Intelligent pulse condition identification method and system, electronic device and medium

Abstract

The application discloses a traditional Chinese medicine pulse condition intelligent discrimination method and system, electronic equipment and medium, and relates to the technical field of traditional Chinese medicine pulse diagnosis. The method comprises the following steps: obtaining individual characteristic parameters and current environment information of a measured person, generating a personalized pressure collection interval for the measured person by using a preset pressure range mapping model; obtaining array type pulse response data; constructing a pressure response three-dimensional model and a spatial form three-dimensional model based on multi-channel single cycle pulse waveform; quantifying pulse position characteristics based on the pressure response three-dimensional model; quantifying pulse number characteristics, pulse shape characteristics, pulse potential characteristics and pulse state characteristics based on the spatial form three-dimensional model; calling a preset opposite pulse condition constraint library to perform logical contradiction elimination on a five-dimensional characteristic set, and retaining a candidate pulse condition set in line with pulse diagnosis rules; and matching the candidate pulse condition set with a clinical standard pulse condition library, and outputting a final pulse condition comprehensive discrimination result. The method improves the completeness and accuracy of pulse condition characteristic expression, and enhances the recognition ability of complex and compatible pulses.

Description

Technical Field

[0001] This invention relates to the field of pulse diagnosis technology in traditional Chinese medicine, and in particular to a method and system for intelligent discrimination of pulse three-dimensional reconstruction and five-dimensional features based on adaptive multi-level pressure control. Background Technology

[0002] Traditional Chinese medicine pulse diagnosis is an important diagnostic method in which physicians use their fingers to palpate the radial artery at the cun, guan, and chi positions to perceive the dynamic characteristics of the pulse in terms of "position, rate, shape, and momentum," thereby judging the state of the body's qi, blood, yin, yang, and organ functions.

[0003] Traditional pulse diagnosis relies heavily on the sensitivity of a physician's fingertips and the accumulation of long-term clinical experience, making it highly subjective. Different physicians often perceive and describe the same pulse differently, making it difficult to guarantee the consistency and repeatability of diagnostic results. This leads to numerous difficulties in the standardization, inheritance, and promotion of pulse diagnosis techniques. Therefore, utilizing modern sensor technology, signal processing technology, and artificial intelligence algorithms to achieve objectification, digitization, and intelligentization of pulse diagnosis has become an important research direction in the field of TCM diagnostic and therapeutic equipment.

[0004] Various forms of pulse diagnosis instruments and their pulse analysis methods have emerged in the prior art. For example, Chinese invention patent application CN121196482A discloses an intelligent TCM pulse diagnosis system based on a hymenoplasmic robot. This system uses a seven-degree-of-freedom robotic arm with a bionic finger component at its end. It simultaneously acquires multimodal signals through a pressure sensor array at the fingertip, a bioelectric detection module, and a near-infrared optical module. Its control unit uses a force control algorithm to simulate three-order standard pulse-taking pressures: superficial (approximately 0.3N), medium (approximately 2.0N), and deep (approximately 6.0N). It then performs wavelet decomposition on the pressure signals and impedance fitting on the bioelectric signals through a feature extraction layer, finally outputting 28 categories of pulse characteristics using a deep neural network model.

[0005] However, the above-mentioned existing technical solutions still have the following shortcomings in practical applications: First, pressure control has limited individual adaptability and lacks dynamic response to environmental and physiological differences. Existing methods typically scan based on preset fixed pressure thresholds (e.g., 0.3N for floating and 6.0N for sinking) or fixed step pressure increments. This approach does not fully consider the changes in vascular depth and tissue compliance caused by individual physiological differences in subjects, such as age, body mass index (BMI), and subcutaneous fat thickness, nor does it take into account the influence of environmental factors such as temperature and air pressure on the state of surface blood vessels. For example, for obese patients or elderly patients with arteriosclerosis, using a uniform standard low pressure threshold often fails to reach effective pulse locations, leading to missed pulse location information or excessively long acquisition time; while for thin patients or patients with exposed blood vessels, the fixed pressure starting point may be too high, directly skipping the floating stage and missing superficial pulse characteristics.

[0006] Secondly, the spatiotemporal representation capabilities of multidimensional pulse information are insufficient, and the feature quantification dimension is singular. Current technologies at the signal analysis level largely rely on two-dimensional time-series waveforms (time-amplitude) acquired by single-point or array sensors. While frequency domain transformation can generate three-dimensional maps, they lack continuous modeling of the spatial distribution of pulse pulsation. For example, current technologies have failed to establish a joint response surface between pressure gradients and sensor spatial arrays, resulting in an inability to accurately quantify the continuous changes in pulse depth. Simultaneously, they have failed to effectively construct a three-dimensional morphological model of space-time-amplitude within a single cycle, making it difficult to objectively characterize key features of traditional Chinese medicine pulse diagnosis, such as pulse width and pulse strength, through geometric quantification methods like spatial integration. This leaves pulse recognition at the level of waveform specificity discrimination, making it difficult to accurately distinguish complex pulses (such as wiry pulses and tense pulses) with similar waveforms but significant differences in spatial distribution and strength characteristics.

[0007] Furthermore, the analytical decision-making logic lacks the constraints of traditional Chinese medicine (TCM) theory, resulting in a weak ability to identify concurrent pulses. Most existing pulse diagnosis systems directly utilize convolutional neural networks or deep neural networks to perform end-to-end classification output on preprocessed waveform data. Although some solutions introduce attention mechanisms or feature matrices, they lack a logical verification mechanism at the decision-making level based on fundamental TCM pulse diagnosis theories (such as the incompatibility of superficial and deep pulses, and the opposition between rapid and slow pulses). This purely data-driven approach easily produces contradictory results that do not conform to clinical reality. Moreover, when faced with multiple pulse characteristics simultaneously present in a patient (i.e., concurrent pulses), it lacks effective feature decoupling and logical fusion methods, leading to limited clinical consistency and interpretability of diagnostic results.

[0008] In summary, existing pulse diagnosis techniques still have significant room for improvement in terms of individualized adaptive pressure acquisition, physical model reconstruction of multi-dimensional pulse information, and multi-feature logical fusion diagnosis based on traditional Chinese medicine theory. Summary of the Invention

[0009] This invention aims to solve the problems of existing pulse diagnosis equipment, such as the inability to adaptively match individual differences leading to the omission of pulse position information, the inability to accurately characterize the spatial distribution and three-dimensional dynamic changes of pulse patterns due to reliance on single-point two-dimensional waveform analysis, and the low recognition rate of complex combined pulses.

[0010] In a first aspect, to solve the above-mentioned technical problems, a method for intelligent pulse diagnosis in Traditional Chinese Medicine is provided, comprising: The individual characteristic parameters of the test subject and the current environmental information are obtained, and a personalized pressure acquisition range is generated to suit the test subject using a preset pressure range mapping model. Based on the pressure response signal fed back when the array sensor first contacts the skin, the personalized pressure acquisition range is optimized to obtain a target pressure acquisition range that fully covers the three phases of floating, middle and sinking. Then, multi-level pressure scanning is performed within the target pressure acquisition range with preset pressure steps to obtain array-type pulse response data under multi-level pressure gradients. The array-type pulse response data is periodically segmented to obtain multi-channel single-cycle pulse waveforms. Based on the multi-channel single-cycle pulse waveforms, a three-dimensional pressure response model and a three-dimensional spatial morphology model are constructed respectively. Based on the aforementioned three-dimensional pressure response model, the optimal pulse-taking pressure value representing the depth of the pulse position is determined by a global extreme value search algorithm, and the pulse position characteristics are quantified accordingly. Based on the aforementioned three-dimensional spatial morphology model, the pulse number characteristics are quantified by the single-cycle time length, the pulse shape characteristics are quantified by the effective amplitude spatial distribution width, the pulse momentum characteristics are quantified by the combination of three-dimensional surface volume and maximum amplitude, and the pulse waveform category is identified by a three-dimensional convolutional neural network model to quantify the pulse state characteristics. A five-dimensional feature set is constructed based on five dimensions: pulse location, pulse number, pulse shape, pulse momentum, and pulse state. A preset opposing pulse pattern constraint library is called to remove logical contradictions from the five-dimensional feature set, retaining a candidate pulse pattern set that conforms to the rules of traditional Chinese medicine pulse diagnosis. The candidate pulse pattern set is then matched with a clinical standard pulse pattern library to output the final comprehensive pulse pattern discrimination result.

[0011] Furthermore, the generation of personalized stress collection intervals adapted to the test subject includes: The individual characteristic parameters of the test subject and the current environmental information are normalized; wherein, the individual characteristic parameters include at least age, gender and body mass index; and the current environmental information includes at least ambient temperature, ambient air pressure and seasonal information. Based on the normalized individual characteristic parameters, a baseline stress range is constructed using a linear regression model, and the upper limit of the baseline stress range is determined. and lower limit :

[0012] In the formula, Normalized Body Mass Index; Normalized age value; Assign a value to gender; , , , , , , , These are the weighted coefficients obtained by fitting clinical data; The baseline pressure range is corrected using the current environmental information to obtain the personalized pressure collection range. The correction formula is as follows:

[0013] In the formula, , These are the lower and upper limits of the personalized pressure acquisition range, respectively; , , These are temperature, air pressure, and seasonal correction factors, respectively. , , These are the corresponding correction weights.

[0014] Furthermore, obtaining the target pressure acquisition range that fully covers the floating, mid-water, and sinking phases includes: Based on the amplitude of the pressure response signal fed back when the array sensor first contacts the skin, the starting pressure value of the personalized pressure acquisition range is adjusted; if the amplitude of the pressure response signal at the first contact exceeds a preset threshold, the current contact pressure is set as the acquisition start pressure, and the acquisition termination pressure remains unchanged. The adjusted pressure range is divided into three equal parts: the floating pressure range, the mid-water pressure range, and the sinking pressure range.

[0015] Furthermore, the step of performing multi-level pressure scanning within the target pressure acquisition range at preset pressure steps to obtain array-type pulse response data under multi-level pressure gradients also includes: A pressure feedback closed-loop control method is employed to continuously monitor the real-time feedback values ​​of the array sensors and calculate the difference between the real-time feedback value and the current target pressure. ; When the difference When the pressure exceeds the first preset threshold, the motor is controlled to run at the first pressurization rate to quickly approach the target pressure. When the difference When the pressure is less than or equal to the first preset threshold, the motor is controlled to run at the second pressurization rate to enter a low-speed fine-tuning state; the second pressurization rate is less than the first pressurization rate. When the pressure change is detected to be stabilizing, pulse signal acquisition is triggered. After collecting pulse data for a preset duration at the current pressure level, the system automatically switches to the next target pressure level.

