finally, the pixel mapping sub-unit maps the original gray values... Substitute the cumulative distribution function and multiply the resulting function value by the maximum gray level. This maps the original grayscale value to a new grayscale value. Through the above mapping operation, regions with concentrated grayscale distribution in the original image are stretched to a wider grayscale range, while regions with sparse grayscale distribution are compressed accordingly, thereby achieving adaptive contrast enhancement within a local area. The adaptive histogram equalization unit also includes a contrast limiting subunit, which is connected to the block histogram statistics subunit and the cumulative distribution function calculation subunit. It is used to perform a clipping operation on the gray-level distribution histogram of each sub-block and redistribute the number of pixels that exceed the preset clipping threshold to other gray-level ranges of the gray-level distribution histogram to suppress the excessive amplification of image noise that may be caused by contrast enhancement processing. Let the preset shear threshold be gray levels in the original histogram The number of pixels is The histogram after clipping The calculation formula is expressed as: ; To evenly redistribute the total number of pixels exceeding the clipping threshold to the incremental values ​​corresponding to each gray level, the calculation formula is as follows: ; in, gray levels in the original gray-level distribution histogram The corresponding number of pixels, i.e., the original statistical value without cropping; This is a preset clipping threshold used to limit the maximum number of pixels at any gray level; The gray levels in the grayscale distribution histogram after cropping are shown. The corresponding number of pixels; For all gray levels that exceed the clipping threshold, calculate the sum of the number of pixels exceeding the threshold.

[0021] This represents the total number of gray levels. The pixel increment value obtained for each gray level after the total number of pixels exceeding the clipping threshold is evenly distributed across all gray levels; In one specific embodiment of the present invention, the contrast limiting subunit includes: Threshold comparator array, including There are 12 parallel comparators, each corresponding to a gray level, used to count the number of pixels at each gray level. With preset shear threshold Compare the results and output the comparison result flag. The shearing execution module, connected to the threshold comparator array, includes... Several parallel multiplexers, each multiplexer uses a flag bit based on the comparison result of the corresponding gray level, when... Time output ,when Time output This yields the cut intermediate histogram values; Beyond the pixel accumulator, connected to a threshold comparator array, it is used to compare all pixels that meet the threshold comparator array. The grayscale level, the parallel computation exceeds the limit. The sum of all additions; The divider module, connected to the pixel accumulator, is used to divide the accumulated sum by the total number of gray levels. , obtain pixel increment value ; Adder array, including A parallel adder is used to combine the clipped intermediate histogram value with the pixel increment value. Add them together and output the final histogram after clipping. ; The contrast limiting subunit iterates through the number of pixels corresponding to each gray level. Those exceeding the preset shear threshold Part of it is removed, temporarily limiting the number of pixels at that gray level to [a certain value]. Then, the number of all cropped pixels is summed to obtain the total number of pixels exceeding the cutoff portion. Next, this total number of pixels is divided by the total number of gray levels. Calculate the pixel increment value evenly distributed to each gray level. Finally, for grayscale levels where the number of pixels does not exceed the clipping threshold, the original number of pixels is multiplied by the increment value. Add them together to get the number of new pixels after cropping. Through the above processing, the overly concentrated gray-level distribution in the original histogram is moderately dispersed, and the gray-level distribution pattern becomes more balanced; The tongue coating-tongue body contrast enhancement hardware processing module also includes an input image caching unit, which is connected to an adaptive histogram equalization unit. The input image caching unit is used to perform caching operations on the tongue image data stream from the image acquisition device and to provide the tongue image data to be processed to the adaptive histogram equalization unit in a continuous data input manner.

