Evaluation recommendation method for intervention effect of music score playing training on cardiovascular health

Abstract

The invention discloses an evaluation recommendation method for a cardiovascular health intervention effect of music score playing training, and relates to the field of biological signal analysis and machine learning. The method comprises the following steps: firstly, collecting resting pulse waves of a subject as a baseline, guiding the subject to play multiple groups of classified test music scores, synchronously recording pulse waves and behavior data before and after playing, and setting a recovery period; preprocessing such as denoising and baseline correction is carried out on the collected signals, and characteristics such as pulse wave morphology and pulse rate variability are extracted; weighting the characteristic change rate by adopting a CRITIC method, calculating a score of a comprehensive intervention effect of the music score, and taking the highest score of the music score as a personalized tag of a subject; and finally, constructing and training a machine learning model, and realizing personalized music score intelligent recommendation of a new subject by taking the resting baseline features as input and the optimal music score tag as output. According to the method, objective quantitative evaluation of the intervention effect is realized, and the pertinence and suitability of cardiovascular health intervention are improved.

Description

Technical Field

[0001] This invention relates to the fields of biosignal analysis, signal processing, and machine learning, and specifically to a method for evaluating and recommending the effects of musical score playing training on cardiovascular health intervention. Background Technology

[0002] Cardiovascular health is a core component of overall health management, especially valuable for middle-aged and elderly individuals, those with chronic diseases, and those in a sub-healthy state. Effective cardiovascular health interventions require overcoming the limitations of traditional static indicator monitoring and incorporating dynamic and individualized physiological feedback mechanisms to achieve precise regulation and scientific management. Pulse wave signals, as a non-invasive, convenient, and information-rich physiological signal, objectively reflect several key cardiovascular function parameters, including vascular elasticity, peripheral resistance, and heart rate variability. They have become a core basis for real-time, dynamic assessment of cardiovascular health status, providing reliable physiological data support for health interventions.

[0003] In the field of health intervention, music training, especially wind instrument practice, has been proven to indirectly improve cardiovascular function due to its precise control of breathing rhythm and positive regulation of psychological state, making it an important direction for non-pharmacological intervention. However, existing music intervention methods still have significant limitations: First, intervention programs are mostly based on a unified music teaching logic or general psychological regulation principles, lacking a quantitative assessment system for individual cardiovascular physiological status, and failing to achieve precise matching of "physiological characteristics - intervention content"; second, the evaluation of cardiovascular health intervention effects relies heavily on subjective questionnaire feedback or periodic physical examination data, making it difficult to capture the dynamic changes of physiological indicators during wind instrument training in real time, and unable to adjust intervention strategies in a timely manner; third, although some studies have used single parameters such as heart rate variability and blood pressure for relaxation training feedback, they have not been combined with structured and personalized music score wind instrument training, resulting in a generalized intervention form and insufficient individual adaptability, making it difficult to fully realize the intervention potential of wind instrument training for cardiovascular health, and limiting its large-scale application in personalized health management.

[0004] Furthermore, existing technologies lack a complete technical system integrating physiological signal acquisition, feature analysis, intervention effect quantification, and personalized recommendations. This makes it impossible to effectively correlate the multidimensional characteristics of pulse wave signals with key parameters such as rhythm, dynamics, and breathing frequency in musical scores, hindering the development of data-driven personalized intervention programs. Therefore, there is an urgent need for a method that can quantitatively assess the intervention effects of musical score playing training on cardiovascular health based on individual dynamic pulse wave signals and achieve personalized intelligent musical score recommendations. This would address the problems of insufficient intervention targeting, outdated assessment methods, and low system integration in existing technologies, promoting the development of non-pharmacological interventions for cardiovascular health towards precision and personalization. Summary of the Invention

