A hierarchical progressive pulse condition recognition method fusing traditional chinese medicine pulse-taking principles

Abstract

The application discloses a hierarchical progressive pulse condition identification method fusing traditional Chinese medicine pulse taking guidelines, and belongs to the technical field of intelligent Chinese medicine diagnosis and treatment. The method comprises the following steps: firstly, controlling a full-automatic pressure module to perform independent gradient hierarchical pressure on a radial artery of a patient and collect waveform signals; then, extracting quantitative characteristic parameters from the waveform signals, performing hierarchical determination based on three guidelines of "floating-sinking, delay number and deficiency and excess", respectively performing attribute determination of pulse position, pulse rate and pulse force dimensions, and outputting a basic guideline pulse label combination; finally, strictly following a diagnosis sequence of "first determining pulse position, then determining pulse rate, and finally determining pulse force and pulse shape" of traditional Chinese medicine, inputting a multi-dimensional characteristic matrix containing pulse shape and rhythm characteristic parameters into a big data model, performing hierarchical progressive reasoning in combination with the guideline label combination of the first stage, and identifying and outputting diagnosis results of 28 standard traditional Chinese medicine pulse conditions or combined pulses. The accuracy, repeatability and medical interpretability of automatic diagnosis are greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of TCM intelligent diagnosis and treatment technology, specifically involving a graded and progressive pulse recognition method that integrates the principles of traditional Chinese medicine pulse diagnosis. Background Technology

[0002] As the core of the four diagnostic methods of traditional Chinese medicine ("inspection, auscultation, inquiry, and palpation"), pulse diagnosis relies heavily on high-precision pulse recognition algorithms for its objectification and digital transformation. In traditional clinical practice, TCM practitioners strictly follow the diagnostic logic of "grasping the essentials and understanding the details," that is, by establishing a progressive diagnostic sequence of "first identifying the pulse location, then the pulse rate, and finally the pulse strength and shape," they first establish the macroscopic essential attributes, and then further refine and screen out the specific standard pulse and combined pulses.

[0003] However, existing intelligent pulse diagnosis and recognition methods generally suffer from the following insurmountable technical defects in their software algorithm architecture: First, existing pulse recognition algorithms typically employ a single deep learning model for end-to-end direct classification. This design completely ignores the hierarchical guidance of traditional Chinese medicine's pulse diagnosis framework, attempting to directly map the collected pulse wave time-domain data into 28 complex pulse patterns. Because the 28 standard pulse patterns and concurrent pulses in traditional Chinese medicine are intertwined and overlapping in terms of time-domain shape and rhythm characteristics, the lack of a top-level framework for filtering and differentiation makes the model highly susceptible to feature confusion when faced with complex concurrent pulses, leading to a high misdiagnosis rate. Furthermore, the classification results completely lack medical interpretability and fail to meet the requirements of high-fidelity, standardized clinical intelligent diagnosis at the medical level.

[0004] Secondly, traditional intelligent diagnostic systems generally lack top-level qualitative identification of the core pulse strength dimension of "deficiency and excess." In the TCM syllabus, "floating and sinking" determines pulse location, "slow and rapid" determines pulse rate, and "deficiency and excess" determines pulse strength; these three together constitute the top-level foundation of pulse diagnosis. Existing algorithms either only roughly focus on pulse location and pulse rate, or extract pulse strength characteristics along with complex pulse shape and rhythm features at the same level. This results in the algorithm's inability to output a macroscopic outline label combination composed of "floating, sinking, slow, rapid, deficient, and excess" in the first stage. Due to the lack of outline-level pre-filtering for the "deficiency and excess" dimension, the subsequent refined screening branches for the 28 standard pulses and concurrent pulses cannot be accurately and dynamically activated. This makes the entire system extremely vulnerable to noise and feature decoupling when dealing with physiological pulse strength changes caused by patient differences in constitution and emotional fluctuations, severely restricting the accuracy and repeatability of intelligent pulse diagnosis systems in clinical dataset collection and automated diagnosis. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing a hierarchical and progressive pulse identification method that integrates the principles of traditional Chinese medicine pulse diagnosis. This new method improves the accuracy and repeatability of clinical dataset collection and automated diagnosis.