[0022] The tongue coating thickness grading feature extraction acceleration unit includes a texture feature extraction acceleration module, which is connected to the tongue coating-tongue body contrast enhancement hardware processing module. It is used to extract tongue coating thickness grading features from the enhanced tongue image in real time. The texture feature extraction acceleration module calculates the texture feature parameters of the tongue coating area and maps the texture feature parameters to the tongue coating thickness grading result output. The texture feature extraction acceleration module includes a gray-level co-occurrence matrix hardware computing array, which is used to perform parallel computing operations on the gray-level co-occurrence matrix feature parameters of the tongue coating region in multiple preset directions within a single hardware clock cycle. The feature parameters include contrast parameters, energy parameters, and entropy parameters. For a given image of a tongue coating region, let its gray-level co-occurrence matrix be... ,in and These represent the grayscale index variables of the two pixels. The spatial distance between pixel pairs The directional angle between pixel pairs; Calculating contrast parameters based on the gray-level co-occurrence matrix Energy parameters and entropy parameter The calculation formulas are expressed as follows: ; ; ; in, To be at a given spatial distance and direction angle Under the condition of gray level With gray level The normalized probability values ​​that appear at the same time are the elements of the normalized gray-level co-occurrence matrix; The contrast parameter reflects the clarity of the tongue coating image texture and the depth of the texture grooves; It is an energy parameter, also known as the second moment of the angle, which reflects the uniformity of gray-level distribution and the fineness of texture in the tongue coating image; The entropy parameter reflects the complexity and non-uniformity of the tongue coating texture. The natural logarithm operator; A grayscale co-occurrence matrix hardware computing array scans an image of the tongue coating region and counts data within a preset spatial distance. and direction angle Under the condition, gray level With gray level The frequency of adjacent occurrences, after normalization, forms a gray-level co-occurrence matrix. Subsequently, based on this matrix, three texture feature parameters are calculated: contrast parameter... The contrast parameter is obtained by weighted summation of the squares of the grayscale differences across the matrix elements. A larger grayscale difference indicates a higher contrast parameter value, signifying clearer tongue coating texture; energy parameter... The energy parameter value is obtained by summing the squares of the matrix elements. A more concentrated distribution of matrix elements indicates a higher energy parameter value, suggesting a more uniform and regular tongue coating texture; the entropy parameter... The entropy parameter is obtained by summing the products of matrix elements and their natural logarithms and taking the negative value. The more dispersed the matrix elements are, the higher the entropy parameter value, indicating that the tongue coating texture is more complex and disordered. In one specific embodiment of the present invention, the gray-level co-occurrence matrix hardware computing array includes multiple parallel gray-level co-occurrence matrix computing units, each computing unit corresponding to a preset direction angle. Each computing unit internally includes: Pixel-to-pixel sampling module, used to sample according to a preset spatial distance and direction angle Read pixel pairs sequentially from the image of the tongue coating area ; The frequency statistics module, connected to the pixel pair sampling module, includes a two-dimensional counter array, where the row index of the two-dimensional counter array corresponds to the gray level. The column index corresponds to the gray level. For each pair of pixels read, the count value at the corresponding position is incremented by one; The normalization module, connected to the frequency statistics module, is used to divide the frequency statistics values ​​in the two-dimensional counter array by the total number of pixel pairs to obtain the normalized gray-level co-occurrence matrix element values. ; The feature calculation module, connected to the normalization module, includes multiple parallel multiply-accumulate operators, each calculating according to the contrast parameter. Energy parameters and entropy parameter The calculation formula performs multiplication and addition operations on the elements of the normalized gray-level co-occurrence matrix; The multiple preset directions include four directions: 0°, 45°, 90° and 135°. The gray-level co-occurrence matrix hardware computing array contains four gray-level co-occurrence matrix computing units, which correspond to the above four directions respectively. The four computing units output the contrast parameters, energy parameters and entropy parameters of their respective directions in parallel within the same clock cycle. The texture feature extraction acceleration module also includes a feature fusion and hierarchical decision logic unit. The feature fusion and hierarchical decision logic unit is connected to the gray-level co-occurrence matrix hardware computing array. It is used to perform weighted fusion processing on the feature parameters in multiple directions output by the gray-level co-occurrence matrix hardware computing array to generate comprehensive texture feature parameters, and output the tongue coating thickness grading result according to the comprehensive texture feature parameters. The tongue coating thickness grading result includes three grading categories: thin coating grading, slightly thick coating grading, and thick coating grading. For the There are several preset directions, and their corresponding contrast parameters are set as follows: Energy parameters are The entropy parameter is First, the feature parameters are normalized to obtain the normalized feature parameters. , , The normalization calculation formulas are expressed as follows: ; ; ; Then the directional comprehensive eigenvalue in this direction The calculation formula is expressed as: ; All Directional comprehensive feature value in each preset direction Weighted fusion is performed to obtain comprehensive texture feature parameters. The calculation formula is expressed as: ; The feature fusion and hierarchical decision logic unit integrates texture feature parameters. The results are compared with preset grading threshold ranges, and the corresponding tongue coating thickness grading results are mapped and output based on the comparison results. in, For all The maximum value among the original values ​​of the contrast parameter in each preset direction; For all The maximum value among the original values ​​of the entropy parameter in each preset direction; , , These are the preset weighting coefficients corresponding to the normalized contrast parameter, energy parameter, and normalized entropy parameter, respectively, and the sum of the three is equal to one. For the first A directional comprehensive feature value in a preset direction comprehensively reflects the overall characteristics of the tongue coating texture in that direction; For the first The direction weight coefficients corresponding to each preset direction; To eliminate differences in the dimensions and numerical ranges of different feature parameters, the contrast and entropy parameters are normalized by dividing them by their respective maximum values ​​in all directions. The energy parameter, since its value range is between zero and one, does not require additional normalization. Secondly, within each preset direction, the three normalized feature parameters are weighted according to preset coefficients. , , Perform a linear weighted summation to obtain the directional comprehensive eigenvalue for that direction. The magnitude of the weighting coefficients reflects the relative importance of different texture features in determining tongue coating thickness. Finally, all... The directional comprehensive characteristic values ​​of each direction are weighted according to their respective directional weighting coefficients. Perform weighted summation to obtain comprehensive texture feature parameters. The feature fusion and hierarchical decision-making logic unit will... The values ​​are compared with the pre-defined grading threshold range, based on... The threshold interval is mapped to the corresponding tongue coating thickness grading result; In a specific embodiment of the present invention, the preset hierarchical threshold range is pre-defined in the following manner: A sample set of tongue images was collected, with the results of the tongue coating thickness grading labeled by TCM experts. The sample set included several samples of thin coating, slightly thick coating, and thick coating. Each tongue image in the sample set is sequentially input into the texture feature extraction acceleration module to obtain the comprehensive texture feature parameters corresponding to each sample. ; Statistical analysis of the samples of thin moss The distribution range of the values ​​was used, with the upper and lower quartiles of this range as the threshold intervals corresponding to the thin moss grading; the values ​​of the thick moss samples were statistically analyzed. The distribution range of the values ​​is used as the upper and lower quartiles of the distribution range as the threshold intervals corresponding to the thick moss classification; the range between the upper limit of the thin moss threshold interval and the lower limit of the thick moss threshold interval is used as the threshold interval corresponding to the slightly thick moss classification; the calibrated classification threshold intervals are stored in the threshold register of the feature fusion and classification decision logic unit.