[0005] The main objective of this invention is to overcome the shortcomings and deficiencies of existing technologies and provide a method for evaluating and recommending the effects of musical score playing training on cardiovascular health intervention. This method combines individual dynamic pulse wave signals to match personalized musical scores for subjects, significantly improving the targeting and suitability of cardiovascular health intervention. At the same time, by integrating modern signal processing technology and machine learning algorithms, it can not only evaluate and recommend the effects of cardiovascular health intervention, but also customize exclusive music intervention programs for individuals, thereby promoting the implementation of personalized health management systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] This invention provides a method for evaluating and recommending the effects of musical score playing training on cardiovascular health intervention, comprising the following steps:

[0008] S1. Have the subject sit quietly in a quiet environment until their physiological state is stable, and collect their pulse wave signal at rest as a baseline reference signal; then guide the subject to play multiple sets of preset test scores in sequence, and record the behavioral data during the performance of each set of scores, as well as the pulse wave signal before and after the performance. A recovery period is set between each set of scores to eliminate cross-interference.

[0009] S2. Perform signal denoising, baseline correction, data segmentation and truncation and signal quality assessment on all collected pulse wave signals in sequence, remove signals with substandard quality and obtain effective pulse wave signals;

[0010] S3. Based on the preprocessed effective pulse wave signal, extract feature parameters related to cardiovascular health status, including pulse wave morphological features and pulse rate variability features.

[0011] S4. The extracted feature parameters are weighted using a weighting method based on the correlation of indicators to obtain the intervention effect score corresponding to each test score. The test score with the highest score is used as the sample label of the subject.

[0012] S5. Construct a machine learning model, using the feature parameters corresponding to the resting baseline pulse wave signal of the subject as the model input, and the highest score musical score label as the model output target. After training the machine learning model, personalized training musical score recommendations are realized for new subjects.

[0013] As a preferred technical solution, the specific process of step S1 is as follows:

[0014] S101. Classify the sheet music according to the number of beats per minute, playing dynamics, playing air pressure range, and breathing frequency to form multiple sets of test sheet music;

[0015] S102. After the subject sat quietly with his / her eyes closed, the pulse wave signal in the resting state was collected by a photoplethysmography sensor as a baseline reference signal.

[0016] S103. After the subject sits still to eliminate cross-interference, the pulse wave signal is collected before the performance. Then, the subject plays the designated score using a wind instrument. The pulse wave signal is collected immediately after the performance. This process is repeated until all test scores are played.

[0017] As a preferred technical solution, the specific process of step S2 is as follows:

[0018] S201. A multi-order low-pass Butterworth filter is used to filter the pulse wave signal to remove high-frequency interference noise.

[0019] S202. Use cubic spline interpolation to perform baseline correction on the filtered signal to remove baseline drift interference.

[0020] S203. The baseline-removed signal is segmented using the sliding window method, and a uniform window size and step size are set.

[0021] S204. The attractor reconstruction method is used to evaluate the quality of each segmented signal and select the effective signal segments that meet the preset quality threshold.

[0022] As a preferred technical solution, step S204 specifically includes:

[0023] First, based on the segmented pulse wave fragments, a correlation signal is constructed by setting a time delay, and then a new variable is obtained through projection transformation. The specific variables are used as the horizontal and vertical coordinates respectively to form an image of the reconstructed attractor.

[0024] Next, three key feature points are identified in the reconstructed image to construct an ideal triangle, and the discrete lines corresponding to the three sides of the triangle are obtained.

[0025] Then, the trajectory of the attractor is divided into corresponding segments according to the three feature points mentioned above. The error between each trajectory segment and the corresponding side of the triangle is calculated. The maximum value of the error of all trajectory segments on each side is taken, and the three maximum values ​​are combined to obtain the global error of the attractor.

[0026] Finally, a quality threshold is set. If the global error does not exceed the threshold, the quality of the pulse wave signal segment is deemed to meet the standard; if it exceeds the threshold, it is considered to be substandard and is removed.

[0027] As a preferred technical solution, the specific process of step S3 is as follows:

[0028] S301. Detect the main wave peak of the screened pulse wave signal, and use the identified main wave peak as the reference point to locate a single complete pulse wave cycle.

[0029] S302. Based on a single complete pulse wave cycle after localization, extract the pulse wave morphological features within that cycle; take the average of the morphological features of all valid pulse wave cycles as the standardized morphological features of the corresponding samples.