[0016] Furthermore, the pressure response three-dimensional model is reconstructed using the collected pressure value as the first dimension, the sensor spatial position as the second dimension, and the pulse signal amplitude as the third dimension; the spatial morphology three-dimensional model is reconstructed using the sensor spatial position as the first dimension, the time sampling points within a single cycle as the second dimension, and the pulse signal amplitude as the third dimension.

[0017] Furthermore, the construction of the three-dimensional pressure response model specifically includes: Peak value detection is performed on the single-cycle pulse waveforms collected by each sensor channel under each pressure level to locate the time point corresponding to the main peak. The pulse signal amplitudes of all sensor channels at the main peak time point are extracted to form a spatial discrete data set. ,in, For the first Level of pressure value acquisition, For the first Spatial location identifier of each sensor This corresponds to the pulse amplitude under different pressures and locations; The spatial discrete data set is smoothed using an interpolation method to generate a three-dimensional pressure response map with pressure value as the first dimension, sensor spatial location as the second dimension, and pulse amplitude as the third dimension.

[0018] Furthermore, constructing the three-dimensional model of the spatial morphology specifically includes: Select the continuous array of sensors covering the main pulsating region of the pulse. Single-cycle pulse waveform data from each sensor channel, of which ; As stated The spatial arrangement of the sensors is based on the first dimension coordinate. Using the normalized time sampling points within a single period as the second dimension coordinate The pulse amplitude of each sensor at each time point The third dimension coordinates form a discrete set of points; The discrete point set is reconstructed by bicubic spline interpolation to obtain a three-dimensional spatial morphology map.

[0019] Furthermore, the quantified pulse position features specifically include: In the three-dimensional pressure response model, the point of maximum amplitude is searched using the global extremum search algorithm; the pressure value corresponding to this point is the optimal pulse-taking pressure. ; Iterating through the optimal pulse sampling pressure The amplitude response of all sensor channels is calculated. If the amplitude response curves of all channels are within... neighborhood interval If the internal peak characteristics are present, then it is confirmed. To achieve the optimal pulse pressure value, where To collect pressure step values; The optimal pulse-taking pressure By comparing the pulse position with the pressure ranges of the floating, middle, and sinking pulses calculated through pressure range mapping, the category and depth of the pulse position can be determined.

[0020] Furthermore, the quantified pulse shape features specifically include: In the aforementioned three-dimensional spatial model, for each sampling moment within a single period Extract the spatial distribution range where the amplitude reaches a preset proportion of the maximum amplitude at that moment, and calculate the spatial length of this range as the effective spatial width at a single moment. ; The effective spatial width of a single time moment for all sampling times within a single period Calculate the mean value as the pulse space width. It satisfies: In the formula, This represents the total number of sampling points for a single-cycle signal. The pulse space width It is compared with a preset pulse width threshold and mapped to the corresponding qualitative description of the pulse shape.

[0021] Furthermore, the quantified pulse characteristics specifically include: Extract the maximum amplitude of all data points in the three-dimensional spatial model. ; The volume of the three-dimensional model of the spatial morphology was calculated using the integral method. ; For the maximum amplitude With the volume Perform normalized weighted combination to calculate the pulse index. It satisfies: In the formula, and These are the amplitude reference value and the volume reference value, respectively. and As a weighting factor; According to the pulse index Determine the strength level of the pulse.

[0022] Furthermore, the step of calling a preset library of opposing pulse constraints to perform logical contradiction removal on the five-dimensional feature set includes: After traversing the five-dimensional feature mapping, the resulting feature combinations are used to detect whether there are any opposing feature pairs that do not conform to the rules of traditional Chinese medicine pulse diagnosis. ; If the aforementioned pair of opposing features exists, then compare the features. and Each feature has a confidence level, and features with higher confidence levels are retained while features with lower confidence levels are removed to obtain a set of candidate pulse patterns. The confidence level is calculated by the degree of fit between the quantified feature value and the center of its respective judgment interval.

[0023] Furthermore, the step of matching the candidate pulse pattern set with the clinical standard pulse pattern database and outputting the final pulse pattern comprehensive judgment result specifically includes: Calculate the similarity score between the candidate pulse pattern set and each standard pulse pattern in the clinical standard pulse pattern library, which satisfies: In the formula, The first candidate pulse dimensional features, The first standard pulse dimensional features, For the first Preset weights for dimensional features This is a similarity calculation function; The standard pulse pattern with the highest similarity score is selected as the final comprehensive discrimination result.

[0024] Further, the periodic segmentation of the array-type pulse response data includes: Wavelet transform is used to decompose the original pulse signal into multiple layers to remove approximate components representing baseline drift and detail components representing high-frequency noise. The periodic low-frequency fluctuations caused by respiration were simulated and eliminated using cubic spline interpolation. The 3σ criterion is used to detect abnormal amplitude points in the signal, and the detected abnormal points are corrected by linear interpolation. The continuous pulse signal is segmented into single cycles based on the valley point detection algorithm, and the segmented single-cycle waveform is resampled and normalized to ensure that each cycle signal has a uniform sampling length.

[0025] A second aspect of the present invention provides a traditional Chinese medicine pulse diagnosis intelligent system, comprising: The pulse acquisition probe contains multiple array-type pressure sensing units, which are used to simultaneously acquire multi-channel pulse pressure signals at the radial artery cun, guan, and chi positions. The pressure control and drive module is used to receive control commands, drive the pulse acquisition probe to perform step-by-step pressurization, and provide real-time feedback of the current pressure value. The data processing and 3D reconstruction module is communicatively connected to the pulse acquisition probe and is used to execute the pressure response 3D model and spatial morphology 3D model construction steps in the method, as well as the five-dimensional feature quantification extraction steps of pulse position, pulse number, pulse shape, pulse momentum and pulse state. The intelligent diagnostic decision-making module is communicatively connected to the data processing and 3D reconstruction module. This module incorporates a conflicting pulse pattern constraint library and a clinical standard pulse pattern library, and is used to execute the pulse pattern comprehensive discrimination step in the method. The human-computer interaction and display module is used to input the individual characteristic parameters of the test subject and environmental information, and to display the pulse analysis results and three-dimensional reconstruction map.

[0026] Furthermore, the pressure sensing units are arranged in a three-row uniform array, wherein the first row has 3 pressure sensing units, the second row has 6 pressure sensing units, and the third row has 3 pressure sensing units.

[0027] Furthermore, it also includes a standard biomechanical pulse wave verification device, which comprises: Servo drive module, used to generate periodic motion with controllable displacement; A pressure output head, connected to the servo drive module, is used to apply dynamic pressure with a set amplitude and frequency to the pressure sensing unit to simulate pulse signals with different pulsation intensities and pulsation rates. The displacement sensing unit is used to monitor the output displacement of the pressure output head in real time.

[0028] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method when executing the program.

[0029] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the method.

[0030] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention utilizes a multi-level pressure control acquisition mechanism that adapts to both the individual and the environment. It dynamically generates personalized pressure scanning intervals based on the individual's characteristic parameters and real-time environmental information, ensuring complete coverage of the superficial, middle, and deep pulse stages and accurately acquiring array-style pulse response data under multiple pressure levels. Furthermore, by constructing a three-dimensional pressure response model and a three-dimensional spatial morphology model, it achieves a joint representation of pulse signals across multiple dimensions—pressure, space, time, and amplitude—breaking through the limitations of traditional two-dimensional waveform analysis. Further, based on the three-dimensional model, it extracts five-dimensional quantitative features of pulse position, pulse number, pulse shape, pulse momentum, and pulse state. Combined with a constraint library of opposing pulse images for logical contradiction elimination and a weighted matching with a standard pulse image library, it improves the completeness and accuracy of pulse feature expression, effectively enhancing the identification of complex pulse combinations. This achieves objectivity, standardization, and intelligence in pulse diagnosis acquisition and analysis, demonstrating good clinical applicability and repeatability. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is the overall flowchart of the adaptive pulse acquisition and multi-dimensional feature comprehensive analysis disclosed in this invention; Figure 2 This is a flowchart of motor control for pulse signal acquisition under adaptive pressure control disclosed in this invention. Figure 3 This is a schematic diagram of the spatial arrangement of the three-part array sensor of the inch, gauge, and ruler disclosed in this invention; Figure 4 This is a structural diagram of the standard mechanical pulse wave verification device disclosed in this invention; Figure 5 This is a flowchart of the pulse signal preprocessing method disclosed in this invention; Figure 6(a) is a waveform diagram of the original pulse signal disclosed in this invention; Figure 6(b) shows the waveform of the dynamic pulse signal after wavelet decomposition disclosed in this invention; Figure 6(c) is a waveform of the pulse signal after removing the respiratory wave using cubic spline interpolation disclosed in this invention. Figure 7 This is a schematic diagram of the pulse wave trough detection results after noise reduction disclosed in this invention; Figure 8(a) is a schematic diagram of the single-cycle pulse waveform after normalization and resampling disclosed in this invention; Figure 8(b) is a schematic diagram of the single-cycle pulse waveform after normalization and resampling disclosed in this invention; Figure 9 This is a three-dimensional schematic diagram of pulse pressure response under multi-level pressure acquisition disclosed in this invention; Figure 10 This is a schematic diagram of three-dimensional reconstruction of pulse spatial morphology under array sensing disclosed in this invention. Figure 11(a) is an example diagram of the 3D pulse wave morphology of common pulse patterns disclosed in this invention, showing a three-dimensional schematic diagram of a wiry pulse; Figure 11(b) is an example diagram of the 3D pulse wave morphology of common pulse patterns disclosed in this invention, showing a three-dimensional schematic diagram of slippery pulse; Figure 11(c) is an example diagram of the 3D pulse wave morphology of common pulse patterns disclosed in this invention, showing a three-dimensional schematic diagram of a normal pulse. Figure 11(d) is an example of the 3D pulse wave morphology of common pulse patterns disclosed in this invention, showing a three-dimensional schematic diagram of a tight pulse. Figure 11(e) is an example diagram of the 3D pulse wave morphology of common pulse patterns disclosed in this invention, showing a three-dimensional schematic diagram of a weak pulse; Figure 11(f) is an example diagram of the 3D pulse wave morphology of common pulse patterns disclosed in this invention, showing a three-dimensional schematic diagram of soft pulse. Figure 12 This invention discloses a pulse morphology classification model based on multi-scale 3D CNN. Figure 13 The training loss and accuracy curves of the pulse classification model disclosed in this invention are shown. Figure 14 This is a schematic diagram illustrating the application scenario of the system disclosed in this invention in a real clinical environment; Figure 15 This is a screenshot of the pulse diagnosis analysis report interface disclosed in this invention. Detailed Implementation

[0033] 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 are within the scope of protection of the present invention.

