An arteriovenous fistula home monitoring method and system based on audio analysis

Abstract

The application provides an arteriovenous fistula home monitoring method and system based on audio analysis, and relates to the technical field of vital sign monitoring. The method comprises the following steps: collecting an analog voltage waveform; converting the analog voltage waveform into discrete voltage values and arranging the discrete voltage values in chronological order to obtain an audio vibration waveform sequence; constructing a frequency energy distribution map through Fourier transform, and constructing a harmonic energy set based on the frequency energy distribution map; fitting the attenuation slope of adjacent order energy peaks in the ascending direction of order, extracting the overall harmonic attenuation trend of the fistula audio signal, and generating a harmonic attenuation characteristic parameter; discretely extracting the fistula blood flow state change in the continuous time period according to the sampling time sequence to generate a blood flow time sequence fluctuation parameter; and classifying and determining the blood flow time sequence fluctuation parameter through a decision tree classification model to generate an arteriovenous fistula monitoring result.

Description

Technical Field

[0001] This invention relates to the field of vital sign monitoring technology, and in particular to a home monitoring method and system for arteriovenous fistulas based on audio analysis. Background Technology

[0002] Arteriovenous fistulas (AVFs) are crucial vascular access points for hemodialysis in patients with end-stage renal disease, and their patency directly determines dialysis effectiveness and patient quality of life. Due to repeated punctures and hemodynamic changes, AVFs are highly susceptible to intimal proliferative stenosis. If not detected and intervened in time, this can progress to thrombosis or vascular occlusion, rendering dialysis impossible and requiring emergency surgical reconstruction. Therefore, continuous and objective monitoring of AVF function is of great significance for early detection of stenosis, prevention of access failure, and reduction of hospitalization rates and medical costs.

[0003] Currently, the routine clinical assessment of arteriovenous fistula function mainly relies on patients visiting the hospital regularly, where professional medical staff use Doppler ultrasound to scan the fistula vessels and directly determine whether stenosis exists by measuring parameters such as blood flow velocity, blood flow rate, and vessel wall morphology. Alternatively, medical staff use a stethoscope attached to the skin surface of the fistula site to perform qualitative analysis of the intensity, pitch, and continuity of blood flow murmurs based on subjective auditory experience, while also using palpation to perceive changes in the strength of the fistula thrill, thereby comprehensively assessing the patency of blood flow in the fistula.

[0004] However, current technologies heavily rely on the on-site operation and subjective judgment of professional medical staff. Patients need to visit medical institutions regularly and cannot conduct continuous fistula function monitoring independently at home, resulting in long monitoring cycles and difficulty in capturing the dynamic changes in the fistula's condition. Furthermore, both ultrasound detection and auscultation interpretation depend on the operator's skill level and subjective auditory identification ability, lacking standardized objective quantitative indicators. This can easily lead to missed diagnoses of early stenosis or biased assessments of the condition, failing to meet patients' actual needs for real-time, objective, and continuous fistula risk assessment and self-management in everyday home settings. Summary of the Invention

[0005] To address the limitations of existing technologies that heavily rely on on-site operations and subjective judgment by professional medical personnel, requiring patients to regularly visit medical institutions and hindering their ability to conduct continuous arteriovenous fistula (AVF) function monitoring at home, resulting in long monitoring cycles and difficulty in capturing dynamic changes in AVF status. Furthermore, both ultrasound detection and auscultation interpretation depend on the operator's skill level and subjective auditory perception, lacking standardized objective quantitative indicators. This can easily lead to missed diagnoses of early stenosis or biased disease assessments, failing to meet patients' practical needs for real-time, objective, and continuous AVF risk assessment and self-management in everyday home settings. Therefore, this invention provides a home monitoring method and system for arteriovenous fistulas based on audio analysis.

[0006] The technical solutions provided by the embodiments of the present invention are as follows: A first aspect of this invention provides a home monitoring method for arteriovenous fistulas based on audio analysis, comprising: S1: Acquire analog voltage waveform; S2: Combined with the preset conversion frequency, the analog voltage waveform is converted into discrete voltage values, and the discrete voltage values ​​are arranged in chronological order to obtain an audio vibration waveform sequence. S3: Based on the audio vibration waveform sequence, a frequency energy distribution map is constructed through Fourier transform, and a harmonic energy set is constructed based on the frequency energy distribution map; S4: Based on the harmonic energy set, the attenuation slope of adjacent order energy peaks is fitted according to the order rising direction, and the overall harmonic attenuation trend of the arteriovenous fistula audio signal is extracted to generate harmonic attenuation characteristic parameters. S5: Based on the harmonic attenuation characteristic parameters, the discreteness of the changes in the blood flow state of the arteriovenous fistula in a continuous time period is extracted according to the sampling time sequence to generate blood flow temporal fluctuation parameters. S6: Combining preset fluctuation standard thresholds, the blood flow temporal fluctuation parameters are classified and judged through a decision tree classification model to generate arteriovenous fistula monitoring results.

[0007] A second aspect of the present invention provides a home monitoring system for arteriovenous fistulas based on audio analysis, comprising: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the home monitoring method for arteriovenous fistulas based on audio analysis as described in the first aspect.

[0008] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the home monitoring method for arteriovenous fistula based on audio analysis as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this invention, by acquiring the vibration waveform attached to the surface of the arteriovenous fistula and performing discrete transformation, the audio vibration signal is segmented and mapped to the frequency dimension for analysis. Based on the extraction of the reference frequency and energy peak features, the overall harmonic attenuation trend is fitted along the order direction. Furthermore, the discreteness of blood flow state changes within a continuous time period is captured by combining the time axis. Thus, the abstract dynamics of the arteriovenous fistula are transformed into specific fluctuation parameters. Combined with a classification model, an objective judgment is made and the stenosis risk result is output. This invention breaks free from the constraints of relying on on-site operation and subjective experience judgment by medical staff, and overcomes the shortcomings of patients not being able to conduct continuous monitoring independently at home. It realizes objective, quantitative, and continuous self-testing of arteriovenous fistula function and risk screening. Attached Figure Description

[0010] 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.

[0011] Figure 1 This is a flowchart illustrating a home monitoring method for arteriovenous fistulas based on audio analysis, provided as an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram of a home monitoring system for arteriovenous fistulas based on audio analysis, provided as an embodiment of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0014] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0015] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0016] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0017] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0018] Reference manual attached Figure 1 The diagram shows a flowchart of a home monitoring method for arteriovenous fistula based on audio analysis provided by an embodiment of the present invention.

[0019] This invention provides a home monitoring method for arteriovenous fistulas based on audio analysis. This method can be implemented using a home monitoring device for arteriovenous fistulas based on audio analysis, which can be a terminal or a server. The processing flow of the home monitoring method for arteriovenous fistulas based on audio analysis may include the following steps: S1: Acquire analog voltage waveform.

[0020] S2: Combined with the preset conversion frequency, the analog voltage waveform is converted into discrete voltage values, and the discrete voltage values ​​are arranged in chronological order to obtain an audio vibration waveform sequence.

[0021] In one possible implementation, S2 specifically includes sub-steps S201 to S204: S201: Combined with the preset frequency band range, perform bandpass frequency domain filtering on the frequency band components within the analog voltage waveform, remove frequency band components that exceed the preset frequency band range, and generate the target waveform segment.

[0022] Optionally, the preset frequency band is specifically defined as the filtering range determined based on the mechanical vibration characteristics of blood turbulence.

[0023] It should be noted that those skilled in the art can set the preset conversion frequency according to actual needs, and this invention does not limit this.