[0030] S303. Based on the pulse wave signal after peak detection, calculate the pulse rate variability characteristics of the signal, wherein the pulse rate variability characteristics cover time domain characteristics, frequency domain characteristics and nonlinear domain characteristics.

[0031] As a preferred technical solution, in step S301, a first-order derivative peak detection algorithm with adaptive threshold is used to detect the main wave peak of the pulse wave signal, specifically as follows:

[0032] First, the discrete signal is divided into multiple segments of equal length. A time window of a specific length is set at the beginning of each segment. The signal amplitude characteristics and instantaneous heart rate are estimated by finding local maxima, and the initial amplitude threshold and time interval threshold are determined. These two thresholds are dynamically adjusted according to the subsequently detected heartbeat characteristics.

[0033] Next, the entire discrete sequence is scanned, and zero-crossing points are found in the difference sequence. The amplitude of the point is then checked in the original signal to see if it exceeds the current amplitude threshold. At the same time, it is confirmed that the interval between the point and the previous accepted peak point meets the time interval threshold requirement. Only points that meet both of these conditions will be determined as contraction peak points.

[0034] Finally, for each pair of adjacent peak points, the signal segment between them is extracted, and the minimum point in the signal segment is located. This minimum point is used as the starting point of the corresponding pulse wave cycle. This starting point is usually located at the lowest point of each cycle, corresponding to the starting position of the signal waveform.

[0035] As a preferred technical solution, in step S303, the time-domain features include:

[0036] Mean PP interval: It reflects the overall rhythm of the pulse wave by calculating the average of all pulse intervals;

[0037] Normal PP interval standard deviation: By calculating the standard deviation of all pulse intervals, it reflects the overall variability of pulse interval over a longer time scale;

[0038] Root mean square of the difference between adjacent pulse intervals: By calculating the root mean square of the difference between adjacent pulse intervals, it reflects the short-term fluctuation characteristics and is mainly used to assess the level of parasympathetic nerve activity.

[0039] Percentage of adjacent pulse intervals with a difference greater than 20ms: Statistical analysis of the percentage of intervals with a difference greater than 20ms between adjacent pulse intervals out of the total number of intervals;

[0040] The frequency domain features include:

[0041] Low-frequency characteristic values: correspond to a specific frequency range, reflecting the combined regulatory function of the sympathetic and vagus nerves, and are related to the activity of the sympathetic nervous system;

[0042] High-frequency characteristics: corresponding to another specific frequency range, reflecting the regulatory activity of the vagus nerve, and related to respiratory arrhythmias;

[0043] The ratio of low-frequency to high-frequency power reflects the balance between the sympathetic nervous system and the vagus nervous system, and is an indicator of the overall balance of the autonomic nervous system.

[0044] The nonlinear domain features include:

[0045] Sample entropy: It quantifies the complexity of a signal by comparing the similarity probabilities of templates of different lengths in a time series. It has low dependence on data length, is computationally stable, and the larger the value, the more complex the signal.

[0046] Multiscale entropy: As an extension of sample entropy, it calculates and integrates the sample entropy at each scale by transforming the original sequence at different scales, reflecting the overall trend over a long time scale and the detailed features of a short time series.

[0047] Approximate entropy: It quantifies the regularity and complexity of a signal by statistically analyzing the probability of new patterns appearing in a sequence. The larger the value, the more difficult the signal is to predict and the weaker its regularity.

[0048] As a preferred technical solution, the specific process of step S4 is as follows:

[0049] S401. Based on the extracted standardized pulse wave morphological features and pulse rate variability features, calculate the rate of change of the above features relative to the resting baseline before performance after the subject plays each set of test scores.

[0050] S402. Based on the multi-dimensional feature change rate dataset, the weight of each feature parameter is calculated using a weight determination method based on indicator correlation.