[0006] To achieve the above effects, this invention discloses a graded and progressive pulse identification method that integrates the principles of traditional Chinese medicine pulse diagnosis, comprising the following steps: S1: Data acquisition: Independent gradient graded pressure is applied to the three parts of the radial artery of the patient (cun, guan, chi), and the raw pulse wave signals of the three parts are acquired and converted into waveform signals; S2: Three-category classification and recognition: Extract quantitative feature parameters from waveform signals, and perform hierarchical judgment based on the three categories of "floating and sinking - slow and rapid - real and fickle". Perform floating and sinking attribute judgment in the pulse position dimension, slow and rapid attribute judgment in the pulse rate dimension, and real and fickle attribute judgment in the pulse force dimension respectively, and output a basic category pulse label combination composed of "floating, sinking, slow, rapid, fickle and fickle". S3: Graded and progressive standard pulse and combined pulse identification: Strictly following the TCM diagnostic sequence of "first identifying pulse location, then pulse rate, and finally pulse strength and shape", a multi-dimensional feature matrix including pulse shape feature parameters and rhythm feature parameters is input into the big data model. Through graded and progressive reasoning, the model finally identifies and outputs diagnostic results of 28 standard TCM pulse patterns or combined pulses.

[0007] Furthermore, in S1, the gradient-graded pressure strictly simulates the three-stage pulse-taking technique of traditional Chinese medicine, with specific pressure standards as follows: floating pulse: 15 mmHg; middle pulse: 50-60 mmHg; deep pulse: 90-120 mmHg.

[0008] Furthermore, in S2, the quantized feature parameters comprehensively cover all three major dimensions, specifically including: Pulse position characteristic parameters: floating and sinking coefficient, floating amplitude, middle amplitude, and sinking amplitude; Pulse rate characteristic parameters: mean pulse rate, pulse rate coefficient of variation, rhythm regularity; Pulse strength characteristic parameters: virtual and real coefficient, maximum pulse strength, minimum pulse strength, pulse strength variation coefficient.

[0009] Furthermore, the rules for performing hierarchical determination based on the characteristic parameters of the three major outlines are as follows: If the buoyancy coefficient meets the preset shallow threshold and the floating amplitude is greater than the sinking amplitude, then it is determined that the basic vein label contains "floating vein". If the buoyancy coefficient meets the preset depth threshold and the sinking amplitude is greater than the floating amplitude, then it is determined that the basic vein label contains "sinking vein". If the average pulse rate is lower than the preset slow pulse threshold, then the basic pulse label is determined to contain "slow pulse". If the average pulse rate is higher than the preset pulse frequency threshold, then the basic pulse label is determined to contain "pulse". If the virtual-to-real coefficient is greater than the preset filling threshold and the maximum pulse force is higher than the rigidity limit, then it is determined that the basic pulse label contains "real pulse". If the virtual-real coefficient is lower than the preset dispersion threshold and the minimum pulse force is lower than the flexibility limit, then the basic pulse label is determined to contain "virtual pulse".

[0010] Furthermore, the multidimensional feature matrix input in S3 specifically includes: Pulse characteristic parameters: rise time, fall time, dicrotic wave height, pulse width, and pulse length; Rhythmic characteristic parameters: number of intervals, interval duration, and interval regularity.

[0011] Furthermore, the process of identifying the graded and progressive standard pulse and concurrent pulse is as follows: the big data model, based on the "floating-sinking-slow-number-real" outline label combination output in the first stage, performs time-frequency joint reasoning on the pulse shape feature parameters and the rhythm feature parameters, and comprehensively determines the rhythm abnormal concurrent pulse, including knotted pulse, intermittent pulse, and rapid pulse, or the morphologically abnormal standard pulse, including slippery pulse, surging pulse, and soft pulse.