[0034] This invention aims to provide a method and system for adaptive multi-level pressure control of pulse diagnosis, including three-dimensional reconstruction and five-dimensional feature intelligent discrimination. It deeply integrates the individualized diagnostic principles of traditional Chinese medicine pulse diagnosis with modern sensing technology, three-dimensional modeling technology, and artificial intelligence decision-making technology. This constructs a closed-loop intelligent analysis system encompassing individualized pressure adaptive acquisition, spatiotemporal reconstruction of multi-dimensional pulse data, quantitative extraction of five-dimensional pulse features, and joint diagnosis based on logical constraints.

[0035] Please see Figure 1 The intelligent pulse diagnosis method for traditional Chinese medicine provided by this invention mainly includes the following four processing steps: The individual and environment-adaptive multi-level pressure control pulse acquisition steps are as follows: First, the individual characteristic parameters of the subject and the current environmental information are acquired, and a personalized pressure acquisition interval adapted to the subject is generated using a preset pressure range mapping model; then, based on the pressure response signal fed back when the array sensor first contacts the skin, the personalized pressure acquisition interval is optimized to obtain a target pressure acquisition range that fully covers the superficial, middle, and deep pulse stages; finally, for the optimized personalized pressure acquisition range, multi-level pressure scanning is performed within the target pressure acquisition range with preset pressure steps to obtain array-type pulse response data under multi-level pressure gradients, thereby reconstructing the complete dynamic response process of the pulse from superficial to deep.

[0036] The process involves processing and reconstructing multi-dimensional pulse data from an array sensor. Noise reduction and period segmentation are performed on the array-based pulse response data to obtain multi-channel single-cycle pulse waveforms. Based on these waveforms, a pressure response 3D model and a spatial morphology 3D model are constructed. The pressure response 3D model is reconstructed using the acquired pressure value as the first dimension, the sensor's spatial position as the second dimension, and the pulse signal amplitude as the third dimension, to visually reflect the dynamic response characteristics of the pulse as pressure changes. The spatial morphology 3D model is reconstructed using the sensor's spatial position as the first dimension, the time sampling points within a single cycle as the second dimension, and the pulse signal amplitude as the third dimension, to reflect the spatial distribution and continuous waveform variation characteristics of the pulse.

[0037] The five-dimensional features extraction steps of pulse location, number, shape, momentum, and state are as follows: Based on the three-dimensional pressure response model, the optimal pulse-taking pressure value representing the depth of the pulse location is determined by a global extreme value search algorithm, and the pulse location features are quantified accordingly; Based on the three-dimensional spatial morphology model, the pulse number features are quantified by the single-cycle time length, the pulse shape features are quantified by the effective amplitude spatial distribution width, the pulse momentum features are quantified by the three-dimensional curved surface volume and the maximum amplitude, and the pulse state features are quantified by identifying the pulse waveform category through a three-dimensional convolutional neural network model, thus realizing the quantitative extraction of the five-dimensional features of pulse location, number, shape, momentum, and state.

[0038] The steps for joint diagnosis and comprehensive judgment of the five-dimensional features of pulse location, number, shape, momentum, and state are as follows: Based on the five-dimensional features of pulse location, number, shape, momentum, and state, a five-dimensional feature set is constructed. A pre-set opposing pulse constraint library is called to eliminate logical contradictions in the five-dimensional feature set, retaining the candidate pulse set that conforms to the rules of traditional Chinese medicine pulse diagnosis. The candidate pulse set is matched with the clinical standard pulse library, and the final comprehensive judgment result of pulse is output.

[0039] The above four steps will be explained in detail below through specific embodiments.

[0040] Step 100: Acquisition of pulse images for multi-level pressure control that adapts to individual and environmental conditions.

[0041] The signal acquisition method simulates the traditional Chinese medicine pulse diagnosis method of "three parts and nine pulses," employing multi-probe synchronous acquisition to simultaneously obtain pulse signals from the radial artery at the cun, guan, and chi positions. Each probe incorporates multiple pressure sensing units, forming a high-density sensor array capable of comprehensively capturing the spatial distribution characteristics of the pulse. First, during the pre-acquisition phase, the subject's basic physiological parameters and real-time environmental information are acquired. Combined with a pre-set pressure mapping model, the pressure range for pulse diagnosis is dynamically adjusted, significantly improving the accuracy and individual adaptability of pulse signal acquisition. The pressure application process uses step-by-step pressure control, applying pulse diagnosis pressure in stages to replicate the actual pulse diagnosis process of a physician lifting, pressing, and searching.

[0042] Step 101: Spatial arrangement of the three-part array sensor (inch, gauge, ruler). In one specific embodiment, such as Figure 3 As shown, a three-probe array sensing structure (cun, guan, chi) is employed, with each probe containing 12 pressure sensing units. These 12 sensing units are arranged in a 3-6-3 configuration, specifically, the first row contains 3 sensing units, the second row contains 6 sensing units, and the third row contains 3 sensing units. The sensing units are evenly distributed in an array. This arrangement ensures sufficient spatial resolution for pulse signal acquisition to capture subtle spatial changes in the radial direction of the pulse, while avoiding signal crosstalk caused by excessively close spacing between adjacent sensing units, thus achieving accurate capture of the spatial distribution characteristics of the cun, guan, and chi pulses. A silicone finger mold can also be placed at the bottom of each probe to simulate the tactile sensation of a human finger.

[0043] In a further design, the data acquisition device also integrates temperature and barometric pressure sensors. The temperature sensor detects the ambient temperature in real time, while the barometric pressure sensor detects the ambient barometric pressure in real time. These sensors are installed inside the device housing, and their readings accurately reflect the actual conditions of the measurement environment.

[0044] like Figure 2 As shown, the entire pressurization and acquisition process employs a pressure feedback closed-loop control method. The lower-level computer continuously monitors the signal fed back from the pressure sensor and dynamically adjusts the motion state and pressurization rate of the actuator motor based on the difference between the feedback value and the target pressure value.

[0045] Step 102: Acquisition of Individual Characteristic Parameters and Environmental Information To achieve adaptive adjustment of pressure control, the individual characteristics of the test subjects and their environmental information were first collected. The individual characteristics included basic physiological parameters such as gender, age, height, and weight. This information was manually entered by the test subjects through a human-computer interaction interface before measurement. Height and weight data were further used to calculate the Body Mass Index (BMI), calculated as weight divided by the square of height, with units of kg / m². 2 Environmental information includes external factors such as the current season, ambient temperature, and air pressure. The season information is automatically calculated by the system based on the current date and mapped to four categories: spring, summer, autumn, and winter. Ambient temperature is obtained in real-time through the aforementioned temperature sensor, and ambient air pressure is obtained in real-time through the aforementioned air pressure sensor.

[0046] Step 103: Mapping of Individual Pulse Taking Pressure Range After acquiring the aforementioned individual and environmental information, the host computer generates a personalized pressure range suitable for the subject using a preset pressure range mapping model, ensuring that this range accurately covers the pressure ranges corresponding to the floating, mid-water, and sinking states. Simultaneously, to ensure the subject's safety, the system controls the upper pressure limit to below 40 kPa, ensuring that the collected pressure remains within the safe threshold range for human skin.

[0047] Specifically, the pressure range mapping process is as follows: First, the collected individual feature parameters and environmental information are normalized to eliminate the influence of differences in parameter units on the algorithm. The normalization formula is as follows:

[0048] In the formula, These are the original parameter values. and These represent the lower and upper limits of a reasonable value for this parameter, determined through statistical analysis of clinical data. The normalized values ​​range from [0, 1].

[0049] In one specific embodiment, the types of individual and environmental factor parameters considered, their names, and their value ranges are shown in Table 1: Table 1. Names and value ranges of the individual and environmental factors considered.

[0050] Based on normalized individual characteristic information, a baseline stress range is constructed using a linear regression model. The lower limit of the baseline stress range is... and upper limit Calculate according to the following formulas:

[0051] In the formula, Normalized Body Mass Index; Normalized age value; Assign a value to the gender, 0 for female and 1 for male; , , , , , , , These are weighted coefficients obtained by fitting clinical data, which is derived from fitting a large amount of clinical data. In a preferred embodiment, Take 2, Take 5. Take 2, Take 8, Take 5. Take 5. Take 2, Take 20.

[0052] Subsequently, the baseline pressure range is corrected using current environmental information to obtain a personalized pressure collection range. The correction formula is:

[0053] In the formula, , These are the lower and upper limits of the personalized pressure collection range, respectively. , , These are temperature, air pressure, and seasonal correction factors, each adjusted according to... Perform calculations, where These are the normalized values ​​for the corresponding parameters; , , The correction weights are respectively for temperature, pressure, and season. In a preferred embodiment, Take 0.2, Take 0.1, Take 0.15.

[0054] Based on the revised personalized pressure detection interval, the pulse pressure ranges corresponding to the superficial, middle, and deep pulses are divided using a proportional three-part division method. The division method is as follows:

[0055] in, Indicates the pressure interval. Indicates the float pressure range. Indicates the range of pressure to be taken. Indicates the range of settling pressure.

[0056] To verify the effectiveness of the aforementioned pressure mapping method, this embodiment selected three representative groups for testing: young, thin women; middle-aged men of standard build; and elderly, overweight men. These three groups exhibit significant differences in body shape, subcutaneous tissue thickness, and vascular elasticity, covering a range of common clinical individual differences. The experimental results are shown in Table 2. Table 2. Results of pulse pressure mapping in different populations and environments.