[0051] S403. Based on the determined objective weights of each feature, the multidimensional feature change rates generated by each subject after playing each set of test scores are weighted and summed to calculate the comprehensive intervention effect score corresponding to the score. Finally, the test score with the highest comprehensive intervention effect score is selected for each subject, and the unique label of the score is used as the subject's personalized training sample label.

[0052] As a preferred technical solution, in step S402, the weights of each feature parameter are calculated using a weight determination method based on index correlation, specifically as follows:

[0053] First, an evaluation matrix is ​​constructed with the subjects playing different musical scores as rows and the rate of change of each feature as columns;

[0054] Secondly, the evaluation matrix is ​​standardized to eliminate the differences in the dimensions of each feature;

[0055] Next, the correlation coefficient matrix between each feature is calculated to quantify the conflict or contrast strength between features.

[0056] Then, by combining the standard deviation of the standardized features with the conflict between features, the amount of information contained in each feature is calculated;

[0057] Finally, based on the proportion of each feature's information content to the total information content, the final objective weight of each feature parameter is determined.

[0058] As a preferred technical solution, step S5 specifically includes:

[0059] S501. Integrate the data of all subjects, and use the standardized morphological features and pulse rate variability features obtained by processing the resting baseline pulse wave signal of each subject through steps S2 and S3 as the input feature vector of the machine learning model, and use the corresponding highest score musical score label as the output target label to form a structured "feature-label" dataset.

[0060] S502. Divide the training set and validation set according to the subjects to ensure that the data of the same subject belongs to only one set;

[0061] S503. Train a random forest classification model using the training set data, and fine-tune the hyperparameters using grid search combined with five-fold cross-validation. Select the parameter combination with the highest cross-validation accuracy to determine the optimal model.

[0062] S504. Collect the resting pulse wave signal of new subjects, and after preprocessing and feature extraction, input it into the optimal model to output the corresponding best musical score number, thereby realizing personalized recommendations.

[0063] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0064] (1) This invention extracts the morphological features, pulse rate variability time-frequency domain features and nonlinear features of pulse wave signals, and uses the CRITIC method to weight and fuse the multi-dimensional feature change rates to construct a quantitative intervention effect scoring system based on the dynamic changes of physiological signals. This achieves an objective, precise and repeatable evaluation and recommendation of the cardiovascular health intervention effect of music playing training, overcoming the limitations of traditional methods that rely on subjective questionnaires or single parameters and are difficult to monitor in real time.

[0065] (2) This invention innovatively correlates the cardiovascular physiological characteristics of an individual at rest (analyzed by pulse wave signals) with the physiological response after playing a specific musical score, and uses a machine learning model to establish a mapping relationship between “individual baseline characteristics and the best intervention score”, thus constructing a data-driven personalized score recommendation system. This system can accurately match the most suitable playing training content for subjects with different cardiovascular function states, significantly improving the pertinence and suitability of health intervention.

[0066] (3) This invention integrates modern signal processing, feature engineering and machine learning technologies to form a complete technical loop from high-quality physiological signal acquisition, preprocessing and quality assessment, to multi-dimensional feature extraction and fusion, and then to personalized model training and verification. This method has the advantages of being non-invasive, convenient and systematic, and provides a scientific, efficient and scalable quantitative tool and implementation plan for non-drug intervention in cardiovascular health. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of this application, 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 is a flowchart of a recommended method for evaluating the effect of musical score playing training on cardiovascular health intervention disclosed in this invention;

[0069] Figure 2 is an experimental flowchart of an embodiment of the present invention;

[0070] Figure 3 is a flowchart of pulse wave signal preprocessing in an embodiment of the present invention;

[0071] Figure 4 is a waveform diagram of the pulse wave signal after preprocessing in an embodiment of the present invention;

[0072] Figure 5 is a waveform diagram after pulse wave peak detection in an embodiment of the present invention. Detailed Implementation

[0073] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0074] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0075] Example 1

[0076] like Figure 1 As shown in this embodiment, a recommended method for evaluating the effect of sheet music playing training on cardiovascular health intervention includes the following steps:

[0077] S1. Experimental preparation stage: Subjects were asked to sit quietly in a quiet environment until their physiological state stabilized, and their pulse wave signals at rest were collected as baseline reference signals. Then, the subjects were guided to play multiple sets of preset test scores in sequence, and their behavioral data during the performance of each set of scores, as well as their pulse wave signals before and after the performance, were recorded simultaneously. A recovery period was set between each set of scores to eliminate cross-interference.