[0012] Compared with the prior art, the present invention has the following significant advantages: This invention innovatively constructs a hierarchical, progressive reasoning framework. By strictly adhering to the diagnostic sequence of Traditional Chinese Medicine (TCM) – "first identify the pulse location, then the pulse rate, and finally the pulse strength and shape" – it significantly improves the accuracy of identifying complex pulse patterns. It can automatically complete standardized data collection across the three pulse positions (cun, guan, chi) and the nine pulse types (superficial, middle, and deep), and classify 28 standard TCM pulse characteristics based on the three principles of "superficial / deep, slow / rapid, and weak / full" – and automatically generate corresponding pathological diagnostic reports. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0014] This embodiment discloses a hierarchical, progressive pulse identification method that integrates the principles of traditional Chinese medicine pulse diagnosis. Its core lies in changing the traditional end-to-end direct classification black-box mapping mode of deep learning models, transforming the "grasping the essentials and understanding the details" diagnostic logic of traditional Chinese medicine clinical practice into a multi-level distributed reasoning architecture. The specific implementation process is as follows: Step S1: Data Acquisition This step achieves fully automated, standardized pulse diagnosis through the coordination of hardware and underlying control algorithms. First, the fully automated pressurization module (such as a servo motor) drives a multimodal composite sensing unit integrated with a digital pressure core to perform independent gradient-level pressurization on the cun, guan, and chi positions of the patient's radial artery. To accurately simulate the traditional Chinese medicine pulse-taking techniques of "superficial, middle, and deep" pulse diagnosis, the system employs PID closed-loop control logic throughout the entire pressurization cycle. Floating stage (S11): The servo motor drives the sensing unit to lightly press on the skin. The algorithm automatically adjusts the pressurization speed based on the real-time pressure feedback from the sensor until the pressure reaches the preset floating standard of 15 mmHg. At this time, the algorithm controls the motor to automatically lock the pressure and maintain it for 30 seconds, while simultaneously acquiring the raw pulse wave signals from the cun, guan, and chi positions.

[0015] Mid-stage (S12): After the locking pressure is completed, the motor continues to press down to the muscle layer. When the pressure feedback loop reaches the mid-stage standard of 50-60 mmHg, the pressure is automatically locked and stabilized for 30 seconds, and the pulse wave signal in this state is continuously collected.

[0016] Sinking stage (S13): The motor finally presses down to the rib layer. When the pressure feedback loop reaches the sinking standard of 90-120 mmHg, it automatically locks the pressure and stabilizes for 30 seconds to complete the last round of data acquisition. Then the entire system is depressurized.

[0017] After acquisition, the decoupling algorithm converts the raw pulse wave signal into a standard waveform signal. Specifically, it decouples the signal into a static pressure baseline signal reflecting macroscopic pressure and a dynamic vibration waveform signal reflecting vascular volume fluctuations.

[0018] Step S2: Three-category classification and identification In order to provide a powerful pre-filtering mechanism for the subsequent 28 standard pulses, this step extracts the pre-quantitative feature parameters from the decoupled signal and inputs them into the preset pulse expert decision system, which performs a graded judgment based on the three categories of "floating and sinking (pulse position) - slow and rapid (pulse rate) - weak and strong (pulse strength)".

[0019] Specifically, the top-level outline quantization feature parameters extracted from the waveform signal comprehensively cover the following three dimensions: Pulse position characteristic parameters: floating and sinking coefficient, floating amplitude, middle amplitude, and sinking amplitude; Pulse rate characteristic parameters: mean pulse rate, pulse rate coefficient of variation, rhythm regularity; Pulse strength characteristic parameters: virtual and real coefficient, maximum pulse strength, minimum pulse strength, pulse strength variation coefficient.