[0057] As shown in Table 2, the personalized pressure ranges generated by the system exhibit good discriminative power under different test individuals and environmental conditions. This method can adaptively map pulse sampling pressure to a suitable range based on the characteristics of different populations, demonstrating good population adaptability. Furthermore, the obtained personalized pressure detection range can be stably controlled within the safe and effective range of 8–40 kPa, ensuring both the effectiveness and integrity of pulse signal acquisition and safety during actual use.

[0058] Step 104: Adaptive Stepped Pressurization Control Method Driven by Multi-Stage Pressure This invention employs a pressure feedback closed-loop control method. During the control process, the signal fed back from the pressure sensor is continuously monitored, and the motion state of the actuator motor is dynamically adjusted based on the feedback value. The personalized pressure range generated above will also be adjusted again based on the feedback value when the sensor first comes into contact with human skin.

[0059] Specifically, when the probe descends and initially contacts the subject's skin, the sensor begins to return a pressure response signal. If the returned signal amplitude is already high at this point, it indicates that the subject's pulse is too superficial. If pressure is still applied starting from the lower limit of the originally calculated range, the previously acquired pressure levels may be invalid, or even fail to acquire a stable pulse signal due to insufficient pressure. Therefore, the system adjusts the pressure acquisition range: setting the starting pressure to the current contact pressure value while keeping the ending pressure unchanged. This mechanism effectively avoids pulse misjudgment caused by improper pressure range settings.

[0060] For the finally determined target pressure acquisition range, the system generates a target pressure sequence at equal intervals. In a preferred embodiment, the pressure step value is set to 1 kPa.

[0061] Figure 2 The adaptive rate control logic during the pressurization process is further demonstrated. At each target pressure point... At this point, the lower-level computer continuously collects the output value of the pressure sensor at a fixed sampling frequency to obtain the feedback value of the current pressure. Calculate the difference between the actual pressure and the target pressure. :

[0062] When the difference When the absolute value is large, the system controls the motor to run at a higher pressurization rate, so that the applied pressure quickly approaches the target pressure value, thereby shortening the acquisition waiting time; when the difference is large... As the pressure gradually decreases, the system reduces the motor's pressurization rate, putting the motor into a low-speed fine-tuning state. This low-speed fine-tuning process effectively reduces pressure overshoot and oscillation errors caused by the elastic rebound of human tissue. When the pressure change is detected to be stabilizing, i.e., the pressure fluctuation amplitude is less than the preset stability threshold, the system automatically activates the signal acquisition mode. To ensure that the signals acquired at each pressure level contain a sufficient number of complete pulse cycles, pulse data is continuously acquired for 5 seconds at each pressure level.

[0063] After the current pressure point is collected, the system automatically enters the next pressurization phase, increasing the pressure to the next target pressure value that increases by 1 kPa, and repeats the above collection process. This process continues until the entire pressure range is scanned, thereby obtaining the complete pulse response signal of the subject under a continuous pressure gradient.

[0064] To protect human safety, this invention incorporates multiple safety thresholds for anomaly protection during the pressurization process, as detailed below: First, the maximum pressurization pressure is set to 40 kPa. When the sensor pressure feedback value exceeds this safety limit, the lower-level computer immediately stops pulse output and sends a shutdown signal to the motor, causing the motor to stop running.

[0065] Secondly, the system monitors the status of the driver and the operating status of the motor in real time. If an abnormality is detected in the driver or an abnormal state such as motor stall or overcurrent, the protection process is triggered, and the data acquisition process is stopped immediately.

[0066] In addition, to prevent the motor from failing to automatically stop pressurizing in the event of an extreme malfunction, the data acquisition device is equipped with a manual pressure relief device. This device can be triggered in an emergency and can release the pressure applied to the wrist to 0 within one second.

[0067] After the pulse signal acquisition is completed, the lower-level computer controls the motor to run in reverse, causing the acquisition probe to gradually rise and the pressure to gradually decrease to the initial safe pressure range, thus completing one complete acquisition cycle.

[0068] To verify the accuracy and reliability of the aforementioned acquisition device in pulse signal acquisition, this invention also designs a standard mechanical pulse wave verification device. For example... Figure 4As shown, the device mainly consists of a servo drive module, a pressure output head, and a displacement sensing unit. This device can output a standard simulated pulse signal with stable amplitude and frequency. Its working principle is as follows: the amplitude of the simulated pulse generated by the device is controlled by the output displacement of the servo motor. The greater the motor displacement, the higher the dynamic pressure amplitude applied to the sensor by the drive head, thus simulating pulse pressure changes under different pulsation intensities in the human body. The standard pulse rate is set by the frequency of the motor's periodic movement; the higher the movement frequency, the faster the simulated pulse rate. This device allows for the customization of various standard signals combining pulse pressure and pulse rate to simulate the diversity of clinical pulse signals, thereby achieving quantitative and standardized verification of sensor acquisition performance and subsequent pulse analysis algorithms.

[0069] Verification experiments were conducted under six different combinations of standard pulse pressure and pulse rate, and the errors between the signal analysis results and the standard signals were compared. The experimental results are shown in Table 3. Table 3. Verification of the effectiveness of pulse sensor signal acquisition

[0070] As shown in Table 3, the measurement error in pulse pressure calculation is consistently within 10%, indicating that the acquisition device possesses good measurement consistency and stability. In pulse rate measurement, the measurement error remains consistently within ±2 bpm, maintaining high calculation accuracy regardless of various pulse pressure and pulse rate combinations. These results fully demonstrate the reliability of the acquisition device in signal acquisition, providing a solid hardware foundation for subsequent accurate processing and feature extraction of pulse signals.

[0071] Step 200: Preprocessing and 3D Reconstruction of Multidimensional Pulse Data from Array Sensors During pulse wave acquisition, various factors, such as electromagnetic interference from the surrounding environment, inherent noise of the acquisition equipment, and the subject's limb micro-movements and respiratory activities, can significantly affect the pulse wave signal, leading to high-frequency noise and baseline drift in the acquired raw signal. To ensure the accuracy of subsequent feature extraction, the raw pulse wave signal must first be preprocessed.

[0072] like Figure 5 As shown, the preprocessing process mainly includes two core steps: noise reduction and period segmentation. Subsequently, based on the preprocessed multidimensional pulse data, two three-dimensional structural maps with different physical meanings are constructed according to different dimensional combinations: a pressure response three-dimensional map and a spatial morphology three-dimensional map.

[0073] Step 201: A joint denoising method based on wavelet transform and spline interpolation This embodiment employs a combination of wavelet transform and cubic spline interpolation to perform joint denoising on the original signal. First, the original pulse signals acquired at each pressure level are decomposed into six levels using the sym6 wavelet basis. The decomposition process can be represented as follows:

[0074] In the formula, Represents the sixth-layer approximate component; Representing the Layer detail components.

[0075] Based on the frequency band division principle of wavelet decomposition and combined with the sampling frequency of the pulse acquisition device of this invention, the frequency range and main signal components corresponding to each decomposition component are shown in Table 4: Table 4 Frequency range of each component after wavelet decomposition

[0076] As shown in Table 4, the frequency range of approximate component A6 is approximately 0–0.78 Hz, mainly reflecting the slow variation trend of the signal and baseline drift; the frequency range of high-frequency detail components D1 and D2 is approximately 12.5–50 Hz, mainly containing high-frequency noise and power frequency interference. Therefore, during signal reconstruction, the system removes components A6, D1, and D2, retaining only components D3, D4, and D5 for signal reconstruction, thereby effectively suppressing baseline drift and high-frequency noise.

[0077] Even after wavelet processing, the signal may still contain periodic low-frequency fluctuations caused by the subject's respiratory activity. To eliminate such respiratory wave interference, this invention employs cubic spline interpolation technology for further signal processing. Specifically, the method involves first detecting the troughs in the signal as characteristic anchor points of the respiratory wave, then using a cubic spline function to perform curve fitting on these troughs to generate a smooth respiratory baseline estimation curve. Subsequently, subtracting this respiratory baseline estimation curve from the original signal yields the clean pulse signal after removing the respiratory wave.

[0078] Figure 6(a) shows the original pulse signal waveform, which shows obvious baseline drift and high-frequency spikes. Figure 6(b) shows the dynamic pulse signal waveform after wavelet decomposition and reconstruction, which shows that high-frequency noise has been effectively suppressed. Figure 6(c) shows the pulse signal waveform after further removal of respiratory waves by cubic spline interpolation, which shows that the signal baseline has been uniformly aligned on a horizontal straight line, and the overall waveform is clear and smooth.

[0079] To quantitatively verify the effectiveness of the joint denoising method, a signal-to-noise ratio (SNR) comparison analysis was performed on the pulse signals before and after denoising. The SNR calculation formula is as follows:

[0080] In the formula, and The values ​​represent signal power and noise power, respectively. The results show that the average signal-to-noise ratio (SNR) of the original pulse signal was 12.3 dB, which increased to an average of 33.5 dB after denoising, with a maximum of 35.7 dB. The SNR improvement after denoising was 18.6–23.2 dB, fully demonstrating that this combined denoising method can effectively suppress various interferences and improve the quality of the pulse signal, laying the foundation for subsequent pulse analysis.

[0081] Even after noise reduction, the signal may still contain a small number of abnormal amplitude points, such as spikes or sudden drops, caused by momentary sensor malfunctions or sudden micro-movements of the subject's limbs. If these outliers are not processed, they will severely interfere with the accuracy of subsequent single-cycle segmentation, leading to deviations in feature extraction.

[0082] This invention employs the 3σ criterion (Laida criterion) to detect and remove outliers from the normalized signal. First, the mean of the signal within the current analysis window is calculated. and standard deviation It will exceed the range. Signal points identified as outliers are processed using a linear interpolation method. This method fits a straight line to the adjacent valid signal points before and after the outlier, and the interpolated value at the outlier's location is used as a replacement value. This process ensures signal continuity and integrity, effectively preventing outliers from interfering with subsequent single-cycle segmentation.

[0083] Step 202: Single-cycle segmentation and extraction of pulse signal The acquired continuous pulse signal typically contains multiple cardiac cycles. Directly analyzing the continuous signal as a whole makes it difficult to accurately characterize the detailed morphological features of the pulse wave within a single cycle. Therefore, it is necessary to segment the continuous signal into independent single-cycle waveforms.