[0023] The multi-center data consistency verification module includes a feature comparison unit connected to the tongue coating thickness grading feature extraction acceleration unit. The feature comparison unit performs real-time consistency verification on the thickness grading results output by the tongue coating thickness grading feature extraction acceleration unit after processing the tongue images collected from different sub-centers. Thickness grading results that exceed the preset deviation range are marked to ensure the comparability of multi-center clinical research data. The feature comparison unit includes a reference feature value storage array and a comparison logic module. The reference feature value storage array is used to store a pre-set reference feature vector for tongue coating thickness grading. The comparison logic module is connected to the reference feature value storage array and is used to perform a parallel comparison operation between the thickness grading features obtained after the tongue image collected by the current sub-center is processed by the tongue coating thickness grading feature extraction acceleration unit and the reference feature vector, and output a consistency verification conclusion based on the comparison result. Let the reference eigenvector be ,in The total dimension of the feature vector. For the first Reference feature values ​​in each dimension; Let the thickness grading feature vector of the current subcenter be... ,in For the first Measured feature values ​​in each dimension; The characteristic deviation calculated by the comparison logic module The calculation formula is expressed as: ; in, The dimension index variable of the feature vector, with a value range of 1 to... ; Reference feature vector The Middle Reference feature values ​​for each dimension, which are pre-calibrated by the central laboratory based on standard tongue image samples and stored in a reference feature value storage array; The measured feature vectors obtained by processing tongue images collected at the current branch center after being processed by the tongue coating thickness grading feature extraction acceleration unit. The Middle Measured feature values ​​in each dimension; For the first The absolute deviation between the measured feature value and the reference feature value in each dimension; The characteristic deviation is the value for all. The sum of the absolute deviation values ​​of each dimension comprehensively reflects the overall degree of deviation between the current measured data of the sub-center and the reference standard. The comparison logic module first reads the pre-stored reference feature vector from the reference feature value storage array. Simultaneously, it receives the measured feature vector output by the tongue coating thickness grading feature extraction acceleration unit. Subsequently, the comparison logic module calculates the measured feature values ​​dimension by dimension. With reference eigenvalues The absolute value of the difference between them, and all The feature bias is obtained by summing the absolute deviation values ​​of each dimension. Characteristic deviation The numerical value directly reflects the degree of deviation between the current tongue image data of the sub-center and the standard reference data in the feature space. The smaller the value, the closer the data of the current branch center is to the reference standard. The larger the value, the more significant the deviation; The multi-center data consistency verification module also includes a deviation flag register, which is connected to the comparison logic module. When the comparison logic module detects that the deviation value between the thickness grading feature of the current sub-center and the reference feature vector exceeds the preset deviation range, it adds abnormal flag information to the thickness grading result. Let the preset deviation threshold be Then the deviation flag register will output the flag status. The calculation formula is expressed as: ; in, This is the characteristic deviation. The preset deviation threshold is set in advance by the clinical research protocol according to the research accuracy requirements and stored in the deviation flag register; The offset flag register is used to set the output status of the offset flag register, and its value is a Boolean logic value of one or zero. This indicates that the thickness grading results of the current sub-center exceed the preset deviation range, and anomaly marker information needs to be added. This indicates that the thickness grading results of the current sub-center are within the preset deviation range, and no additional anomaly marker information is required. The feature deviation is calculated in the comparison logic module. After the calculation, the deviation flag register will The current value and the preset deviation threshold Perform numerical comparisons. When the characteristic deviation... Greater than the preset deviation threshold When this occurs, it indicates that the deviation between the tongue image data collected by the current sub-center and the standard reference data has exceeded the range allowed by the clinical research protocol, and the deviation flag register outputs... A value of one is used to add anomaly marker information to the thickness grading result. When the feature deviation... Less than or equal to the preset deviation threshold When the deviation flag register outputs, it indicates that the current data deviation of the sub-center is within an acceptable range. A value of zero indicates that the thickness grading result is output in a normal state.