[0020] The expert decision-making system for the framework strictly adheres to the following hierarchical judgment rules based on the aforementioned three framework characteristic parameters: Pulse position (superficial / deep) attribute determination: Check the superficial / deep coefficient. If it meets the preset superficial threshold and the superficial amplitude is greater than the deep amplitude, then the basic pulse label is determined to contain "superficial pulse". Conversely, if the superficial / deep coefficient meets the preset deep threshold and the deep amplitude is greater than the superficial amplitude, then the basic pulse label is determined to contain "deep pulse".

[0021] Pulse rate (slow pulse) attribute determination: Check the average pulse rate. If it is lower than the preset slow pulse threshold, the basic pulse label is determined to contain "slow pulse"; if it is higher than the preset pulse frequency threshold, the basic pulse label is determined to contain "fast pulse".

[0022] Pulse strength (real / virtual) attribute determination: Check the real / virtual coefficient. If it is greater than the preset fullness threshold and the maximum pulse strength is higher than the preset rigidity upper limit, then the basic pulse label is determined to contain "real pulse". If the real / virtual coefficient is lower than the preset dispersion threshold and the minimum pulse strength is lower than the preset flexibility lower limit, then the basic pulse label is determined to contain "virtual pulse".

[0023] Finally, step S2 integrates the judgment results of the above three dimensions and outputs a specific basic framework label combination consisting of "floating, sinking, slow, number, virtual, and real".

[0024] Step S3: Graded progressive identification of standard pulse and concurrent pulse Guided by the outline labels in the first stage, this step conducts a refined screening of 28 complex pulse patterns. The algorithm control layer first dynamically activates the corresponding target sub-network from a preset model library or adjusts the weights of the conditional attention mechanism in the big data model based on the basic pulse label combination output in step S2, thus instantaneously splitting the vast 28-pulse search space.

[0025] At this point, the system continues to extract the remaining microscopic details from the waveform signal to construct a multidimensional feature matrix for input into the big data model. This matrix specifically includes: Pulse characteristic parameters: rise time, fall time, dicrotic wave height, pulse width, pulse length; Rhythmic characteristic parameters: number of intervals, interval duration, and interval regularity.

[0026] In the specific identification and reasoning process, the big data model strictly follows the diagnostic sequence of traditional Chinese medicine: "first identify the pulse location, then the pulse rate, and finally the pulse strength and shape." The model uses the "floating / sinking - slow / rapid - weak / full" outline label combination output from step S2 as a strong constraint boundary or prior condition, guiding the deep network of the model to focus on subtle differences in microscopic shape and rhythm. Identification of Abnormal Pulse Morphology: When the outline label combination is determined to be "floating-solid", the big data model combines the rise time, pulse width and dicrotic wave height in the pulse shape feature parameters to perform time-frequency joint reasoning, which can accurately determine the "surging pulse" or "slippery pulse" that belongs to the abnormal morphology, without having to perform invalid search in the solution space of "deep pulse system (such as weak pulse, firm pulse)".

[0027] Rhythm Abnormality and Pulse Identification: The big data model performs deep time series modeling based on the number of intervals, interval duration, and interval regularity in the rhythm feature parameters. If the outline label contains "slow pulse" and the detected number of intervals is high and the interval duration is regular, the progressive determination output is "intermittent pulse" among the concurrent pulses; if the outline label contains "frequent pulse" and the intervals are irregular, the determination output is "rapid pulse"; if the outline label contains "slow pulse" and the intervals are irregular, the determination output is "knotted pulse".

[0028] Through the above-mentioned hierarchical and progressive multi-label conditional reasoning, the big data model can accurately identify and automatically output the diagnostic results of 28 standard TCM pulse patterns or specific combined pulse patterns. Based on this, it can link with the backend knowledge base to automatically generate corresponding digital pathological diagnosis reports.