[0024] Among them, the mechanical vibration characteristics of blood turbulence refer to the physical characteristics of local mechanical vibration fluctuations generated when blood flows through the narrow lesion in the lumen of an arteriovenous fistula, due to the drastic changes in blood flow velocity and direction, transforming from an orderly laminar flow state to a turbulent and irregular flow state.

[0025] Specifically, a piezoelectric vibration pickup device is connected and activated via a physical cable, attached 3 cm proximal to the arteriovenous fistula anastomosis site in the forearm. This piezoelectric vibration pickup device integrates a high-sensitivity piezoelectric ceramic wafer and a charge amplification circuit. Its hardware sampling frequency is fixed at 4000 Hz. Using continuous direct memory access mode, it continuously reads analog voltage waveforms for 5 seconds from the hardware interface register. The read analog voltage waveform data format is a 64-bit double-precision floating-point array, with data units in millivolts. The system's non-volatile memory is then used to retrieve preset frequency band configuration parameters. The preset frequency band consists of two boundary parameters: a lower frequency band threshold and an upper frequency band threshold. The lower frequency band threshold is set to 20 Hz, and the upper frequency band threshold is set to 800 Hz. The data setting for this range is based on experimental statistical results of the core energy spectrum distribution characteristics of turbulent mechanical vibrations excited by blood flowing through the arteriovenous fistula stenosis.

[0026] It should be noted that those skilled in the art can set the upper and lower limit thresholds of the frequency band according to actual needs, and this invention does not impose any limitations on them.

[0027] Specifically, a finite impulse response bandpass filter architecture is established in memory, containing 256 filter orders. 257 bandpass tap coefficients are retrieved from memory, and the bandpass tap coefficient sequence is subjected to discrete convolution with the input analog voltage waveform array. The specific calculation process is as follows: extract the waveform node voltage value at the current moment, multiply the waveform node voltage value with the corresponding bandpass tap coefficient to obtain the single-order product value, and then perform a summation operation on all 257 single-order product values ​​within the sliding window at the current moment to calculate the filtered center node voltage value. In practical computing scenarios, when the voltage values ​​of three consecutive nodes in a simulated voltage waveform array are 15 mV, 12 mV, and 10 mV, and the corresponding three bandpass tap coefficients are 0.5, 0.3, and 0.2, multiplying 15 mV by 0.5 yields the first product of 7.5 mV, multiplying 12 mV by 0.3 yields the second product of 3.6 mV, and multiplying 10 mV by 0.2 yields the third product of 2 mV. Then, the summation operation of 7.5 mV, 3.6 mV, and 2 mV is performed to calculate the voltage result of the local convolution center node as 13.1 mV. By traversing the entire simulated voltage waveform array point by point using an incrementing data pointer, the calculated center node voltage results are reassembled and combined according to the corresponding timestamp order and stored in a newly created 1D array memory space to generate the target waveform segment.

[0028] Furthermore, the audio acquisition module of the monitoring system is configured with two acquisition channels, with channel 1 and channel 2 having identical parameter settings. The sampling frequency of each channel is 4000 Hz, the gain coefficient is set to 2.5, and the lower limit of the bandpass filter is 20 Hz, and the upper limit is 800 Hz. The parameter configuration is determined based on the mechanical vibration characteristics of blood turbulence, aiming to effectively acquire the effective vibration signal within the 20 Hz to 800 Hz frequency band generated by arteriovenous fistula blood flow, while simultaneously eliminating interference noise outside this frequency range. All of the above parameters are used to standardize the physical extraction state of the front-end signal.

[0029] S202: Combined with the preset conversion frequency, the target waveform segment is sampled at equal intervals to extract the instantaneous voltage amplitude corresponding to the sampling node, and the instantaneous voltage amplitude is converted into a decimal digital value to generate a discrete voltage value.

[0030] Optionally, the preset conversion frequency is a fixed sampling frequency set according to the channel performance parameters.

[0031] Specifically, the target waveform segment is retrieved from a one-dimensional array memory space. The configuration file of the hardware analog-to-digital converter (ADC) controller is read, and the preset conversion frequency is extracted from it. This preset conversion frequency is set to a fixed 2000 Hz. This value is set according to the Nyquist sampling theorem, which states that the sampling frequency must be strictly greater than twice the upper limit threshold of the preset frequency band of 800 Hz. The internal clock divider is activated, and the time interval between adjacent sampling points is calculated to be 0.5 milliseconds based on the preset conversion frequency of 2000 Hz. The module controls the data interception pointer to perform equally spaced discrete sampling operations on the target waveform segment according to the time interval step of 0.5 milliseconds. In a single sampling action, the instantaneous voltage amplitude of the continuous waveform corresponding to the 0.5 millisecond time node is read, and the specific voltage floating-point value is extracted. Subsequently, the module reads the system-set analog-to-digital conversion reference voltage value of 3.3 volts and the quantization resolution of 16-bit unsigned integer format. The extracted instantaneous voltage amplitude is divided by the 3.3-volt analog-to-digital conversion reference voltage to obtain the voltage relative scaling factor. Then, the voltage relative scaling factor is multiplied by the maximum decimal value of 65535 in 16-bit unsigned integer format, and the result of the multiplication is truncated by rounding down to convert it into an absolute decimal number.

[0032] Furthermore, in a practical computing scenario, when the instantaneous voltage amplitude corresponding to a sampling node of the target waveform segment is 1.65 volts, and the analog-to-digital conversion reference voltage is 3.3 volts, with a maximum decimal value of 65535, the voltage quantization submodule divides 1.65 volts by 3.3 volts to calculate a voltage relative scaling factor of 0.5. Then, it multiplies 0.5 by 65535 to obtain 32767.5. The module performs a floor operation on 32767.5, ultimately converting it to a decimal number of 32767. This value is defined as the discrete voltage value at that node. The voltage quantization submodule continuously performs the above division, multiplication, and floor operations in 0.5 millisecond steps until all sampling nodes within the entire target waveform segment have been processed. The decimal numbers generated in each calculation are sequentially stored in the discrete data buffer, generating a data set consisting of a series of discrete voltage values.

[0033] S203: Read the timestamp tag corresponding to the discrete voltage value.

[0034] Specifically, connected to the discrete data buffer, based on all the discrete voltage values ​​stored in the buffer, the module reads the timestamp tag corresponding to the value synchronously appended to the end of the memory address of each discrete voltage value by the system real-time clock generator. The data format of the timestamp tag is a 64-bit integer absolute time constant containing year, month, day, hour, minute, second and millisecond.

[0035] S204: Based on the recorded timestamp tags, the discrete voltage values ​​are sorted in ascending order according to time sequence, and all the sorted discrete voltage values ​​are merged and recombined into a one-dimensional array structure to construct an audio vibration waveform sequence.

[0036] Specifically, an independent sorting workspace is created internally, and a quicksort logic architecture is used to perform position swapping operations on all discrete voltage values ​​containing timestamp tags. The specific calculation process is as follows: the module randomly selects the record timestamp tag of the first value in the discrete voltage value set as the benchmark comparison time node, and the module traverses all other discrete voltage values, performing a subtraction operation between each extracted record timestamp tag and the benchmark comparison time node.

[0037] Furthermore, when the time difference obtained from the subtraction operation is negative, the module determines that the discrete voltage value is earlier than the benchmark comparison time point in the time dimension, and moves its memory pointer to the left region of the workspace. When the time difference obtained from the subtraction operation is positive, the module determines that the discrete voltage value is later than the benchmark comparison time point in the time dimension, and moves its memory pointer to the right region of the workspace. The module recursively performs the above timestamp subtraction comparison and pointer movement operations within the left and right regions until the timestamp labels of all discrete voltage values ​​show a strict monotonically increasing state from small to large, completing the ascending order arrangement.