[0024] The hardware processing module for enhancing the contrast between tongue coating and tongue body, the acceleration unit for extracting the grading features of tongue coating thickness, and the multi-center data consistency verification module are all integrated into the same intelligent analysis chip; In one specific embodiment of the present invention, the intelligent analysis chip is implemented using a field-programmable gate array (FPGA). The hardware processing module for enhancing tongue coating-tongue body contrast, the acceleration unit for extracting tongue coating thickness grading features, and the multi-center data consistency verification module are all written in hardware description language and synthesized into FPGA logic circuits.

[0025] In another specific embodiment, the intelligent analysis chip is implemented using an application-specific integrated circuit (ASIC), and each functional module is integrated on a single silicon substrate in the form of a solidified logic gate circuit.

[0026] In another specific embodiment, the intelligent analysis chip is implemented using an embedded artificial intelligence processor based on a domestically produced NPU architecture. The texture feature extraction acceleration module calls the convolution acceleration unit of the NPU to complete the parallel calculation of the gray-level co-occurrence matrix.

[0027] The intelligent analysis chip also includes a multi-center data synchronization interface, which is connected to a multi-center data consistency verification module. This interface is used to receive tongue image data from different clinical research sub-centers and to package the thickness grading results after verification by the multi-center data consistency verification module with the sub-center identification information corresponding to each sub-center before outputting them to the outside. The intelligent analysis chip also includes a tongue image region segmentation preprocessing module, which is connected to the input image cache unit and is used to separate the tongue coating region and the tongue body region from the original tongue image before contrast enhancement processing.