[0078] Furthermore, step S1 is performed as follows:

[0079] S101. Classify musical scores according to beats per minute, playing dynamics, playing air pressure range, and breathing frequency;

[0080] S102. The subject is asked to sit quietly with his / her eyes closed for a period of time. After the countdown ends, the subject's pulse wave signal is collected using a photoplethysmography sensor.

[0081] S103. The subject sits in front of the screen and remains still for a period of time to eliminate cross-interference. After the countdown ends, the pulse wave signal is collected before the performance. After the collection, the subject plays the designated score using wind instruments such as trombone or saxophone. The pulse wave signal is collected immediately after the performance.

[0082] S2. Pulse wave signal preprocessing: All acquired pulse wave signals are sequentially subjected to signal denoising, baseline correction, data segmentation and truncation, and signal quality assessment. Substandard signals are removed to obtain valid pulse wave signals.

[0083] Furthermore, step S2 is as follows:

[0084] S201. Suppose the collected discrete pulse wave signal sequence is... ,in This is the sampling point index. The output signal sequence after filtering is... The implementation of this digital filter is defined by the following linear constant-coefficient difference equation:

[0085]

[0086] Among them, coefficient and The design parameters of the 7th-order low-pass Butterworth filter are uniquely determined.

[0087] The original input signal sequence Substituting into the aforementioned difference equation, the output signal sequence after filtering out high-frequency noise can be obtained through recursive calculation. .

[0088] S202. Use cubic spline interpolation to process the filtered pulse wave signal. To perform baseline removal, the signal is first detected. For all trough points, construct a cubic spline interpolation function using the trough points as interpolation nodes. Then, the baseline estimate is calculated using a cubic spline function:

[0089]

[0090] Subtracting the estimated baseline from the original signal yields the baseline-removed signal:

[0091]

[0092] S203. The pulse wave signal after baseline drift elimination is segmented using the sliding window method. The 1-minute pulse wave signal is divided into 10-second windows with a step size of 5 seconds.

[0093] S204. The quality of the segmented pulse wave signal is evaluated using the attractor reconstruction method, as follows:

[0094] Assuming the segmented pulse wave is Its average period is Define time delay Define the following two signals:

[0095]

[0096]

[0097] and Origin Therefore, in order to eliminate the constant vertical translation effect, the attractor is projected onto a new plane, and the following new variables are defined:

[0098]

[0099]

[0100]

[0101] by The x-axis is... Projecting the image onto the ordinate yields the reconstructed attractor image. High-quality pulse wave signal segments will form a reconstructed attractor with a typical triangular shape. Therefore, a good indicator of pulse wave signal quality is the degree to which the attractor approximates an ideal triangle. The three vertices of the ideal triangle are named... , and , Defined as The coordinates of the point with the smallest y-axis coordinate can be expressed as: ; Defined as The coordinates of the point with the largest x-axis coordinate can be expressed as: ; Defined as The coordinates of the point with the largest x-axis coordinate can be expressed as: .

[0102] definition For crossing and a straight line, For crossing and a straight line, For crossing and a straight line, , and In discrete form, it can be obtained through the following formula:

[0103]

[0104]

[0105]

[0106] exist After calculating the sides that form the ideal triangle on the plane, the core idea is to quantify the distances between the attractor trajectory and each side of the ideal triangle. To do this, the attractor trajectory is divided into several segments, each located at a point... , and Between. Located at point and The trajectory segment and the straight line between Compare them. The trajectory and the straight line can be obtained through formula (12). The error. Among them, Point and The attractor trajectory segment between them. The same analysis is performed on the other two sides of the triangle, calculated by formulas (13) and (14) respectively, where It is a point and The attractor trajectory segment between them, and It is a point and The attractor trajectory segment between them.