[0038] Furthermore, in a practical computing scenario, when there are three discrete voltage values ​​of 32767, 15000, and 45000, with corresponding timestamps of 500 milliseconds, 300 milliseconds, and 700 milliseconds respectively, the sequence construction submodule selects 500 milliseconds as the baseline comparison time node. The module calculates 300 milliseconds minus 500 milliseconds to obtain -200 milliseconds, determining that 15000 should be placed on the left. The module calculates 700 milliseconds minus 500 milliseconds to obtain +200 milliseconds, determining that 45000 should be placed on the right. The final ascending order result is 15000, 32767, and 45000. After the arrangement is completed, all discrete voltage values ​​are extracted sequentially, their timestamps are removed, and only the pure decimal values ​​are retained. Following the pattern of continuously increasing physical memory addresses, the decimal values ​​are merged and recombined into a new one-dimensional array structure. After encapsulating the start and end address pointers, an audio vibration waveform sequence is constructed.

[0039] In this embodiment of the invention, by performing bandpass frequency domain filtering based on the mechanical vibration characteristics of blood turbulence on the original analog voltage waveform, environmental and circuit noise outside the frequency band is effectively eliminated, significantly improving the signal-to-noise ratio of the arteriovenous fistula blood flow vibration signal. Simultaneously, by combining equally spaced discrete sampling and high-precision analog-to-digital conversion, the continuous analog waveform is quantized into high-resolution decimal discrete voltage values, and reconstructed into a one-dimensional array structure of audio vibration waveform sequences through timestamp sorting. This provides a standardized, high-fidelity time-domain data foundation for subsequent frequency domain analysis, ensuring the accuracy and reliability of arteriovenous fistula blood flow state feature extraction.

[0040] S3: Based on the audio vibration waveform sequence, a frequency energy distribution spectrum is constructed through Fourier transform, and a harmonic energy set is constructed based on the frequency energy distribution spectrum.

[0041] The Fourier transform is a mathematical analysis method that converts a signal from the time domain to the frequency domain. In this patent, it is applied to an audio vibration waveform sequence, decomposing it into combinations of different frequency components, thereby revealing the energy distribution of each frequency component in the original waveform.

[0042] Optionally, the audio vibration waveform sequence specifically includes: a time series index, a discrete amplitude sequence, and an envelope amplitude.

[0043] Optionally, the harmonic energy set specifically includes: harmonic frequency components, power spectral density, and high-frequency harmonic amplitude.

[0044] In one possible implementation, S3 specifically includes sub-steps S301 to S306: S301: Slice the audio vibration waveform sequence according to the preset length.

[0045] Specifically, the audio vibration waveform sequence is extracted, and a preset length parameter is read from the configuration table. This preset length parameter is set to 1024 sampling points. This value is set based on the underlying logic requirement of the Fast Fourier Transform algorithm that the input data length must be an integer power of 2. The frequency domain conversion submodule performs a segmentation operation on the long audio vibration waveform sequence according to the fixed span of 1024 sampling points. The module allocates an independent memory segment for each of the 1024 discrete voltage value sequences segmented, and obtains an independent waveform segment.

[0046] It should be noted that those skilled in the art can set the preset length according to actual needs, and this invention does not limit it.

[0047] S302: By using Fourier transform, frequency domain dimension mapping operation is performed on the segmented audio vibration waveform sequence to extract the frequency nodes corresponding to the frequency domain dimension and calculate the amplitude energy value corresponding to the frequency nodes.

[0048] Specifically, for each waveform segment, a Hamming window function is introduced for smoothing. The module performs term-by-term multiplication of the 1st to 1024th discrete voltage values ​​within the waveform segment with the 1024 weighting coefficients corresponding to the Hamming window function, calculating a smoothed waveform segment with suppressed edge spectral leakage. Subsequently, the Fast Fourier Transform butterfly operation core is activated internally, allocating the real node values ​​within the smoothed waveform segment to the real part register of the complex number operation architecture, and initializing all imaginary part registers to 0. Through alternating operations of multi-level complex multiplication and addition, the discrete voltage values ​​in the time domain are converted into a sequence of complex results in the frequency domain. For the complex result corresponding to each frequency node, the real and imaginary parts of the complex number are extracted. The module multiplies the real part value by itself to obtain the square of the real part, and multiplies the imaginary part value by itself to obtain the square of the imaginary part. Then, the square of the real and imaginary parts is summed, and finally, the square root of the sum is taken to calculate the amplitude energy value corresponding to this frequency node.

[0049] S303: Arrange the frequency nodes and amplitude energy values ​​in two dimensions to generate a frequency domain energy matrix.

[0050] Specifically, in a practical computing scenario, when a waveform segment undergoes a Fast Fourier Transform (FFT) and the complex result output at the 300 Hz frequency node has a real part of 4 and an imaginary part of 3, multiplying 4 by itself yields 16, and multiplying 3 by itself yields 9. Then, summing 16 and 9 yields 25. The module performs a square root operation on 25, ultimately calculating the amplitude energy value corresponding to the 300 Hz frequency node as 5. Using all frequency node values ​​within the range of 0 Hz to 1000 Hz as the row index scalar of the matrix, and the time sequence number of the waveform segment as the column index scalar, the corresponding amplitude energy value calculated by the square root operation is filled into the matrix cell where the row and column indices intersect. A 2D arrangement operation is then performed to generate a complete frequency domain energy matrix.

[0051] S304: Traverse the elements within the frequency domain energy matrix, compare and select the largest amplitude energy value, extract the frequency node scalar at the horizontal axis position corresponding to the largest amplitude energy value, and generate the fundamental frequency signal value.

[0052] Specifically, the frequency domain energy matrix is ​​locked, and an internal register is initialized to record the local highest energy, with its initial value set to an absolute value of 0. A matrix traversal pointer is configured to read the amplitude energy values ​​arranged within the frequency domain energy matrix sequentially, starting from row 1, column 1, from left to right and top to bottom. For each read operation, the module performs a subtraction comparison between the current amplitude energy value and the recorded value stored in the internal register. If the difference between the current amplitude energy value and the recorded value is greater than 0, the current amplitude energy value is considered larger, and the module performs a value overwrite operation, storing the current amplitude energy value in the internal register to replace the original value. Simultaneously, the module reads the specific value of the frequency node corresponding to the matrix row index where the current amplitude energy value is located and stores this specific value in the synchronously updated frequency register. If the difference between the current amplitude energy value and the recorded value is less than or equal to 0, the module does not perform any overwrite operation, and the pointer continues to move to the next matrix element.

[0053] Furthermore, after the fundamental frequency analysis submodule iterates through all elements within the entire frequency domain energy matrix using the aforementioned subtraction comparison and conditional overwrite operations, the final value remaining in the internal register is the selected maximum amplitude energy value. The module directly reads the frequency node scalar bound to this maximum amplitude energy value from the synchronously updated frequency register and defines this frequency node scalar as the core periodic frequency of the arteriovenous fistula blood thrill. In actual computational scenarios, when there are three significant amplitude energy values ​​of 150, 500, and 200 within the elements of the frequency domain energy matrix, corresponding to frequency nodes of 50 Hz, 100 Hz, and 150 Hz respectively, and the initial value recorded in the internal register is 0, the fundamental frequency analysis submodule first reads 150 and subtracts 150 from 0. If the difference is positive, the internal register is updated to 150, and the frequency register is updated to 50 Hz. Next, the module reads 500 and subtracts it from 150. If the difference is positive, the internal register is overwritten with 500, and the frequency register is overwritten with 100 Hz. Finally, the module reads 200 and subtracts it from 500. If the difference is negative, no overwriting is performed. After the traversal is complete, the frequency node scalar corresponding to the maximum value of 100 Hz is extracted from the frequency register, generating the final fundamental frequency signal value of 100 Hz.