[0084] This invention employs a trough point detection-based method for period segmentation. Specifically, the trough search function from the peak-finding algorithm is used to scan the preprocessed signal and identify all local minima, i.e., trough points, in the signal. The signal segment between two adjacent trough points is defined as a complete pulse cycle. The trough point detection results are as follows: Figure 7 As shown in the figure, the detected trough points are marked with dots, clearly demonstrating the effect of periodic segmentation.

[0085] After segmenting all cycles, the slight differences in the duration of each cycle result in inconsistent sampling points for each single-cycle signal, making direct comparison and analysis impossible. Therefore, resampling and normalization of the segmented single-cycle signals are necessary. The resampling process uniformly interpolates and maps single-cycle signals of different lengths to signals with a fixed number of sampling points. In a preferred embodiment, the fixed number of sampling points is set to 150. This length preserves key morphological features such as the main pulse wave and dicrotic wave while effectively controlling the computational complexity of subsequent 3D reconstruction. The normalization process linearly scales the amplitude of each single-cycle signal, uniformly mapping its amplitude range to the range [0,1].

[0086] Figures 8(a)-8(b) Two standard single-cycle waveforms obtained after normalization and resampling are shown. It can be seen that the processed single-cycle signals are comparable on both the time and amplitude axes, providing a standardized input data format for subsequent 3D reconstruction and feature extraction.

[0087] Step 203: Pressure Response and Three-Dimensional Reconstruction of Spatial Morphology (1) Construction of a three-dimensional pressure response diagram: The three-dimensional pressure response diagram is used to characterize the distribution of pulse amplitude under different pulse-taking pressures and different sensor spatial locations, revealing the dynamic response characteristics of pulse position with pressure changes, and providing an objective basis for the quantitative determination of superficial, middle and deep pulses.

[0088] To highlight the core physiological characteristics of pulse pulsation and reduce data redundancy, this invention selects data from the main peak time point of the single-cycle pulse waveform at each pressure level to construct a three-dimensional pressure response graph. The main peak corresponds to the rising phase of the pulse pulsation cycle, and its amplitude can reflect the energy intensity of the pulse pulsation to the greatest extent, making it the most representative characteristic moment within a single cycle.

[0089] The specific construction process is as follows: For the pulse signals collected at each pressure level, peak detection is first performed on the single-cycle signals of each sensor channel to locate the time sampling point corresponding to the main peak. Select pulse signal amplitude data from all sensor channels at this time point.

[0090] Subsequently, the actual pressure applied by the acquisition probe is used as the first coordinate axis, i.e., the X-axis, with a value range covering the pressure intervals covered during the actual acquisition process, discretized at 1 kPa intervals; the spatial arrangement of each sensor is used as the second coordinate axis, i.e., the Y-axis, arranged sequentially according to the sensor channel number; and the pulse signal amplitude at the corresponding pressure and position is used as the third coordinate axis, i.e., the Z-axis. This constitutes a spatially discrete data set, denoted as […]. ,in, For the first Level of pressure value acquisition, For the first Spatial location identifier of each sensor In order to be under pressure Spatial location The amplitude of the main wave peak below.

[0091] After extracting and constructing data from the cun, guan, and chi probes respectively, the amplitude signals from the three probes are weighted and fused at the same pressure level. The weighting fusion can employ equal-weighted averaging or an adaptive weighting method based on sensor sensitivity, ultimately forming a comprehensive pressure response characterization that reflects the overall pulse pattern and positional characteristics. .

[0092] To improve the spatial resolution and visual smoothness of the 3D image, an interpolation method is used to smooth the discrete grid data points. In a preferred embodiment, a bicubic interpolation algorithm is used to refine the grid in both the pressure and spatial dimensions, enabling the 3D surface to more accurately reproduce the continuous variation of pulse amplitude across different spatial locations and pressure levels.

[0093] like Figure 9 As shown, the reconstructed 3D surface fully presents the dynamic law of pulse amplitude variation with pressure and spatial position. Along the X-axis, i.e., the pressure axis, the surface exhibits an arched trend that first rises and then falls, intuitively reflecting the dynamic change characteristic of pulse amplitude first increasing and then decreasing with increasing pulse-taking pressure. The pressure value corresponding to the apex of this arch is the region where the optimal pulse-taking pressure is located. Along the Y-axis, i.e., the spatial array axis, the surface shows the spatial distribution difference of pulse amplitude on the multi-channel sensor array of Cun, Guan, and Chi at the same pulse-taking pressure level, reflecting the positional offset information of the strongest pulse point relative to the center of the sensor array.

[0094] (2) Construction of three-dimensional spatial morphology: To further characterize the joint distribution characteristics of pulse waves in the spatial and temporal dimensions within a single cycle, this invention combines multi-sensor spatial array data with segmented single-cycle pulse sequence signals to construct a three-dimensional map of pulse spatial morphology.

[0095] according to Figure 3 The spatial arrangement of the sensors shown indicates that the six sensor channels in the middle row most effectively cover the main pulsating area of ​​the pulse. Therefore, data from these six sensors are selected as the data basis for constructing the three-dimensional spatial morphology diagram.

[0096] Let the first The single-cycle signal of each sensor channel contains There are 1,000 equally spaced sampling points, of which Take values ​​from 1 to 6. Denote the time series amplitude vector of this channel as: ,in, Indicates the first The sensor at the first Pulse amplitude values ​​at each time sampling point.

[0097] Taking the spatial arrangement of the six sensors as the first dimension, i.e., the X-axis, let the first dimension be denoted as the first dimension. The position quantization value of each sensor is The geometric spacing of each sensor is set according to the actual physical arrangement to quantify spatial attributes and provide a spatial scale benchmark for subsequent pulse width feature calculation. The normalized time sampling points within a single cycle are used as the second dimension coordinate, i.e., the Y-axis, and the time coordinate vector for a single cycle is... The pulse wave amplitude of each sensor at each time point is used as the third-dimensional coordinate, i.e., the Z-axis. This forms a three-dimensional discrete point set, with each data point corresponding to a point in three-dimensional space. After performing the above processing on the data at the cun, guan, and chi positions respectively, the data at the same spatial location and time coordinate are fused by channel to obtain fused data that reflects the overall spatial characteristics of the cun, guan, and chi pulse. .

[0098] Subsequently, bicubic spline interpolation is used to reconstruct the fused discrete point set. In a preferred embodiment, the number of sampling points in the spatial dimension is expanded from the original 6 to 600, while the number of sampling points in the temporal dimension remains unchanged. After this interpolation process, the spatial resolution of the 3D image is significantly improved, and the surface becomes smoother and more continuous.

[0099] like Figure 10 As shown, this three-dimensional spatial morphology fully characterizes the spatiotemporal distribution features of a single-cycle pulse wave. Along the X-axis, i.e., the spatial dimension, the lateral extension of the surface reflects the pulsation width in the radial artery, visually displaying the changes in pulse width. Along the Y-axis, i.e., the time dimension, the surface presents the temporal evolution of the pulse waveform, fully restoring key morphological structures such as the steepness of the rising limb, the sharpness of the main peak, and the location and amplitude of the dicrotic wave. The Z-axis height, i.e., the amplitude dimension, corresponds to the pulsation intensity of the pulse wave, and the overall height of the surface reflects the strength of the pulse pulsation energy.

[0100] Figures 11(a) to 11(f) show examples of three-dimensional pulse wave morphology for six common pulse types. Figure 11(a) illustrates a wiry pulse, characterized by a steep waveform, sharp main peak, and large spatial gradient; Figure 11(b) illustrates a slippery pulse, with a smooth and rounded waveform, a distinct dicrotic wave, and relatively uniform spatial amplitude distribution; Figure 11(c) illustrates a level pulse, characterized by a gentle and natural waveform with all features centered; Figure 11(d) illustrates a tense pulse, characterized by a tense pulse shape, a steep ascending limb, and high spatial concentration; Figure 11(e) illustrates a weak pulse, characterized by a relatively low overall amplitude and a small surface volume; and Figure 11(f) illustrates a soft pulse, characterized by a soft pulse shape with relatively gentle ascending and descending limbs. These intuitive three-dimensional morphological differences provide a rich feature base for subsequent automatic pulse state recognition based on deep learning models.

[0101] Step 300: Extraction of pulse position and dynamic features Based on the reconstructed three-dimensional pressure response diagram and three-dimensional spatial morphology diagram, the present invention further extracts the five-dimensional quantitative features of pulse position, number, shape, potential and state.

[0102] Step 301: Pulse location feature extraction and determination of optimal pulse-taking pressure Pulse location characteristics reflect the depth of the pulse's manifestation and are an important basis for determining the location of pathogenic factors, both superficial and deep. This invention, based on a three-dimensional pressure response graph, analyzes the variation patterns of pulse amplitude under different pulse-taking pressures to determine the optimal pulse-taking pressure value that maximizes pulse intensity, and uses this value to quantify pulse depth.

[0103] The specific extraction method is as follows: First, a global extremum search algorithm is performed on the three-dimensional pressure response surface to find the global maximum point on the surface. The coordinates corresponding to this maximum point are the pulse-taking pressure and its corresponding spatial location that maximizes the pulse intensity. The search process can be represented as follows:

[0104] in, To obtain the optimal sampling pressure value, This marks the spatial location where the strongest pulsation occurs under this pressure. For the pressure response three-dimensional surface at point The magnitude function at that point.

[0105] Subsequently, to eliminate false maximum pressure points caused by local sensor fluctuations or noise interference, the system verifies the validity of the search results. Specifically, it iterates through... The amplitude response of all sensing units under the same pressure value is compared to verify whether the response curves of each sensor channel are all within the range of [value missing]. neighborhood space The internal peak characteristics are obvious, among which To collect pressure step values.

[0106] If the response curves of all Y-axis positions, i.e., different sensor channels, reach their respective local maximum values ​​within this neighborhood interval, then it is confirmed that... This is the effective optimal pulse pressure point. If only a few sensors show a peak at this point while other sensors respond slowly or too low, then this point is determined to be a spurious extreme point caused by local fluctuations or noise. The system will then perform a neighborhood search again in the remaining region after excluding this point to find the true optimal pressure point that meets the conditions.