[0107]

[0108]

[0109]

[0110] Since the attractor trajectory tends to traverse each side of the triangle multiple times, it is necessary to calculate the error of each side across all corresponding trajectory segments and take the maximum value as the final error for that side. The calculated maximum error value... , and This can be represented as a point in a three-dimensional Cartesian coordinate system. It can be expected that if the error value is small, the point will be closer to the origin; conversely, if the error value is large, the point will be farther from the origin. The magnitude of the vector from the origin to this error point is calculated using the following formula, and this magnitude is defined as the global error of the attractor:

[0111]

[0112] In high-quality pulse wave signals, most of the time intervals The values ​​tend to be low, thus forming a positively skewed distribution. As a measure of attractor quality, [the value is used to]... The distribution defines a threshold. :

[0113]

[0114] in, An attractor indicating "good" quality. An attractor that indicates "bad" quality.

[0115] S3. Feature Parameter Extraction: Based on the preprocessed effective pulse wave signal, feature parameters related to cardiovascular health status are extracted. The feature parameters include pulse wave morphological features and pulse rate variability features.

[0116] Furthermore, step S3 is as follows:

[0117] S301. A first-order derivative peak detection algorithm with adaptive threshold is used to detect the peak of the pulse wave signal. The core idea is to calculate the first derivative of the photoplethysmography pulse wave signal and adaptively determine the threshold to achieve accurate positioning of the peak point.

[0118] For the preprocessed signal The first difference is calculated using the following formula:

[0119]

[0120] differential sequence The zero-crossing points are initially identified as candidate locations near the contraction peaks in the corresponding original signal. To avoid false zero-crossing points introduced by noise, this method employs an adaptive threshold mechanism to filter out true feature points.

[0121] The algorithm first processes the entire discrete signal Divide the text into several paragraphs of equal length. At the beginning of each paragraph, define a length of... A discrete-time window is defined, within which the amplitude characteristics of the signal and the instantaneous heart rate are estimated using simple methods such as finding local maxima, and an initial amplitude threshold is calculated. and time interval threshold These thresholds will be dynamically updated in subsequent processing based on newly detected cardiac characteristics.

[0122] The final peak detection scans the entire discrete sequence, in the difference sequence. The zero-crossing point is located in the middle, and in the original signal The check determines whether the corresponding amplitude exceeds the current amplitude threshold. At the same time, it was confirmed that the index difference between this point and the previous accepted peak point met the threshold. The requirement is that only discrete points that simultaneously satisfy both amplitude and interval conditions are allowed. Only then will it be determined as the true peak of the systolic phase.

[0123] The starting point is then determined by detecting the minimum value between each pair of adjacent peaks. This starting point is generally located at the lowest point within each cycle, corresponding to the beginning position of the signal waveform. For each pair of adjacent peaks, the signal segment between them is extracted, and the minimum point within that signal segment is located, which is then used as the starting point of the corresponding cycle.

[0124] S302. Based on a single pulse wave cycle, extract its morphological features, including the peak systolic blood pressure. Peak diastolic blood pressure Double notch Pulse interval Augmentation Index Substitution Enhancement Index The ratio of the second beat notch to the peak contraction Negative relative growth index Peak duration of systole Double notch time diastolic peak time Features such as morphological features are obtained by taking the mean of the feature value sequence for all periods.

[0125] S303. Calculate multiple PRV features related to cardiovascular health in the pulse wave signal. The core is to extract the pulse wave... Interval, or time difference between adjacent peaks. This invention extracts the time-domain, frequency-domain, and nonlinear-domain features of PRV. The specific features and calculation methods are as follows:

[0126] S303.1 Extracting the temporal features of PRV:

[0127] (1) Mean interval of PP ( The average PP interval is used to reflect the overall rhythm of the pulse wave, and is calculated as follows:

[0128]

[0129] in For the first The interval between pulse beats This represents the total number of periods.