[0054] S305: Extract the coordinates of the first to fifth order multiples of the fundamental frequency signal value.

[0055] S306: Within the frequency domain energy matrix, retrieve the amplitude energy value corresponding to the product coordinate, and merge the obtained multi-order values ​​into a one-dimensional structure in ascending order to generate a set of harmonic energy.

[0056] Specifically, the module retrieves the baseband signal value and internally has a fixed set of integer multipliers. This set contains only five integer scalars: 1, 2, 3, 4, and 5. The retrieved baseband signal value is then monotonically multiplied by each of these five integer scalars to calculate the specific frequency coordinates corresponding to each order. That is, the module multiplies the baseband signal value by 1 to obtain the coordinates of the first-order multiplier product, multiplies the baseband signal value by 2 to obtain the coordinates of the second-order multiplier product, and so on until the baseband signal value is multiplied by 5 to obtain the coordinates of the fifth-order multiplier product.

[0057] Furthermore, carrying these five specifically calculated multiplicative coordinates, the module reconnects to the frequency domain energy matrix. Each multiplicative coordinate is used as a query keyword, and a precise numerical matching search is performed within the frequency index axis of the frequency domain energy matrix. Once the corresponding horizontal axis frequency index is found, the module reads the amplitude energy value at the intersection of the row and column corresponding to that index position, extracting a total of five multi-order energy values. Subsequently, a new 1D contiguous memory block is allocated, and the five extracted multi-order values ​​are sequentially stored in consecutive address bits of this 1D structure memory block in strict physical ascending order from order 1 to order 5, completing the value merging and finally encapsulating the 1D structure. When the input fundamental frequency signal value is 100 Hz as calculated above, 100 Hz is multiplied by 1, 2, 3, 4, and 5 respectively, and the coordinates of the first-order multiple product are calculated as 100 Hz, the second-order multiple product coordinates as 200 Hz, the third-order multiple product coordinates as 300 Hz, the fourth-order multiple product coordinates as 400 Hz, and the fifth-order multiple product coordinates as 500 Hz. The frequency doubling extraction submodule retrieves these 5 product coordinates within the frequency domain energy matrix. If the retrieved amplitude energy value is 500 for 100 Hz, 400 for 200 Hz, 250 for 300 Hz, 100 for 400 Hz, and 500 for 50 Hz, the module stores these 5 specific values ​​(500, 400, 250, 100, and 50) into a 1D array memory in chronological order, generating a harmonic energy set containing these 5 values.

[0058] In this embodiment of the invention, by slicing and performing a Fast Fourier Transform on the audio vibration waveform sequence, the time-domain waveform is accurately mapped to the frequency domain dimension, and a frequency-domain energy matrix is ​​constructed. This achieves the transformation of the arteriovenous fistula blood flow vibration signal from a complex time-domain waveform to a clear frequency-domain energy distribution. Based on this, the fundamental frequency signal with the highest energy is selected by traversing the matrix, and the energy values ​​corresponding to the product coordinates of the first to fifth order multiples of the fundamental frequency signal are extracted, constructing a standardized harmonic energy set. This processing effectively removes noise and irrelevant frequency components from the signal, providing structured frequency-domain feature data for subsequent quantitative analysis of the harmonic attenuation law of the arteriovenous fistula blood flow state, significantly improving the objectivity and consistency of arteriovenous fistula stenosis feature extraction.

[0059] S4: Based on the harmonic energy set, the attenuation slope of adjacent order energy peaks is fitted according to the order rising direction, and the overall harmonic attenuation trend of the arteriovenous fistula audio signal is extracted to generate harmonic attenuation characteristic parameters.

[0060] Among them, the arteriovenous fistula audio signal refers to the electrical signal characterization of the mechanical vibration waveform reflecting the blood flow state inside the fistula, which is collected by a vibration pickup device attached to the surface of the arteriovenous fistula.

[0061] Specifically, based on the harmonic energy set, the energy peak corresponding to each order is extracted in ascending order. The energy peak corresponding to the next order is subtracted from the energy peak corresponding to the previous order. The energy change amplitude parameter between adjacent order nodes is read to generate the difference between adjacent orders.

[0062] Specifically, the generated harmonic energy set is loaded and identified as a 1D vector containing 5 elements. An internal difference storage queue is initialized. Following the physical storage sequence of the vector elements (i.e., in ascending order from 1st to 5th order), the corresponding energy peaks at each order node are extracted one by one. The internal arithmetic logic unit initiates a difference subtraction operation mechanism. For two adjacent order nodes, the energy peak corresponding to the next order is strictly extracted as the minuend, and the energy peak corresponding to the preceding order is extracted as the subtrahend. The energy peak corresponding to the next order is subtracted from the energy peak corresponding to the preceding order, and the resulting signed real number is recognized by the module as the energy change amplitude parameter between the two adjacent order nodes.

[0063] Furthermore, for the five elements in the one-dimensional vector, four complete difference subtraction operations need to be performed. The energy change amplitude parameters with positive and negative signs obtained from each operation are pushed into the difference storage queue in the order of calculation to obtain four independent difference values.

[0064] For example, in a real-world computational scenario, when the harmonic energy set data passed from the aforementioned module is invoked—that is, the first-order energy peak value is 500, the second-order energy peak value is 400, the third-order energy peak value is 250, the fourth-order energy peak value is 100, and the fifth-order energy peak value is 50—the following subtraction operations are performed: the first subtraction operation is performed, subtracting the first-order energy peak value from the second-order energy peak value of 400, resulting in an energy change amplitude parameter of -100. The second subtraction operation is performed, subtracting the second-order energy peak value from the third-order energy peak value of 250, resulting in an energy change amplitude parameter of -150. The third subtraction operation is performed, subtracting the third-order energy peak value from the fourth-order energy peak value of 100, resulting in an energy change amplitude parameter of -150. The fourth subtraction operation is performed, subtracting the fourth-order energy peak value from the fifth-order energy peak value of 50, resulting in an energy change amplitude parameter of -50. The four specific real numbers -100, -150, -150, and -50 are calculated and concatenated in order to generate a complete sequence of adjacent order difference values.

[0065] Furthermore, a virtual 2D Cartesian coordinate system is constructed in memory by invoking a sequence of adjacent order differences containing four real numbers. The logical order span indices corresponding to these four adjacent order differences are read and defined as 1, 2, 3, and 4 respectively. These four order indices are directly assigned as the horizontal coordinate values ​​of the virtual coordinate system. Simultaneously, each of the four adjacent order differences is assigned as its corresponding vertical coordinate value. Through this pairing relationship between the horizontal and vertical coordinates, four discrete points with precise coordinate positioning are established within the 2D coordinate system.

[0066] Furthermore, the least squares linear regression kernel is activated for these four discrete points. First, the total sum of all x-coordinate values ​​and the total sum of all y-coordinate values ​​are calculated. Then, the x-coordinate and y-coordinate of each discrete point are multiplied to obtain a cross-product term, and the total sum of all cross-product terms is calculated. Simultaneously, the total sum of squared terms multiplied by each x-coordinate value itself is calculated. The module multiplies the total sum of x-coordinate values ​​by the total sum of y-coordinate values ​​and divides this by the total number of discrete points (4) to obtain the average baseline bias term. The average baseline bias term is subtracted from the total sum of cross-product terms to obtain the regression numerator bias. Next, the total sum of x-coordinate values ​​is multiplied by itself and divided by 4 to obtain the x-axis baseline bias term. The x-axis baseline bias term is subtracted from the total sum of squared x-coordinate terms to obtain the regression denominator bias. Finally, the regression numerator bias is divided by the regression denominator bias to calculate the real number representing the slope of the regression line, and this real number is directly generated and encapsulated as the decay slope coefficient.