[0107] If no suitable pressure point is found after traversing the entire pressure range, the system considers the signal acquisition quality to be poor and that there is significant interference during the acquisition process. The system will then prompt the user to re-acquire the signal.

[0108] Obtain the effective optimal pulse pressure value Then, it is compared with the pressure ranges for floating, mid-water, and sinking calculated in step 100 through pressure range mapping to determine... The pressure range into which the pulse falls determines whether the pulse is classified as superficial, middle, or deep.

[0109] To meet the needs of refined diagnosis in traditional Chinese medicine clinical practice, this invention further divides the superficial and deep pulse positions into three levels. Specifically: Floating pulse position according to In the float pressure range The relative position within is divided into three levels: slightly floating, floating, and slightly floating. near The lower limit is slightly floating, located at The middle part is floating, and the one near it is floating. The upper limit is considered to be slightly skewed.

[0110] According to the pulse position of deep type In the sinking pressure range The relative position within is divided into three levels: slightly sinking, sinking, and partially sinking. near The lower limit is slightly heavy, located at The middle part is sinking, close to The upper limit is considered to be slightly heavier.

[0111] This grading mechanism can more accurately describe the differences in pulse depth and better reflect the severity and urgency of the symptoms.

[0112] Step 302: Pulse count feature extraction Pulse rate, referring to the frequency of pulse beats, is one of the important indicators for judging cold / heat and deficiency / excess. This invention calculates pulse rate based on the sampling length of a single-cycle pulse signal. ,in, Pulse rate, measured in beats per minute; The length of the sampling points after resampling of a single-cycle signal is determined by the single-cycle segmentation and resampling steps. This represents the time interval between adjacent sampling points, measured in seconds (s), and is calculated from the sampling frequency of the acquisition device. Different pulse rate values ​​can be mapped to different qualitative descriptions of the pulse pattern.

[0113] In one specific embodiment: when A pulse rate <60 is considered a slow pulse.

[0114] When 60≤ If the pulse rate is less than 70, it is considered bradycardia.

[0115] When 70 < A pulse ≤90 is considered normal.

[0116] When 90 < When the value is ≤120, it is determined to be a pulse.

[0117] when A pulse rate greater than 120 is considered a rapid pulse.

[0118] Step 303: Pulse Shape Feature Extraction Pulse shape features are used to reflect the thickness and spatial fullness of blood vessels. This invention quantifies the pulse space width by extracting it from a three-dimensional spatial morphology diagram.

[0119] The specific extraction method is as follows: For each sampling moment within a single period We extract the spatial distribution interval where the third dimension, i.e., the amplitude dimension, reaches 80% of the maximum amplitude at that moment. The left and right boundaries of this interval are denoted as... and Calculate the length of this interval in the spatial dimension, which serves as the effective spatial width at a single time point. .

[0120] For all within a single period The arithmetic mean of the effective spatial widths at each sampling time point is used as the pulse spatial width of that single-cycle pulse wave. :

[0121] in, This represents the total number of sampling points for a single-cycle signal. For the first The effective space width is calculated at each sampling time.

[0122] Based on the calculated pulse space width Combining the traditional Chinese medicine definition of pulse width, it can be mapped to different qualitative descriptions of pulse shape.

[0123] In one specific embodiment: when When the value is less than 2.2, it is determined to be a thready pulse.

[0124] When 2.2 < A pulse value ≤3.0 is considered normal.

[0125] when When the value is greater than 3.0, it is determined to be a wide pulse.

[0126] Step 304: Extraction of pulse characteristics Pulse strength refers to the intensity and dynamic trend of the pulse, and is an important objective basis for differentiating between deficiency and excess. This invention combines two parameters—maximum amplitude and three-dimensional volume—extracted from a three-dimensional spatial diagram to jointly quantify pulse strength characteristics.

[0127] First, extract the maximum amplitude. By traversing the amplitude data of all data points in the 3D spatial morphology graph, the global amplitude extremum is extracted, which is the maximum amplitude value for that single cycle.

[0128] in, The third in the three-dimensional diagram of spatial morphology The spatial sampling point, the first The amplitude at each time sampling point.

[0129] Secondly, calculate the total volume of the 3D graph. The volume of the 3D plot is obtained by performing double numerical integration along the spatial and temporal dimensions on the amplitude surface.

[0130] For each sampling time First, the spatial amplitude curve at that moment is calculated using the trapezoidal numerical integration method. Cross-sectional area enclosed by the spatial coordinate axes :

[0131] In the formula, The spacing between the spatial sampling points after interpolation; and These are the left and right boundary coordinates of the effective pulse width in the spatial dimension; and For the corresponding spatial sampling point index; For spatial location At the moment The amplitude.

[0132] Subsequently, the trapezoidal numerical integration method was used again along the time dimension to analyze the cross-sectional area sequence. Integrating, we obtain the total volume of the 3D graph. :

[0133] In the formula, The time interval between adjacent sampling points; This represents the total number of sampling points for a single-cycle signal.

[0134] To integrate the above two parameters into a unified pulse potential quantification index, the following steps are taken: and Perform normalized weighted combination to calculate the pulse index. :

[0135] In the formula, The amplitude reference value is determined by the average of the maximum amplitude values ​​of healthy individuals obtained from a large number of clinical trial data. This is a volume reference value, also derived from clinical data statistics. and As a weighting factor, in a preferred embodiment, the amplitude and volume contribute equally to the pulse potential, i.e. and Both are 0.5.

[0136] Based on the calculated pulse index The pulse is mapped to three levels: when When the pulse reaches the preset threshold for a solid pulse, it is determined to be a solid pulse, indicating that the pathogenic factors are strong while the body's vital energy is not weak.

[0137] when When the pulse is within the preset normal pulse threshold range, it is determined to be a normal pulse state, indicating that the Qi and blood are balanced.

[0138] when When the pulse rate is below the preset threshold for a weak pulse, it is determined to be a weak pulse, indicating insufficient vital energy.

[0139] In one specific embodiment: when A value less than 0.6 indicates a weak pulse.

[0140] When 0.6 < A pulse value ≤1 is considered a normal pulse.

[0141] when When the value is greater than 1, it is determined to be a strong pulse.

[0142] This pulse index can effectively distinguish between excess and deficiency pulse patterns, providing an objective quantitative basis for TCM differentiation of excess and deficiency.

[0143] Step 305: Pulse feature extraction Pulse characteristics refer to the waveform morphology of the pulse wave over time. The waveform characteristics of a single-cycle pulse wave differ significantly under different pathological or physiological conditions. For example, a wiry pulse is characterized by a steep waveform and a sharp main peak; a slippery pulse is characterized by a rounded waveform and a smooth descending branch; and a choppy pulse is characterized by a disordered waveform and a halting ascending branch.

[0144] This invention employs a multi-scale three-dimensional convolutional neural network as a pulse state classification model, using a reconstructed and standardized three-dimensional pulse spatial morphology map as model input. Through deep learning, it automatically mines the deep features of different pulse states in the spatial, temporal, and intensity dimensions, and finally outputs the category of the pulse state and its corresponding classification probability.

[0145] like Figure 12 The diagram shows the structure of a pulse morphology classification model based on a multi-scale 3D convolutional neural network. The classification model contains three parallel 3D convolutional modules. Each parallel module consists of three 3D convolutional layers and one 3D max pooling layer stacked together. Each module uses 3D convolutional kernels of different sizes, corresponding to different receptive field scales, enabling the model to simultaneously capture both local fine features and global contour features in the 3D pulse atlas.

[0146] The feature maps output by the three parallel modules are concatenated and fused along the channel dimension to form a multi-scale fused feature tensor. This fusion mechanism enhances the model's ability to represent complex pulse morphologies. Subsequently, the fused feature tensor is flattened and input into a fully connected layer, where it is mapped through a nonlinear transformation to the final pulse classification result and the probability distribution of each category.

[0147] The pulse recognition dataset was constructed based on pulse data from 2000 clinically collected individuals. The samples cover different age groups, genders, constitution types, and various clinically common pulse types, ensuring the diversity and representativeness of the dataset. After preprocessing and 3D reconstruction of the original data, several experienced TCM experts manually screened and labeled the data based on temporal waveform features and 3D morphological features. The final result is a single-cycle 3D atlas dataset containing multiple standard pulse types, including normal pulse, wiry pulse, slippery pulse, tense pulse, weak pulse, and soft pulse. Figure 11 shows examples of the 3D reconstruction results for some of the pulse types.

[0148] During model training, the dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio, used for model parameter optimization, hyperparameter tuning, and final performance evaluation, respectively. Furthermore, the initial learning rate was set to 0.001, cross-entropy loss was used, the number of training epochs was set to 100, the batch size was set to 16, and the Adam optimizer was selected. An early stopping strategy was employed to monitor the validation set loss during training; if the validation set loss did not decrease for several consecutive epochs, training was terminated early to prevent overfitting.

[0149] Figure 13 The graph shows the training loss versus accuracy of the pulse classification model. Figure 13 The upper curve in the graph represents the change in the model's loss function value on the training and validation sets as a function of the number of training epochs, while the lower curve represents the change in classification accuracy as a function of the number of training epochs. Figure 13 It can be seen that the model reached convergence after about 70 rounds of training, triggering the early stopping mechanism and stopping training.

[0150] The trained model is evaluated on the test set using accuracy. Recall rate The F1 score is used as the evaluation metric. The calculation methods for each metric are as follows:

[0151] in, This represents the number of true positives, i.e., the number of samples that the model correctly predicts as positive. This represents the number of true negative examples, that is, the number of samples that the model correctly predicts as negative. This indicates the number of false positives, that is, the number of samples that the model incorrectly predicted as positive. This indicates the number of false negatives, that is, the number of samples that the model incorrectly predicted as negative.

[0152] The experimental results are shown in Table 5: Table 5 Validation of Pulse Morphology Classification Results

[0153] As shown in Table 5, the model performs stably on the pulse shape classification task, with all comprehensive evaluation indicators at a high level. This result demonstrates that the technical solution combining the three-dimensional spatial morphological atlas constructed in this invention with a multi-scale three-dimensional convolutional neural network can effectively capture the morphological differences of different pulse shapes in three-dimensional space, achieving reliable automatic identification of pulse waveform features.