[0028] In one specific embodiment of the present invention, the tongue image region segmentation preprocessing module employs a threshold segmentation algorithm based on the HSV color space. This algorithm first converts the input RGB tongue image to the HSV color space, extracting the hue and saturation channels. Then, it uses preset hue and saturation threshold ranges to perform preliminary segmentation of the tongue region, separating the tongue from the facial background. Furthermore, within the tongue region, based on the differences in hue and saturation distribution between the tongue coating and the tongue body, an adaptive thresholding method is used to determine the segmentation boundary between the tongue coating region and the tongue body region. After segmentation, a tongue coating region mask and a tongue body region mask are generated.

[0029] In another specific embodiment, the tongue image region segmentation preprocessing module is implemented using a lightweight convolutional neural network model. This model is pre-trained on a tongue image sample dataset and directly outputs the segmentation results of the tongue coating region and the tongue body region through hardware inference within the chip.

[0030] The segmented tongue coating region mask and tongue body region mask are respectively fed into the adaptive histogram equalization unit and the texture feature extraction acceleration module, serving as the input limiting regions for subsequent contrast enhancement processing and texture feature extraction. The hardware circuit of the tongue image region segmentation preprocessing module is integrated with the tongue coating-tongue body contrast enhancement hardware processing module, the tongue coating thickness grading feature extraction acceleration unit, and the multi-center data consistency verification module in the same intelligent analysis chip.

[0031] Example 2, as Figure 4 As shown, this invention provides an intelligent analysis method for clinical research data in traditional Chinese medicine, comprising the following steps: S1. Through the tongue coating-tongue body contrast enhancement hardware processing module, the input multicenter clinical study tongue image is subjected to contrast adaptive enhancement processing, and the contrast between the tongue coating area and the tongue body area is enhanced at the moment of image acquisition. S2. Through the tongue coating thickness grading feature extraction acceleration unit, the tongue coating thickness grading features are extracted in real time from the enhanced tongue image. Based on the texture feature extraction acceleration module, the texture feature parameters of the tongue coating area are calculated, and the texture feature parameters are mapped to the tongue coating thickness grading results and then output. S3. Through the multi-center data consistency verification module, real-time consistency verification processing is performed on the thickness grading results corresponding to the tongue images collected from different sub-centers, and anomaly marking is performed on the thickness grading results that exceed the preset deviation range.

[0032] Example 3: The present invention provides a computer-readable storage medium storing a computer program thereon, characterized in that when the computer program is loaded into the intelligent analysis chip of the intelligent analysis system for clinical research data of traditional Chinese medicine, the intelligent analysis chip is configured to execute the intelligent analysis method for clinical research data of traditional Chinese medicine.

[0033] The embodiments disclosed in this invention are preferred embodiments, but are not limited thereto. Those skilled in the art can easily understand the spirit of this invention based on the above embodiments and make different extensions and variations, but as long as they do not depart from the spirit of this invention, they are all within the protection scope of this invention.

Claims

1. A smart data analysis system for clinical research in traditional Chinese medicine, characterized in that, include: The hardware processing module for enhancing the contrast between tongue coating and tongue body includes an adaptive histogram equalization unit, which is used to perform adaptive contrast enhancement processing on the input tongue image. The adaptive histogram equalization unit enhances the contrast between the tongue coating area and the tongue body area at the moment of image acquisition. The tongue coating thickness grading feature extraction acceleration unit includes a texture feature extraction acceleration module, which is connected to the tongue coating-tongue body contrast enhancement hardware processing module. It is used to extract tongue coating thickness grading features from the enhanced tongue image in real time. The texture feature extraction acceleration module calculates the texture feature parameters of the tongue coating area and maps the texture feature parameters to the tongue coating thickness grading result output. The multi-center data consistency verification module includes a feature comparison unit connected to a tongue coating thickness grading feature extraction acceleration unit. The feature comparison unit performs real-time consistency verification on the thickness grading results output by the tongue coating thickness grading feature extraction acceleration unit after processing the tongue images collected from different sub-centers. Thickness grading results that exceed the preset deviation range are marked. The tongue coating-tongue body contrast enhancement hardware processing module, the tongue coating thickness grading feature extraction acceleration unit, and the multi-center data consistency verification module are all integrated into the same intelligent analysis chip.