[0067] For example, in a practical calculation scenario, the previously calculated x and y coordinates are input, i.e., the first point has x coordinate 1 and y coordinate -100, the second point has x coordinate 2 and y coordinate -150, the third point has x coordinate 3 and y coordinate -150, and the fourth point has x coordinate 4 and y coordinate -50. The total sum of the x coordinates is: 1 + 2 + 3 + 4 = 10. The total sum of the y coordinates is: -100 + (-150) + (-150) + (-50) = -450. The cross-product term is: 1 × (-100) + 2 × (-150) + 3 × (-150) + 4 × (-50), and the final result is -1050. The sum of the squared x coordinates is: 1 + 4 + 9 + 16 = 30. The average baseline bias term is: 10 × (-450) / 4 = -1125. Subtracting -1125 from the cross-product term (-1050) yields a regression numerator bias of 75. The horizontal axis baseline bias term is calculated as 10 multiplied by 10 and divided by 4, equaling 25. Subtracting 25 from the horizontal axis square term (30) yields a regression denominator bias of 5. The module divides the regression numerator bias of 75 by the regression denominator bias of 5, ultimately calculating a decay slope coefficient of 15.

[0068] Specifically, for the attenuation slope coefficient, the test reference constant within the preset sample dataset is obtained, the absolute deviation between the attenuation slope coefficient and the test reference constant is calculated, and the absolute deviation is weighted and calculated with the proportional mapping weight component to generate harmonic attenuation characteristic parameters.

[0069] Furthermore, the calculated attenuation slope coefficient of 15 is extracted and connected in parallel to a pre-set sample database in the system backend. This database stores a large amount of real acoustic attenuation statistics from clinically diagnosed normal arteriovenous fistula patients. Following the data query command, a test baseline constant specifically configured for the attenuation slope is retrieved from the pre-set sample database. This test baseline constant is a healthy reference baseline derived by arithmetically averaging the harmonic attenuation performance of 1000 normal fistula vessels in a non-stenotic state. The specific value of this test baseline constant is set to 10.

[0070] Further, the attenuation slope coefficient and the test reference constant are extracted. The attenuation slope coefficient is subtracted from the test reference constant to obtain the initial deviation value. Then, the module initiates absolute value correction logic, removing the sign bit of the initial deviation value and retaining only the absolute magnitude to generate the absolute deviation. Next, a preset proportional mapping weight component is retrieved from system memory. This component is used to adjust the dominant position of the deviation in the comprehensive feature evaluation system. Based on the correlation analysis of long-term clinical follow-up data with the sensitivity to slope variation, this proportional mapping weight component is fixed at 0.8. The calculated absolute deviation is then multiplied by the proportional mapping weight component using pure arithmetic weighting. The final floating-point result is assigned to a new variable node to generate harmonic attenuation characteristic parameters.

[0071] For example, in a real-world computing scenario, when the attenuation slope coefficient of 15 from the pre-calculation process is input, and the module successfully extracts the test baseline constant of 10 from the database, the trend feature quantification submodule subtracts 10 from 15, calculating an initial deviation value of positive 5. After absolute value positiveing ​​logic processing, the absolute deviation remains at 5. Subsequently, the module extracts a proportional mapping weight component with a set value of 0.8, and performs a multiplication operation between the absolute deviation of 5 and the proportional mapping weight component of 0.8, i.e., 5 multiplied by 0.8, resulting in a final result of 4.

[0072] In this embodiment of the invention, by performing adjacent difference operations on the peak values ​​of each order in the harmonic energy set, the gradual attenuation amplitude of harmonic energy with increasing order is quantified. Based on this, a two-dimensional coordinate system is constructed for linear regression fitting, and the attenuation slope coefficient reflecting the overall attenuation slope is extracted. This coefficient is further compared with a health benchmark constant determined based on large-sample clinical data, the absolute deviation is calculated and weighted, and standardized harmonic attenuation characteristic parameters are generated. This process transforms the abstract harmonic attenuation law in the arteriovenous fistula audio signal into quantifiable characteristic parameters, effectively characterizing the dissipation characteristics of vibration energy when blood flows through the fistula, and providing an objective, continuous, and pathophysiologically closely related quantitative indicator for subsequent assessment of the risk of fistula stenosis.

[0073] S5: Based on the harmonic attenuation characteristic parameters, the discreteness of the changes in the blood flow state of the arteriovenous fistula in continuous time periods is extracted according to the sampling time sequence to generate blood flow temporal fluctuation parameters.

[0074] Optionally, the blood flow temporal fluctuation parameters specifically include: coefficient of variation, range, and standard deviation.

[0075] In one possible implementation, S5 specifically includes sub-steps S501 to S506: S501: Perform mapping matching between harmonic attenuation characteristic parameters and corresponding preset sampling time series.

[0076] It should be noted that those skilled in the art can set the size of the preset sampling time series according to actual needs, and this invention does not limit this.

[0077] S502: According to the sampling time increment rule, the matched feature state items are continuously recombined and spliced ​​to generate a feature fluctuation sequence.

[0078] Specifically, the memory relay station is activated to continuously acquire harmonic attenuation characteristic parameters output by the trend feature quantization submodule within different time periods. An internally pre-established circular data buffer corridor with a capacity of 1000 floating-point nodes is used. A preset sampling time series generated by the internal system clock is retrieved. This preset sampling time series consists of a set of monotonically increasing absolute timestamps with a fixed time interval of 1 minute. A one-to-one mapping and binding operation is performed between each newly acquired harmonic attenuation characteristic parameter and the preset sampling time series timestamp when it arrives at the system bus, transforming individual floating-point numbers into feature status items with time tags.

[0079] Furthermore, the memory management pointer is controlled to push the matched feature status items with time tags into the circular data buffer corridor according to the preset absolute time increment rule of the sampling time sequence, i.e., the time sequence logic from early to late. As subsequent feature status items are continuously input, the module appends the new feature status items to the contiguous physical memory address space at the end of the queue, realizing continuous recombination and splicing. When the circular data buffer corridor is filled with a certain number of feature status items, the entire 1-dimensional data strip extending along the time axis is solidified into a feature fluctuation sequence.

[0080] For example, in a real-world computing scenario, after three consecutive 1-minute monitoring cycles, the time-series sequence construction submodule acquires the pre-loaded harmonic attenuation characteristic parameter value of 4 in the first minute and binds it to the timestamp of the first minute to form the first characteristic state item. In the second minute, it acquires a new harmonic attenuation characteristic parameter value of 4.2 and binds it to the timestamp of the second minute to form the second characteristic state item. In the third minute, it acquires a new harmonic attenuation characteristic parameter value of 4.5 and binds it to the timestamp of the third minute to form the third characteristic state item. Following a time-incrementing rule, the time-series sequence construction submodule sequentially fills 4, 4.2, and 4.5 into contiguous memory, and the resulting composite content is a characteristic fluctuation sequence containing three consecutive values: 4, 4.2, and 4.5.

[0081] S503: Construct a dynamic observation window along the time axis.