[0029] The embodiments described above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A hierarchical, progressive pulse diagnosis method integrating traditional Chinese medicine pulse diagnosis principles, characterized in that... Includes the following steps: S1: Data acquisition: Independent gradient graded pressure is applied to the three parts of the radial artery of the patient (cun, guan, chi), and the raw pulse wave signals of the three parts are acquired and converted into waveform signals; S2: Three-category classification and recognition: Extract quantitative feature parameters from waveform signals, and perform hierarchical judgment based on the three categories of "floating and sinking - slow and rapid - real and false". Perform floating and sinking attribute judgment in the pulse position dimension, slow and rapid attribute judgment in the pulse rate dimension, and real and false attribute judgment in the pulse force dimension respectively, and output the basic category pulse label combination composed of "floating, sinking, slow, rapid, false and true". S3: Graded and progressive standard pulse and combined pulse identification: Strictly following the TCM diagnostic sequence of "first identifying pulse location, then pulse rate, and finally pulse strength and shape", a multi-dimensional feature matrix including pulse shape feature parameters and rhythm feature parameters is input into the big data model. Through graded and progressive reasoning, the model finally identifies and outputs diagnostic results of 28 standard TCM pulse patterns or combined pulses.

2. The graded and progressive pulse identification method integrating traditional Chinese medicine pulse diagnosis principles as described in claim 1, characterized in that: In S1, the gradient-graded pressure strictly simulates the three pulse-taking techniques of traditional Chinese medicine: superficial, middle, and deep. The specific pressure standards are: superficial: 15 mmHg; middle: 50-60 mmHg; deep: 90-120 mmHg.

3. The graded and progressive pulse identification method integrating traditional Chinese medicine pulse diagnosis principles as described in claim 1, characterized in that: In S2, the quantized feature parameters comprehensively cover all three major dimensions, specifically including: Pulse position characteristic parameters: floating and sinking coefficient, floating amplitude, middle amplitude, and sinking amplitude; Pulse rate characteristic parameters: mean pulse rate, pulse rate coefficient of variation, rhythm regularity; Pulse strength characteristic parameters: virtual and real coefficient, maximum pulse strength, minimum pulse strength, pulse strength variation coefficient.

4. The graded and progressive pulse identification method integrating traditional Chinese medicine pulse diagnosis principles as described in claim 3, characterized in that: The rules for performing hierarchical determination based on the characteristic parameters of the three major outlines are as follows: If the buoyancy coefficient meets the preset shallow threshold and the floating amplitude is greater than the sinking amplitude, then it is determined that the basic vein label contains "floating vein". If the buoyancy coefficient meets the preset depth threshold and the sinking amplitude is greater than the floating amplitude, then it is determined that the basic vein label contains "sinking vein". If the average pulse rate is lower than the preset slow pulse threshold, then the basic pulse label is determined to contain "slow pulse". If the average pulse rate is higher than the preset pulse frequency threshold, then the basic pulse label is determined to contain "pulse". If the virtual-to-real coefficient is greater than the preset filling threshold and the maximum pulse force is higher than the rigidity limit, then it is determined that the basic pulse label contains "real pulse". If the virtual-real coefficient is lower than the preset dispersion threshold and the minimum pulse force is lower than the flexibility limit, then the basic pulse label is determined to contain "virtual pulse".

5. The graded and progressive pulse identification method integrating traditional Chinese medicine pulse diagnosis principles as described in claim 1, characterized in that: The multidimensional feature matrix input in S3 specifically includes: Pulse characteristic parameters: rise time, fall time, dicrotic wave height, pulse width, and pulse length; Rhythmic characteristic parameters: number of intervals, interval duration, and interval regularity.

6. The graded and progressive pulse identification method integrating traditional Chinese medicine pulse diagnosis principles as described in claim 5, characterized in that, The process of identifying the graded and progressive standard pulse and concurrent pulse is as follows: The big data model, based on the "floating and sinking - slow and rapid - real and false" outline label combination output in the first stage, performs time-frequency joint reasoning on the pulse shape feature parameters and the rhythm feature parameters, and comprehensively determines the rhythm abnormal concurrent pulse, including knotted pulse, intermittent pulse, and rapid pulse, or the morphologically abnormal standard pulse, including slippery pulse, surging pulse, and soft pulse.