2. The intelligent analysis system for clinical research data in traditional Chinese medicine according to claim 1, characterized in that, The adaptive histogram equalization unit includes a block histogram statistics subunit, a cumulative distribution function calculation subunit, and a pixel mapping subunit; The block histogram statistics subunit is used to perform block division processing on the tongue image and to statistically analyze the pixel gray values ​​in each sub-block to generate a gray distribution histogram corresponding to each sub-block. The cumulative distribution function calculation subunit is connected to the block histogram statistics subunit, and is used to receive the gray-level distribution histogram of each sub-block and calculate the cumulative distribution function corresponding to each sub-block based on the gray-level distribution histogram; The pixel mapping subunit is connected to the cumulative distribution function calculation subunit and is used to perform grayscale remapping operation on the pixel grayscale values ​​in each sub-block according to the calculated cumulative distribution function. For any sub-block, let the gray level within that sub-block be... The probability of its occurrence is Then the cumulative distribution function corresponding to that sub-block The calculation formula is expressed as: ; The pixel mapping subunit is based on the cumulative distribution function. The original grayscale values ​​within each sub-block Perform a grayscale remapping operation; the remapped grayscale value The calculation formula is expressed as: ; in, This represents the total number of gray levels. For the grayscale index variable in the summation operation, grayscale Normalized occurrence probability within a sub-block grayscale The corresponding cumulative distribution function value, This represents the original grayscale value of a pixel within a sub-block of the original tongue image. This indicates that the original grayscale value will be... The function value obtained after substituting into the cumulative distribution function, This is the new grayscale value of the pixel after the grayscale remapping operation. This represents the maximum value of the grayscale level.

3. The intelligent analysis system for clinical research data in traditional Chinese medicine according to claim 2, characterized in that, The adaptive histogram equalization unit further includes a contrast limiting subunit, which is connected to the block histogram statistics subunit and the cumulative distribution function calculation subunit. It is used to perform a clipping operation on the gray-level distribution histogram of each sub-block and redistribute the number of pixels that exceed the preset clipping threshold to other gray-level intervals of the gray-level distribution histogram. Let the preset shear threshold be gray levels in the original histogram The number of pixels is The histogram after clipping The calculation formula is expressed as: ； To evenly redistribute the total number of pixels exceeding the clipping threshold to the incremental values ​​corresponding to each gray level, the calculation formula is as follows: ； in, gray levels in the original gray-level distribution histogram The corresponding number of pixels, The gray levels in the grayscale distribution histogram after cropping are shown. The corresponding number of pixels; For all gray levels that exceed the clipping threshold, calculate the sum of the number of pixels exceeding the threshold.

4. The intelligent analysis system for clinical research data in traditional Chinese medicine according to claim 1, characterized in that, The texture feature extraction acceleration module includes a gray-level co-occurrence matrix hardware computing array, which is used to perform parallel computing operations on the gray-level co-occurrence matrix feature parameters of the tongue coating region in multiple preset directions within a single hardware clock cycle. The feature parameters include contrast parameters, energy parameters, and entropy parameters. For a given image of a tongue coating region, let its gray-level co-occurrence matrix be... ,in and These represent the grayscale index variables of the two pixels. The spatial distance between pixel pairs The directional angle between pixel pairs; Calculate contrast parameters based on the gray-level co-occurrence matrix. Energy parameters and entropy parameter The calculation formulas are expressed as follows: ； ； ； in, To be at a given spatial distance and direction angle Under the condition of gray level With gray level Simultaneous normalized probability values This is the natural logarithm operator.