[0154] Step 400: Joint diagnosis and comprehensive discrimination based on five-dimensional features of bit-state dynamics Traditional pulse diagnosis relies heavily on two-dimensional temporal waveform analysis, failing to fully consider the multidimensional characteristics of pulse in terms of spatial distribution, pressure response, and temporal evolution. This results in incomplete diagnostic information and insufficient feature differentiation. This invention utilizes the diagnostic concept of "positional pattern" in traditional Chinese medicine pulse diagnosis, combined with modern pulse waveform analysis, to construct a five-dimensional joint diagnostic method based on positional pattern, thereby achieving objective pulse identification.

[0155] Step 401: Multidimensional pulse feature quantification and mapping First, establish a mapping relationship between characteristic values ​​and actual qualitative descriptions of pulse patterns, and convert the various quantitative indicators extracted in the previous steps into pulse descriptions that conform to the conventions of traditional Chinese medicine terminology.

[0156] Position-dimensional mapping: Based on optimal pulse-taking pressure The comparison results with the three pressure zones of floating, middle, and deep pulse positions map the pulse positions into three categories: floating, middle, and deep. Further, within the floating and deep categories, finer levels are output, such as slightly floating, floating, slightly floating, slightly deep, deep, and slightly deep. Simultaneously, the degree of fit between the eigenvalue and the center of the interval is calculated as the confidence level for the position dimension. The closer to the center of the interval, the higher the confidence level; the closer to the interval boundary, the lower the confidence level.

[0157] Numerical dimension mapping: based on the calculated pulse rate This is mapped to qualitative descriptions such as slow, gradual, rapid, and fast. Confidence level is determined by the proximity of the pulse rate value to the center of the standard interval.

[0158] Shape-dimensional mapping: based on pulse space width This is mapped to qualitative descriptions such as fine, constant, and wide. The confidence level is determined by the degree to which the width value fits the center of the decision interval.

[0159] Potential dimension mapping: based on pulse potential index This is mapped to qualitative descriptions such as imaginary, constant, and real. The confidence level is determined by the degree of fit between the index value and the center of the decision interval.

[0160] State dimension mapping: The three-dimensional convolutional neural network model directly outputs the pulse shape category, such as string pulse, slippery pulse, and flat pulse, along with the corresponding classification probability value. This classification probability value is directly used as the confidence level of the state dimension.

[0161] Step 402: Logical verification based on the opposing pulse constraint library The pulse characteristics and their confidence levels obtained from mapping the above dimensions are combined into a five-dimensional feature set. Since the discrimination of each dimension is relatively independent, there may be pathologically mutually exclusive or logically contradictory pulse combinations between different dimensions in the feature set. For example, floating pulse and deep pulse both belong to pulse position characteristics, and a single subject cannot simultaneously exhibit both floating pulse and deep pulse; rapid pulse and slow pulse both belong to pulse rate characteristics, and they cannot appear simultaneously.

[0162] To address this issue, this invention constructs a constraint library of opposing pulse patterns based on the fundamental theories of traditional Chinese medicine pulse diagnosis. The constraint library predefines all opposing pulse patterns that do not conform to the rules of traditional Chinese medicine pulse diagnosis.

[0163] The logic verification process is as follows: First, the results of the five-dimensional feature combinations are traversed and checked. If opposing feature pairs are detected... Then, compare the confidence levels carried by the two features. Retain the feature with higher confidence and discard the feature with lower confidence. This process can be formally represented as: If there are opposing characteristics ,and The confidence level is higher than The confidence level is then retained. Remove ;like The confidence level is lower than The confidence level is then retained. Remove .

[0164] After performing the above filtering operation on all detected opposing feature pairs, the feature results of each dimension that do not have logical contradictions are saved to form a candidate pulse set.

[0165] Step 403: Weighted matching based on the clinical standard pulse database To further improve diagnostic accuracy and clinical applicability, this invention utilizes the clinical expertise of a team of senior TCM physicians to construct a standardized pulse database. This database covers common single pulse types and combined pulse types, and provides standardized definitions for the characteristic manifestations of each pulse type across five dimensions.

[0166] The candidate pulse image set obtained after logical verification is compared with each record in the standard pulse image database to calculate similarity, which satisfies the following formula:

[0167] in, For the first The feature weights for each dimension are set by clinical experts based on the importance of each dimension in the overall pulse diagnosis. For similarity calculation functions, distance-based similarity functions can be used for continuous numerical features, while matching-based similarity functions can be used for discrete categorical features. For candidate pulse signs in the first Feature values ​​or category identifiers in each dimension; The standard pulse is at the first Feature values ​​or category identifiers in each dimension.

[0168] After calculating the similarity score between the candidate pulse image set and each record in the standard pulse image database, the standard pulse image with the highest score is selected as the final comprehensive discrimination result.

[0169] This standard pulse database is dynamically updated. As clinical data accumulates, the database can be updated with the characteristics of typical cases that have been reviewed and confirmed by experts, continuously optimizing matching rules and weighting parameters to improve diagnostic accuracy and clinical consistency.

[0170] Figure 14 This invention demonstrates the application scenarios of the system in a real clinical environment; Figure 15 The report presents a typical pulse diagnosis analysis report interface, which includes a three-dimensional reconstruction map of the pulse at the cun, guan, and chi positions, quantitative values ​​and qualitative descriptions of the five-dimensional characteristics of position, number, shape, momentum, and state, and comprehensive pulse diagnosis results.

[0171] Step 404: Clinical Validation and Performance Evaluation To verify the accuracy and reliability of the pulse analysis method of the present invention, a clinical controlled experiment was conducted on 150 subjects covering different age groups and different body types.

[0172] In the experiment, two experienced TCM experts first independently diagnosed the pulse of each subject, providing overall pulse findings and diagnostic conclusions for pulse location, shape, strength, and rate. When the two experts disagreed, a third senior expert was brought in to make the final decision. The resulting expert consensus served as the standard reference result for this experiment.

[0173] Subsequently, pulse signals were acquired and intelligently analyzed on the same subject using the device and method described in this invention. The analysis results automatically output by the system were compared with the standard reference results of expert diagnosis. The matching results of each dimension of features and the overall pulse results are shown in Table 6: Table 6. Verification of the accuracy of pulse number and overall results across various dimensions.

[0174] As shown in Table 6, the overall pulse diagnosis results of this invention show an 82% consistency with expert diagnoses. Among the various sub-dimensions, pulse location and pulse rate show the highest consistency, both exceeding 90%; the consistency for pulse shape and pulse strength is slightly lower, around 80% to 85%. This difference mainly stems from the higher requirements for sensor spatial resolution and pressure control precision in the precise discrimination of pulse shape and strength, a common technical challenge in the field of objective pulse diagnosis. These results verify the good reliability of this invention in intelligent pulse diagnosis and its significant clinical application and auxiliary diagnostic value.

[0175] Step 405, Repeatability and Stability Testing To further verify the repeatability and stability of the pulse recognition system of this invention, four subjects were selected and their pulses were repeatedly collected and analyzed at different time periods under the same environmental conditions. The pulse analysis results are shown in Table 7. Table 7. Stability Test Results of Pulse Recognition under Repeated Acquisitions

[0176] As shown in Table 7, the system demonstrates high consistency in pulse characteristic determination across multiple repeated acquisitions, exhibiting good overall stability and repeatability. While slight fluctuations exist in the quantified values ​​of pulse position and pulse rate, and minor variations in pulse width, the pulse shape recognition results remain highly consistent. This is primarily because the quantification of pulse position and pulse rate is more sensitive to physiological fluctuations such as acquisition pressure, limb micro-movements, and respiratory rhythm, and is more easily affected by short-term, minute changes. Nevertheless, the qualitative judgments for each dimension remain largely consistent across multiple tests, resulting in a stable and reliable overall diagnostic outcome. This result fully demonstrates that the pulse acquisition and analysis scheme proposed in this invention possesses excellent repeatability.

[0177] This invention also provides a TCM pulse diagnosis intelligent system, mainly comprising a pulse acquisition probe, a pressure control and drive module, a data processing and 3D reconstruction module, an intelligent diagnostic decision module, a human-computer interaction and display module, and a standard mechanical pulse wave verification device. The pulse acquisition probe includes multiple array-type pressure sensing units for simultaneously acquiring multi-channel pulse pressure signals at the radial artery's cun, guan, and chi positions. The pressure control and drive module receives control commands, drives the pulse acquisition probe to perform step-by-step pressurization actions, and provides real-time feedback of the current pressure value. The data processing and 3D reconstruction module is communicatively connected to the pulse acquisition probe and performs the pressure response 3D model and spatial morphology 3D model construction steps, as well as the five-dimensional feature quantification extraction steps of pulse position, pulse number, pulse shape, pulse momentum, and pulse state. The intelligent diagnostic decision module is communicatively connected to the data processing and 3D reconstruction module, and incorporates an opposing pulse constraint library and a clinical standard pulse library to perform the comprehensive pulse diagnosis steps. The human-computer interaction and display module allows input of the subject's individual characteristic parameters and environmental information, and displays the pulse analysis results and 3D reconstruction atlas.

[0178] In a preferred embodiment, the pressure sensing units are arranged in a three-row uniform array, wherein the first row has 3 pressure sensing units, the second row has 6 pressure sensing units, and the third row has 3 pressure sensing units.

[0179] The standard mechanical pulse wave verification device includes a servo drive module, a pressure output head, and a displacement sensing unit. The servo drive module is used to generate periodic motion with controllable displacement. The pressure output head is connected to the servo drive module and is used to apply dynamic pressure with a set amplitude and frequency to the pressure sensing unit to simulate pulse signals with different pulsation intensities and rates. The displacement sensing unit is used to monitor the output displacement of the pressure output head in real time.

[0180] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.