[0082] S504: Through a dynamic observation window, local state tracking is performed on consecutive adjacent feature nodes within the feature fluctuation sequence. The relative offset of time-domain fluctuations between feature nodes is calculated, and the difference between adjacent relative offsets is performed to generate a state discrete deviation.

[0083] Specifically, the output feature fluctuation sequence is retrieved, and a dynamic observation window is initialized along the time axis of the sequence, i.e., the direction of memory address increment. The volume of this dynamic observation window is set to cover the width of three consecutive feature nodes. The time step distance of sliding the dynamic observation window backward by one feature node is controlled, and the continuous adjacent feature nodes of the feature fluctuation sequence falling within the window are locally tracked and scanned through the dynamic observation window. During each window stop observation period, the module extracts the values ​​corresponding to two adjacent feature nodes within the window, subtracts the value corresponding to the previous time node from the value corresponding to the later time node, and performs a subtraction difference calculation. The resulting difference is the relative offset of the temporal fluctuation between these two feature nodes.

[0084] Furthermore, since the observation window contains three nodes, the module generates offsets for two adjacent nodes. Then, it subtracts the previous time-domain fluctuation relative offset from the later one, performing a second-order difference operation. When the pre-generated characteristic fluctuation sequence containing three consecutive values ​​(4, 4.2, and 4.5) is invoked, and the dynamic observation window precisely covers these three nodes, the module first tracks the first node (4) and the second node (4.2), subtracting 4 from 4.2 to calculate the first time-domain fluctuation relative offset as 0.2. The module then tracks the second node (4.2) and the third node (4.5), subtracting 4.2 from 4.5 to calculate the second time-domain fluctuation relative offset as 0.3. Subsequently, the state change analysis submodule performs a second-order difference calculation, subtracting the first offset (0.2) from the second offset (0.3) to calculate a second-order difference result of 0.1. This value of 0.1 is used as a local acceleration index and directly generated as the state discrete deviation of the current data segment.

[0085] S505: Perform global time-domain aggregation analysis on the discrete deviation of the state to extract the global discrete evolution trend of the blood flow state over a continuous time period.

[0086] Global time-domain aggregation analysis refers to the overall mathematical statistics and fusion processing of a series of discrete deviations reflecting changes in blood flow state on a continuous time axis, in order to extract the overall discrete evolution law of blood flow state over the entire monitoring period.

[0087] S506: Based on the global discrete evolution trend, perform dynamic mapping processing to generate blood flow temporal fluctuation parameters.

[0088] Specifically, a series of discrete state deviations are extracted, and an internal global statistical calculation architecture is constructed to perform global time-domain aggregation analysis on these deviations. First, all input discrete state deviations are iteratively summed. Then, the sum is divided by the total number of input discrete state deviations to calculate the arithmetic mean baseline value. Next, each discrete state deviation is iterated over again, and subtraction is performed on each deviation with the arithmetic mean baseline value to obtain the centered residual for each deviation. Each centered residual is then multiplied by itself to calculate the corresponding squared residual value. The module performs a summation operation on all squared residual values. Finally, the sum of squared residuals is divided by the difference between the total number of discrete state deviations and 1 to calculate the variance value. This variance value represents the global discrete evolution trend of the blood flow state over a continuous time period.

[0089] Furthermore, a pre-set constant scalar correction coefficient is extracted from the system. This constant scalar correction coefficient is determined by calibrating the background noise drift level of different sensor batches and is set to 1.5. The fluctuation degree quantification submodule performs dynamic mapping processing based on the global discrete evolution trend. Specifically, the calculated variance value is multiplied by the constant scalar correction coefficient.

[0090] For example, suppose the system inputs three discrete state deviations: 0.1 calculated beforehand, and 0.2 and 0.3 calculated over two subsequent time windows. The volatility quantification submodule first calculates the average: 0.1 + 0.2 + 0.3 = 0.6. Dividing 0.6 by 3 gives the arithmetic mean baseline value of 0.2. It then calculates the centered residuals: 0.1 - 0.2 = -0.1, 0.2 - 0.2 = 0, 0.3 - 0.2 = 0.1. Next, it calculates the squared residuals: -0.1 multiplied by itself gives 0.01, 0 multiplied by itself gives 0, and 0.1 multiplied by itself gives 0.01. The module sums all the squared residuals: 0.01 + 0 + 0.01 = 0.02, 0.02 / (3 - 1) = 0.01, and calculates the variance value as 0.01. Then, the constant scalar correction coefficient of 1.5 is called, and 0.01 is multiplied by 1.5 to calculate the final result of 0.015. This 0.015 is generated by the module and established as the blood flow temporal fluctuation parameter in the current long period.

[0091] Among them, dynamic mapping processing refers to the process of mapping the global discrete evolution trend obtained after global time-domain aggregation analysis into the final blood flow time-series fluctuation parameters through specific mathematical transformation rules.

[0092] In this embodiment of the invention, a characteristic fluctuation sequence reflecting the dynamic changes in the blood flow state of the arteriovenous fistula is constructed by mapping, matching, and recombining harmonic attenuation characteristic parameters according to the sampling time sequence. Based on this, a dynamic observation window is constructed along the time axis, and local tracking and secondary difference operations are performed on consecutive adjacent characteristic nodes to generate a state discrete deviation quantity characterizing the acceleration of blood flow fluctuations. Furthermore, a global time-domain aggregation analysis is performed on the state discrete deviation quantities generated by multiple time windows to calculate the variance value reflecting the overall discrete evolution trend, and standardized blood flow temporal fluctuation parameters are generated through dynamic mapping processing. This process organically combines the microscopic fluctuation law and macroscopic evolution trend of the blood flow state of the arteriovenous fistula within a continuous time period, realizing a multi-scale quantitative characterization from instantaneous changes to long-term stability, providing rich and stable temporal characteristic inputs for the subsequent objective assessment of stenosis risk.

[0093] S6: Combining preset fluctuation standard thresholds, the blood flow temporal fluctuation parameters are classified and judged through a decision tree classification model to generate arteriovenous fistula monitoring results.

[0094] Among them, the decision tree classification model refers to a machine learning classification algorithm based on a tree structure.

[0095] Optionally, arteriovenous fistula monitoring results include the risk level of vascular stenosis, patency rate, and abnormal alarm level.

[0096] In one possible implementation, S6 specifically includes sub-steps S601 to S605: S601: Combine the preset fluctuation standard threshold, calculate the difference between each indicator included in the blood flow time-series fluctuation parameter and the corresponding preset fluctuation standard threshold to obtain the difference value.

[0097] S602: Calculate the absolute value of the difference to obtain the fluctuation deviation.

[0098] Specifically, the generated blood flow temporal fluctuation parameters are invoked, and the underlying safety boundary memory is accessed synchronously to extract the preset fluctuation standard threshold. The preset fluctuation standard threshold is set based on the output data of the simulation experimental model of the steady-state tolerance limit of the vascular mechanical properties of the source autofocal fistula. It represents the maximum allowable fluctuation energy without damage to the normal laminar boundary layer, and its specific value is factory-set as a reference value array that matches various indicators.

[0099] Furthermore, an internal multi-channel difference comparator is established. Each indicator contained in the received blood flow temporal fluctuation parameters is placed in the minuend register, and the corresponding preset fluctuation standard threshold is placed in the subtrahend register. Each indicator is subtracted from its corresponding preset fluctuation standard threshold, and the difference operation is directly performed to obtain preliminary algebraic difference values. Subsequently, the module inputs each algebraic difference value to the internal absolute value conversion logic unit, forcibly clearing the sign bit of the algebraic difference value to zero, and calculating the absolute value of the difference as a purely positive number. This absolute value of the difference is then determined as the pure numerical scale indicating whether the current blood flow abnormality exceeds or deviates from the safety benchmark.