5. The intelligent analysis system for clinical research data in traditional Chinese medicine according to claim 1, characterized in that, The texture feature extraction acceleration module also includes a feature fusion and hierarchical decision logic unit. The feature fusion and hierarchical decision logic unit is connected to the gray-level co-occurrence matrix hardware computing array. It is used to perform weighted fusion processing on the feature parameters in multiple directions output by the gray-level co-occurrence matrix hardware computing array to generate comprehensive texture feature parameters, and output the tongue coating thickness grading result according to the comprehensive texture feature parameters. The tongue coating thickness grading result includes three grading categories: thin coating grading, slightly thick coating grading, and thick coating grading. For the There are several preset directions, and their corresponding contrast parameters are set as follows: Energy parameters are The entropy parameter is First, the feature parameters are normalized to obtain the normalized feature parameters. , , The normalization calculation formulas are expressed as follows: ； ； ； Then the directional comprehensive eigenvalue in this direction The calculation formula is expressed as: ； All Directional comprehensive feature value in each preset direction Weighted fusion is performed to obtain comprehensive texture feature parameters. The calculation formula is expressed as: ; The feature fusion and hierarchical decision-making logic unit will integrate the texture feature parameters. The results are compared with preset grading threshold ranges, and the corresponding tongue coating thickness grading results are mapped and output based on the comparison results. in, For all The maximum value among the original values ​​of the contrast parameter in each preset direction For all The maximum value among the original values ​​of the entropy parameter in each preset direction. , , These are the preset weighting coefficients corresponding to the normalized contrast parameter, energy parameter, and normalized entropy parameter, respectively. For the first Directional comprehensive feature values ​​in a preset direction, For the first The direction weight coefficients corresponding to each preset direction.

6. The intelligent analysis system for clinical research data in traditional Chinese medicine according to claim 1, characterized in that, The feature comparison unit includes a reference feature value storage array and a comparison logic module. The reference feature value storage array is used to store a pre-set reference feature vector for tongue coating thickness grading. The comparison logic module is connected to the reference feature value storage array and is used to perform a parallel comparison operation between the thickness grading features obtained after the tongue image collected by the current sub-center is processed by the tongue coating thickness grading feature extraction acceleration unit and the reference feature vector, and output a consistency verification conclusion based on the comparison result.

7. The intelligent analysis system for clinical research data in traditional Chinese medicine according to claim 6, characterized in that, The multi-center data consistency verification module also includes a deviation flag register, which is connected to the comparison logic module. When the comparison logic module detects that the deviation value between the thickness grading feature of the current sub-center and the reference feature vector exceeds the preset deviation range, it adds abnormal flag information to the thickness grading result.

8. The intelligent analysis system for clinical research data in traditional Chinese medicine according to claim 1, characterized in that, The intelligent analysis chip also includes a multi-center data synchronization interface, which is connected to a multi-center data consistency verification module. This interface is used to receive tongue image data from different clinical research sub-centers and to package the thickness grading results after verification by the multi-center data consistency verification module with the sub-center identification information corresponding to each sub-center before outputting them to the outside.

9. A method for intelligent analysis of clinical research data in traditional Chinese medicine, applied to the intelligent analysis system for clinical research data in traditional Chinese medicine as described in any one of claims 1-8, characterized in that, Includes the following steps: S1. Through the tongue coating-tongue body contrast enhancement hardware processing module, the input multicenter clinical study tongue image is subjected to contrast adaptive enhancement processing, and the contrast between the tongue coating area and the tongue body area is enhanced at the moment of image acquisition. S2. Through the tongue coating thickness grading feature extraction acceleration unit, the tongue coating thickness grading features are extracted in real time from the enhanced tongue image. Based on the texture feature extraction acceleration module, the texture feature parameters of the tongue coating area are calculated, and the texture feature parameters are mapped to the tongue coating thickness grading results and then output. S3. Through the multi-center data consistency verification module, real-time consistency verification processing is performed on the thickness grading results corresponding to the tongue images collected from different sub-centers, and anomaly marking is performed on the thickness grading results that exceed the preset deviation range.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is loaded into the intelligent analysis chip of the intelligent analysis system for clinical research data in traditional Chinese medicine, the intelligent analysis chip is configured to execute the intelligent analysis method for clinical research data in traditional Chinese medicine.

Images