[0130] (2) Standard deviation of normal PP interval ( ): This represents the standard deviation of the intervals between all heartbeats in a signal segment, i.e., the standard deviation of the PP interval. It reflects the overall variability of the pulse interval over a longer time scale. The calculation formula is as follows:

[0131]

[0132] in, This is the mean of all calculated PP intervals.

[0133] (3) Root mean square of the difference between adjacent PP periods ( ): This represents the root mean square of the difference between the PP intervals, reflecting short-term fluctuations. It is mainly used to assess the activity level of the parasympathetic nervous system. The calculation formula is as follows:

[0134]

[0135] (4) The ratio of the total number of intervals with an interval difference greater than 20ms between adjacent PP intervals ( The percentage of intervals with an adjacent PP interval greater than 20ms is calculated using the following formula:

[0136]

[0137] NN20 represents the total number of intervals with an adjacent interval greater than 20 ms.

[0138] S303.2 uses the Welch method to perform frequency domain analysis on the pulse wave signal, assuming there is a time series Divide it into L segments, each containing M data points. Then the power spectrum of the p-th segment is:

[0139]

[0140] in, It is a normalization factor that allows for adjustment. The normalization factor ensures that the calculated spectral estimate is asymptotically unbiased. For the window function to be added, you can generally choose Hanning window, rectangular window, Blackman window, Chebyshev window, etc.

[0141] The original signal can be obtained by summing and averaging the power spectra of the L segments. Global spectral estimation:

[0142]

[0143] Based on the above process, frequency domain features can be extracted, which reflect different physiological activities of the human body. The calculation methods for each feature are as follows:

[0144] (1) Low frequency ( The range is 0.04Hz to 0.15Hz, reflecting the combined regulatory function of the sympathetic and vagus nerves, and is mainly related to the activity of the sympathetic nervous system. Its calculation formula can be expressed as:

[0145]

[0146] (2) High frequency ( The range is 0.15Hz to 0.4Hz, mainly reflecting the regulatory activity of the vagus nerve and related to respiratory arrhythmias. Its calculation formula can be expressed as:

[0147]

[0148] (3) The ratio of low-frequency power to high-frequency power ( (I): This reflects the balance between the sympathetic nervous system and the vagus nervous system, and is an indicator of the overall balance of the autonomic nervous system.

[0149] S303.3 Nonlinear characteristic analysis is an important method for studying signal complexity and dynamic changes. Unlike linear analysis methods, which are limited to the overall trend of the signal or the distribution of specific energy ranges, this method emphasizes the assessment of the local dynamics and overall complexity of time series, and can capture deeper information that cannot be revealed by time-domain and frequency-domain analysis. Commonly used nonlinear characteristic analysis methods include: sample entropy, multi-scale entropy, and approximate entropy.

[0150] (1) Sample entropy

[0151] Sample entropy ( Sample entropy is a non-linear metric used to measure the complexity and regularity of time series data. It quantifies the complexity of a signal by comparing the similarity probabilities of templates within the time series. It has low dependence on data length and its calculation method is stable. The calculation method is as follows:

[0152]

[0153] For a given time series The construction length is template vector sequence ,in, This indicates that all sequences of length 1 are extracted sequentially from the time series. The subsequences of the vector template are used to construct a vector template. The maximum difference between all corresponding points in two random templates is calculated, which is defined as the distance, as shown in the following formula:

[0154]

[0155] in, It's a template. and Distance is a measure of the similarity between two templates. Using the maximum difference method can capture larger fluctuations in the sequence.

[0156] For a given vector Calculation with other templates The distance between them is less than or equal to the tolerance. The number of matches. Define the number of matches. The calculation formula is as follows:

[0157]

[0158] in It is an indicator function, and the distance between templates is less than or equal to... hour, ,otherwise It is 0.

[0159] Calculate under a given tolerance The similarity between templates, i.e., sample entropy, is defined as the similarity between templates within the tolerance range. Within the range, the length is The template and length are The ratio of template matching accuracy. The specific formula is as follows:

[0160]

[0161] It is a length of The number of times the template is matched. It is a length of The number of template matches. Sample entropy reflects the complexity of a signal; the higher the sample entropy, the more complex the signal; the lower the sample entropy, the less complex the signal.