[0100] For example, in a real-world computing scenario, when the pre-processing steps are called to obtain the bleeding flow timing fluctuation parameters (taking the coefficient of variation as an example, its value is 0.015), and the corresponding preset fluctuation standard threshold of 0.010 is extracted from the safety boundary memory, 0.015 is subtracted from 0.010, and the algebraic difference is calculated to be positive 0.005. Simultaneously, the range, standard deviation, and other indicators are also subtracted according to their respective dimensions. After processing by the absolute value conversion logic unit, since the value is already positive, the absolute value of this difference remains unchanged at 0.005. Then, the multiple absolute values ​​of the difference, including the value of 0.005, are integrated and output to generate the final fluctuation deviation, which is then pushed into the next level of the system's data pipeline.

[0101] It should be noted that those skilled in the art can set the size of the preset fluctuation standard threshold according to actual needs, and this invention does not limit it.

[0102] S603: Obtain the logical hierarchical decision nodes within a multi-level conditional branch architecture.

[0103] Optionally, the multi-level conditional branching architecture is specifically a logical decision framework built based on a decision tree classification model.

[0104] S604: Perform a comparison operation between the fluctuation deviation and the corresponding logical layer judgment node, filter the associated leaf nodes that meet the logical conditions, and generate node mapping coefficients.

[0105] Specifically, the system calls a fluctuation deviation quantity containing the differences of multiple indicators. A multi-level conditional branching architecture is instantiated in memory. This architecture is a pure logical decision framework built on a decision tree classification model. The decision tree does not perform real-time iterative training; instead, it is fixedly composed of a top-level root node, multiple logical hierarchical decision nodes, and bottom-level related leaf nodes. Pre-defined conditional thresholds are extracted from each logical hierarchical decision node. These thresholds are established before manufacturing by segmenting and calculating historical feature data from 8000 patients based on the principle of minimizing Gini impurity. The threshold for the first-level logical hierarchical decision node is set to 0.003, and the threshold for the second-level left-side logical hierarchical decision node is set to 0.008. The received fluctuation deviation quantity is loaded into the root node, driving the data downwards. At the first-level logical hierarchical decision node, the module extracts a specific indicator component (such as the coefficient of variation deviation) from the fluctuation deviation quantity and performs a numerical comparison with the first-level threshold of 0.003. When this component is less than 0.003, the instruction flow branches to the right and enters the leaf node representing health. When the component is greater than or equal to 0.003, the instruction flow turns to the left branch and continues to descend to the second-level logic layer decision node.

[0106] Furthermore, at the second-level logical layer determination node, the module again compares another related indicator component (or still the same component) in the fluctuation deviation with the second-level threshold of 0.008. If the component is less than 0.008, it flows to the right branch of that layer and enters the first-type abnormal leaf node; if it is greater than or equal to 0.008, it flows to the left branch of that layer and enters the second-type abnormal leaf node. When the data flow reaches the final related leaf node, it reads the dedicated numerical scalar hard-coded inside the leaf node and generates the node mapping coefficient output. When the fluctuation deviation (where the coefficient of variation deviation is 0.005) is received from the front-end module, the decision allocation submodule performs a comparison operation at the first-level node, determines that 0.005 is greater than 0.003, and the data flows to the left branch. Entering the second-level node, it performs a comparison operation, determines that 0.005 is less than 0.008, and therefore the data flows to the right branch of that layer, successfully filtering and falling into the corresponding first-type abnormal related leaf node. The decision allocation submodule reads the preset coefficient parameters inside the leaf node, which are set to 2. Finally, the module uses this value of 2 as the determined classification label and generates specific node mapping coefficients.

[0107] It should be noted that those skilled in the art can set the threshold value of the layered judgment node according to actual needs, and this invention does not limit it.

[0108] S605: Combining the state classification association table in the pre-set sample dataset, the node mapping coefficients are matched and addressed with the reference coefficient values, and the state classification feature code is extracted to generate the arteriovenous fistula monitoring results.

[0109] Optionally, the state classification association table is specifically a set of data in the pre-set sample dataset that stores the mapping relationship between reference coefficient values ​​and classification feature codes.

[0110] Specifically, based on the output node mapping coefficients, the system directly accesses and loads the state classification association table stored in the system's read-only memory. This state classification association table is a structured 2D data set within the pre-set sample dataset that stores a one-to-one correspondence between reference coefficient values ​​and classification feature codes. The state mapping submodule internally initiates a linear addressing mechanism, using the received node mapping coefficients as key index search parameters, and performs a matching addressing operation row by row from top to bottom within the reference coefficient value column of the state classification association table.

[0111] Furthermore, when the internal comparator determines that the reference coefficient value of the current row is numerically equal to the input node mapping coefficient, the matching action is achieved. The read pointer then moves horizontally to the classification feature code column of the current row, directly extracting the physical pathological state classification feature code, composed of a specific Chinese string and numerical combination, stored in that memory unit. Finally, the extracted state classification feature code is encapsulated into a standard external communication data packet, and the final arteriovenous fistula monitoring result is generated for external devices via the display driver interface or remote transmission protocol. In actual operation scenarios, when the node mapping coefficient value of 2 is received from the preceding process, the module scans row by row in the state classification association table to find the record row with a reference coefficient value of 2. Upon locating that row, the module reads the pointer and extracts the Chinese string "moderate stenosis" as the state classification feature code bound to the value 2. This specific pathological qualitative conclusion of "moderate stenosis" is encapsulated into a data frame, ultimately generating and printing the monitoring result "moderate stenosis" on the user terminal display. The advantage of this operational logic is that it avoids complex character translation operations at the underlying software level by using a simple table lookup and direct reading mechanism, thus ensuring the efficiency of the final monitoring report and its accurate correspondence with medical definitions.

[0112] For example, the state classification association table stores a one-to-one correspondence between reference coefficient values ​​and classification feature codes. When the node mapping coefficient is 1, the corresponding classification feature code is described as "normal laminar flow state," indicating that the arteriovenous fistula is in a patent laminar flow state with no significant signs of stenosis. When the node mapping coefficient is 2, the corresponding classification feature code is described as "moderate stenosis," suggesting that the arteriovenous fistula has a certain degree of stenosis and abnormal blood flow. When the node mapping coefficient is 3, the corresponding classification feature code is described as "severe stenosis warning," indicating that the arteriovenous fistula has a serious risk of stenosis and requires timely medical intervention. The deterministic dictionary mapping relationship established within the state mapping submodule for results at different calculation levels directly governs the generation of the final textual conclusion.

[0113] In this embodiment of the invention, a standardized fluctuation deviation is generated by subtracting various indicators in the blood flow temporal fluctuation parameters from a preset fluctuation standard threshold and calculating the absolute value. This effectively eliminates the influence of individual differences among patients and measurement dimensions on the judgment results. Furthermore, based on a pre-set decision tree classification model, the fluctuation deviation is compared layer by layer with multi-level logical hierarchical judgment nodes, automatically selecting relevant leaf nodes that meet the judgment conditions and generating node mapping coefficients. Finally, the node mapping coefficients are converted into intuitive stenosis risk level classification results through table lookup matching. This achieves automated and standardized judgment from continuous quantitative parameters to discrete risk levels, eliminating the reliance on the subjective auscultation experience of medical staff in traditional methods and providing repeatable and traceable objective monitoring conclusions for patients in home settings.

[0114] Reference manual attached Figure 2 The diagram shows a schematic of the structure of a home monitoring system for arteriovenous fistulas based on audio analysis provided by the present invention.