[0181] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0182] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent pulse diagnosis in Traditional Chinese Medicine, characterized in that, include: The individual characteristic parameters of the test subject and the current environmental information are obtained to generate a personalized stress collection range that is suitable for the test subject. Multi-level pressure scanning is performed within the personalized pressure acquisition range to obtain array-type pulse response data under multi-level pressure gradients; The array-type pulse response data is periodically segmented to obtain multi-channel single-cycle pulse waveforms. Based on these multi-channel single-cycle pulse waveforms, a three-dimensional pressure response model and a three-dimensional spatial morphology model are constructed. The three-dimensional pressure response model is reconstructed with the acquired pressure value as the first dimension, the sensor's spatial position as the second dimension, and the pulse signal amplitude as the third dimension. The three-dimensional spatial morphology model is reconstructed with the sensor's spatial position as the first dimension, the time sampling points within a single cycle as the second dimension, and the pulse signal amplitude as the third dimension. Based on the aforementioned three-dimensional pressure response model, pulse position characteristics are quantified; based on the aforementioned three-dimensional spatial morphology model, pulse number characteristics, pulse shape characteristics, and pulse momentum characteristics are quantified. Specifically, pulse shape characteristics are quantified by the effective amplitude spatial distribution width, pulse momentum characteristics are quantified by the combination of three-dimensional surface volume and maximum amplitude, and pulse waveform categories are identified through a neural network model to quantify pulse state characteristics. Pulse state characteristics refer to the waveform shape categories of pulse waves as they change over time, and these categories include at least wiry pulse, slippery pulse, and rough pulse. A five-dimensional feature set is constructed based on the five-dimensional features of pulse location, pulse number, pulse shape, pulse momentum, and pulse state. Logical contradictions in the five-dimensional feature set are eliminated, and the final comprehensive pulse judgment result is output.

2. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 1, characterized in that, The generation of personalized stress collection intervals adapted to the test subject includes: The individual characteristic parameters of the test subject and the current environmental information are normalized; wherein, the individual characteristic parameters include at least age, gender and body mass index; and the current environmental information includes at least ambient temperature, ambient air pressure and seasonal information. Based on the normalized individual characteristic parameters, a baseline stress range is constructed using a linear regression model, and the upper limit of the baseline stress range is determined. and lower limit : In the formula, Normalized Body Mass Index; Normalized age value; Assign a value to gender; , , , , , , , These are the weighted coefficients obtained by fitting clinical data; The baseline pressure range is corrected using the current environmental information to obtain the personalized pressure collection range. The correction formula is as follows: In the formula, , These are the lower and upper limits of the personalized pressure acquisition range, respectively; , , These are temperature, air pressure, and seasonal correction factors, respectively. , , These are the corresponding correction weights.

3. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 2, characterized in that, The personalized pressure acquisition range fully covers the three phases of floating, mid-water, and sinking, specifically including: Based on the amplitude of the pressure response signal fed back when the array sensor first contacts the skin, the starting pressure value of the personalized pressure acquisition range is adjusted; if the amplitude of the pressure response signal at the first contact exceeds a preset threshold, the current contact pressure is set as the acquisition start pressure, and the acquisition termination pressure remains unchanged. The adjusted pressure range is divided into three equal parts: the floating pressure range, the mid-water pressure range, and the sinking pressure range.

4. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 1, characterized in that, The process of performing multi-level pressure scans within the personalized pressure acquisition range to obtain array-type pulse response data under multi-level pressure gradients includes: A pressure feedback closed-loop control method is adopted to continuously monitor the real-time feedback values ​​of the array sensors and calculate the difference between the real-time feedback value and the current target pressure. ; When the difference When the pressure exceeds the first preset threshold, the motor is controlled to run at the first pressurization rate to quickly approach the target pressure. When the difference When the pressure is less than or equal to the first preset threshold, the motor is controlled to run at the second pressurization rate to enter a low-speed fine-tuning state; the second pressurization rate is less than the first pressurization rate. When the pressure change is detected to be stabilizing, pulse signal acquisition is triggered. After collecting pulse data for a preset duration at the current pressure level, the system automatically switches to the next target pressure level.

5. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 1, characterized in that, Constructing the three-dimensional model of the pressure response specifically includes: Peak value detection is performed on the single-cycle pulse waveforms collected by each sensor channel under each pressure level to locate the time point corresponding to the main peak. The pulse signal amplitudes of all sensor channels at the main peak time point are extracted to form a spatial discrete data set. ,in, For the first Level of pressure value acquisition, For the first Spatial location identifier of each sensor This corresponds to the pulse amplitude under different pressures and locations; The spatial discrete data set is smoothed using an interpolation method to generate a three-dimensional pressure response map with pressure value as the first dimension, sensor spatial location as the second dimension, and pulse amplitude as the third dimension.

6. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 1, characterized in that, Constructing the three-dimensional model of the spatial morphology specifically includes: Select a continuous array of sensors covering the pulse pulsation area Single-cycle pulse waveform data from each sensor channel, of which ; As stated The spatial arrangement of the sensors is based on the first dimension coordinate. Using the normalized time sampling points within a single period as the second dimension coordinate The pulse amplitude of each sensor at each time point The third dimension coordinates form a discrete set of points; The discrete point set is reconstructed by bicubic spline interpolation to obtain a three-dimensional spatial morphology map.

7. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 1, characterized in that, The quantified pulse position features specifically include: In the three-dimensional pressure response model, the point of maximum amplitude is searched using a global extremum search algorithm; the pressure value corresponding to this point is the optimal pulse-taking pressure. ; Iterating through the optimal pulse sampling pressure The amplitude response of all sensor channels is calculated. If the amplitude response curves of all channels are within... neighborhood interval If the internal peak characteristics are present, then it is confirmed. To achieve the optimal pulse pressure value, where To collect pressure step values; The optimal pulse-taking pressure By comparing the pulse position with the pressure ranges of the floating, middle, and sinking pulses calculated through pressure range mapping, the category and depth of the pulse position can be determined.

8. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 1, characterized in that, The quantified pulse shape features specifically include: In the aforementioned three-dimensional spatial model, for each sampling moment within a single period Extract the spatial distribution range where the amplitude reaches a preset proportion of the maximum amplitude at that moment, and calculate the spatial length of this range as the effective spatial width at a single moment. ; Effective spatial width at a single moment for all sampling moments within a single period Calculate the mean value as the pulse space width. It satisfies: In the formula, This represents the total number of sampling points for a single-cycle signal. The pulse space width It is compared with a preset pulse width threshold and mapped to the corresponding qualitative description of the pulse shape.

9. The intelligent pulse diagnosis method in Traditional Chinese Medicine according to claim 1, characterized in that, The quantified pulse characteristics specifically include: Extract the maximum amplitude of all data points in the three-dimensional spatial model. ; The volume of the three-dimensional model of the spatial morphology was calculated using the integral method. ; For the maximum amplitude With the volume Perform normalized weighted combination to calculate the pulse index. It satisfies: In the formula, and These are the amplitude reference value and the volume reference value, respectively. and As a weighting factor; According to the pulse index Determine the strength level of the pulse.

10. The intelligent pulse diagnosis method in Traditional Chinese Medicine according to claim 1, characterized in that, The logical contradiction removal of the five-dimensional feature set includes: After traversing the five-dimensional feature mapping, the feature combinations obtained are used to check whether there are any opposing feature pairs that do not conform to the rules of traditional Chinese medicine pulse diagnosis. ; If the aforementioned pair of opposing features exists, then compare the features. and Each feature has a confidence level, and features with higher confidence levels are retained while features with lower confidence levels are removed to obtain a set of candidate pulse patterns. The confidence level is calculated by the degree of fit between the quantified feature value and the center of its respective judgment interval.

11. The intelligent pulse diagnosis method in Traditional Chinese Medicine according to claim 10, characterized in that, The final pulse diagnosis result output specifically includes: The candidate pulse pattern set is matched with the clinical standard pulse pattern database, and the similarity score between the candidate pulse pattern set and each standard pulse pattern in the clinical standard pulse pattern database is calculated, which satisfies: In the formula, The first candidate pulse Dimensional features, The first standard pulse Dimensional features, For the first Preset weights for dimensional features This is a similarity calculation function; The standard pulse pattern with the highest similarity score is selected as the final comprehensive discrimination result.

12. The intelligent pulse diagnosis method for traditional Chinese medicine according to claim 1, characterized in that, The periodic segmentation of the array-type pulse response data includes: Wavelet transform is used to decompose the original pulse signal into multiple layers to remove approximate components representing baseline drift and detail components representing high-frequency noise. The periodic low-frequency fluctuations caused by respiration were simulated and eliminated using cubic spline interpolation. The 3σ criterion is used to detect abnormal amplitude points in the signal, and the detected abnormal points are corrected by linear interpolation. The continuous pulse signal is segmented into single cycles based on the valley point detection algorithm, and the segmented single-cycle waveform is resampled and normalized to ensure that each cycle signal has a uniform sampling length.

13. A traditional Chinese medicine pulse diagnosis intelligent system, characterized in that, include: The pulse acquisition probe contains multiple array-type pressure sensing units, which are used to simultaneously acquire multi-channel pulse pressure signals at the radial artery cun, guan, and chi positions. The pressure control and drive module is used to receive control commands, drive the pulse acquisition probe to perform step-by-step pressurization actions, and provide real-time feedback of the current pressure value. The data processing and three-dimensional reconstruction module is communicatively connected to the pulse acquisition probe and is used to perform the pressure response three-dimensional model and spatial morphology three-dimensional model construction steps in the method as described in any one of claims 1-12, as well as the five-dimensional feature quantification extraction steps of pulse position, pulse number, pulse shape, pulse momentum, and pulse state. The intelligent diagnostic decision module is communicatively connected to the data processing and three-dimensional reconstruction module. The intelligent diagnostic decision module has a built-in opposing pulse constraint library and a clinical standard pulse library, and is used to perform the pulse comprehensive discrimination step in the method as described in any one of claims 1-12. as well as The human-computer interaction and display module is used to input the individual characteristic parameters of the test subject and environmental information, and to display the pulse analysis results and three-dimensional reconstruction map.

14. The intelligent pulse diagnosis system for traditional Chinese medicine according to claim 13, characterized in that, The pressure sensing units are arranged in a three-row uniform array, with three pressure sensing units in the first row, six pressure sensing units in the second row, and three pressure sensing units in the third row.

15. The intelligent pulse diagnosis system for traditional Chinese medicine according to claim 13, characterized in that, It also includes a standard biomechanical pulse wave verification device, which comprises: Servo drive module, used to generate periodic motion with controllable displacement; A pressure output head, connected to the servo drive module, is used to apply dynamic pressure with a set amplitude and frequency to the pressure sensing unit to simulate pulse signals with different pulsation intensities and pulsation rates. The displacement sensing unit is used to monitor the output displacement of the pressure output head in real time.

16. An electronic 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 program, it implements the steps of the method as described in any one of claims 1-12.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-12.

Images