[0115] The present invention also provides an audio analysis-based home monitoring system 20 for arteriovenous fistulas, applied to the above-mentioned audio analysis-based home monitoring method for arteriovenous fistulas, comprising: Processor 201.

[0116] The memory 202 stores computer-readable instructions that, when executed by the processor 201, implement the home monitoring method for arteriovenous fistula based on audio analysis as described in the method embodiment.

[0117] The home monitoring system 20 for arteriovenous fistula based on audio analysis provided by the present invention can perform the above-mentioned home monitoring method for arteriovenous fistula based on audio analysis and achieve the same or similar technical effects. To avoid duplication, the present invention will not elaborate further.

[0118] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0119] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0120] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0121] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0122] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0123] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0124] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0125] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0126] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0127] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0128] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0129] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the home monitoring method for arteriovenous fistula based on audio analysis as described in the method embodiment.

[0131] The present invention provides a computer-readable storage medium that can implement the steps and effects of the home monitoring method for arteriovenous fistula based on audio analysis in the above-described method embodiments. To avoid repetition, the present invention will not repeat the details.

[0132] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0133] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0134] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0135] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0136] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A home monitoring method for arteriovenous fistulas based on audio analysis, characterized in that, include: S1: Acquire analog voltage waveform; S2: Combined with a preset conversion frequency, the analog voltage waveform is converted into discrete voltage values, and the discrete voltage values ​​are arranged in chronological order to obtain an audio vibration waveform sequence; S3: Based on the audio vibration waveform sequence, a frequency energy distribution map is constructed through Fourier transform, and a harmonic energy set is constructed based on the frequency energy distribution map; S4: Based on the harmonic energy set, the attenuation slope of adjacent order energy peaks is fitted according to the order rising direction, and the overall harmonic attenuation trend of the arteriovenous fistula audio signal is extracted to generate harmonic attenuation characteristic parameters. S5: Based on the harmonic attenuation characteristic parameters, the discreteness of the changes in the blood flow state of the arteriovenous fistula in a continuous time period is extracted according to the sampling time sequence to generate blood flow temporal fluctuation parameters. S6: Combining the preset fluctuation standard threshold, the blood flow temporal fluctuation parameters are classified and judged through the decision tree classification model to generate arteriovenous fistula monitoring results.

2. The home monitoring method for arteriovenous fistulas based on audio analysis according to claim 1, characterized in that, The audio vibration waveform sequence specifically includes: a time series index, a discrete amplitude sequence, and an envelope amplitude; The harmonic energy set specifically includes: harmonic frequency components, power spectral density, and high-frequency harmonic amplitude; The harmonic attenuation characteristic parameters specifically include: regression intercept, attenuation coefficient, and residual; The blood flow temporal fluctuation parameters specifically include: coefficient of variation, range, and standard deviation; The results of the arteriovenous fistula monitoring include the risk level of vascular stenosis, patency rate, and abnormal alarm level.

3. The home monitoring method for arteriovenous fistulas based on audio analysis according to claim 1, characterized in that, S2 specifically includes: S201: Combined with a preset frequency band range, perform bandpass frequency domain filtering on the frequency band components within the analog voltage waveform, remove frequency band components that exceed the preset frequency band range, and generate the target waveform segment; S202: Combined with the preset conversion frequency, the target waveform segment is sampled at equal intervals to extract the instantaneous voltage amplitude corresponding to the sampling node, and the instantaneous voltage amplitude is converted into a decimal digital value to generate the discrete voltage value; S203: Read the recording timestamp tag corresponding to the discrete voltage value; S204: Based on the recorded timestamp tag, the discrete voltage values ​​are sorted in ascending order according to the time sequence, and all the sorted discrete voltage values ​​are merged and recombined into a one-dimensional array structure to construct the audio vibration waveform sequence.

4. The home monitoring method for arteriovenous fistulas based on audio analysis according to claim 3, characterized in that, The preset frequency band range is specifically defined as: the filtering range determined based on the mechanical vibration characteristics of blood turbulence; The preset conversion frequency is specifically a fixed sampling frequency set according to the channel performance parameters.

5. The home monitoring method for arteriovenous fistulas based on audio analysis according to claim 1, characterized in that, S3 specifically includes: S301: Slice the audio vibration waveform sequence according to a preset length; S302: Through the Fourier transform, the segmented audio vibration waveform sequence is subjected to frequency domain dimension mapping operation, the frequency nodes corresponding to the frequency domain dimension are extracted, and the amplitude energy value corresponding to the frequency nodes is calculated. S303: Arrange the frequency nodes and the amplitude energy values ​​in a two-dimensional manner to generate a frequency domain energy matrix; S304: Traverse the elements inside the frequency domain energy matrix, compare and filter the amplitude energy value with the largest value, extract the frequency node scalar at the horizontal axis position corresponding to the largest amplitude energy value, and generate the fundamental frequency signal value. S305: Extract the coordinates of the first to fifth order multiples product corresponding to the fundamental frequency signal value; S306: Within the frequency domain energy matrix, retrieve the amplitude energy value corresponding to the product coordinate, and merge the obtained multi-order values ​​into a one-dimensional structure in ascending order to generate the harmonic energy set.

6. The home monitoring method for arteriovenous fistulas based on audio analysis according to claim 1, characterized in that, S5 specifically includes: S501: Perform mapping and matching between the harmonic attenuation characteristic parameters and the corresponding preset sampling time series; S502: According to the sampling time increment rule, the matched feature state items are continuously recombined and spliced ​​to generate a feature fluctuation sequence; S503: Construct a dynamic observation window along the time axis; S504: Through the dynamic observation window, local state tracking is performed on consecutive adjacent feature nodes within the feature fluctuation sequence, the relative offset of time-domain fluctuations between feature nodes is calculated, and the relative offsets of adjacent nodes are differentially divided to generate a state discrete deviation. S505: Perform global time-domain aggregation analysis on the discrete deviation of the state to extract the global discrete evolution trend of the blood flow state within a continuous time period; S506: Based on the global discrete evolution trend, perform dynamic mapping processing to generate the blood flow temporal fluctuation parameters.

7. The home monitoring method for arteriovenous fistulas based on audio analysis according to claim 1, characterized in that, S6 specifically includes: S601: Combining the preset fluctuation standard threshold, the indicators included in the blood flow time-series fluctuation parameters are subtracted from their corresponding preset fluctuation standard thresholds to obtain the difference value; S602: Calculate the absolute value of the difference to obtain the fluctuation deviation; S603: Obtain the logical hierarchical decision nodes within a multi-level conditional branch architecture; S604: Perform a comparison operation between the fluctuation deviation and the corresponding logical hierarchical judgment node, filter the associated leaf nodes that meet the logical conditions, and generate node mapping coefficients; S605: Combining the state classification association table in the preset sample dataset, the node mapping coefficients are matched and addressed with the reference coefficient values, and the state classification feature code is extracted to generate the arteriovenous fistula monitoring results.

8. The home monitoring method for arteriovenous fistulas based on audio analysis according to claim 7, characterized in that, The multi-level conditional branching architecture is specifically a logical decision framework built based on the decision tree classification model; The state classification association table is specifically a data set that stores the mapping relationship between reference coefficient values ​​and classification feature codes within the preset sample dataset.

9. A home monitoring system for arteriovenous fistulas based on audio analysis, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the home monitoring method for arteriovenous fistula based on audio analysis as described in any one of claims 1 to 8.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the home monitoring method for arteriovenous fistula based on audio analysis as described in any one of claims 1 to 8.

Images