[0162] (2) Multiscale entropy

[0163] Multiscale entropy ( Multiscale entropy is used to analyze the complexity of time series at different scales. It is an extension of sample entropy. By weighting the series at different scales, it can capture a richer description of the complexity of the signal. Large-scale entropy can analyze the overall trend and pattern over a long time scale, while small-scale entropy can more comprehensively reveal the detailed features within a short time series.

[0164] First, the original time series... Perform scaling transformations to obtain new sequences at different scales. The formula is as follows:

[0165]

[0166] , As a scale, It is composed of adjacent elements in the original sequence. The average of the data points is used to construct the average of the data points. It is the length of the original time series. It is a scale The length of the new time series. For the new series... The template vector and distance are constructed using sample entropy, and the number of matches is calculated. Similar to formulas (27) and (28), the simplified multi-scale entropy is finally calculated. As shown below:

[0167]

[0168] Different results can be obtained from the above formula. The sample entropy at one scale is thus obtained as multi-scale entropy. Multi-scale entropy provides a method for measuring the complexity of time series data at multiple scales.

[0169] (3) Approximate entropy

[0170] Approximate entropy ( (This is a non-linear indicator that quantifies the regularity and complexity of time series data. It reflects the dynamic characteristics of a system by statistically analyzing the probability of new patterns appearing in the sequence.) A larger value indicates a higher level of signal complexity. The calculation method is as follows:

[0171] Given time series The construction length is template , representing each in the sequence A dimensional vector. As shown below:

[0172]

[0173] Similar to calculating the sample entropy distance, the approximate entropy also uses the Chebyshev distance to represent the similarity between templates, as shown below:

[0174]

[0175] The smaller the Chebyshev distance, the higher the similarity between templates; the larger the distance, the greater the difference. For templates... Calculate other templates Inter-match degree As shown below:

[0176]

[0177] Based on length and Log average of the matching degree of each template and The formula is as follows:

[0178]

[0179]

[0180] The value quantifies the complexity of the time series. A higher value indicates that the signal is not easy to predict; a lower value indicates that the signal has strong regularity and is easy to predict.

[0181] S4. Intervention effect score and label determination: The extracted feature parameters were weighted using the CRITIC method based on the correlation of indicators to obtain the intervention effect score corresponding to each test score. The test score with the highest score was used as the sample label of the subject.

[0182] Furthermore, step S4 is as follows:

[0183] S401. This step aims to quantify the dynamic changes in the subject's cardiovascular physiological characteristics relative to their pre-performance resting baseline after playing each set of test scores. The rate of change is calculated based on the standardized pulse wave morphology and pulse rate variability features extracted in step S3.

[0184] Assume each subject plays the [number]th [performance / event]. A certain characteristic value before the test score (resting state) is The characteristic value after playing the same score is If this characteristic is positively correlated with cardiovascular health, then the rate of change of this characteristic relative to the resting baseline after performance is... The calculation formula is as follows:

[0185]

[0186] If this characteristic is negatively correlated with cardiovascular health, then the rate of change of this characteristic relative to the resting baseline after performance... The calculation formula is as follows:

[0187]

[0188] If the value of a certain feature is zero or close to zero in the resting state, then it is represented by an absolute change:

[0189]

[0190] S402. This step, based on the multi-dimensional feature change rate dataset constructed in S401, uses the CRITIC method to calculate the weights of each feature parameter. This method comprehensively considers the standard deviation of features and the correlation between features, thereby objectively evaluating the contribution of each feature to the cardiovascular health intervention effect score.

[0191] Assume there is a total Group test sheet music, each group Rate of change of each characteristic. Construct an evaluation matrix. ,in:

[0192]

[0193] in Indicates the performance of the first After the set of scores The rate of change of each feature.

[0194] To eliminate the dimensional differences between different features, the evaluation matrix is ​​normalized using the range standardization method:

[0195]

[0196] in , The standardized value. and They represent the first The minimum and maximum values ​​of the column. The standardized matrix is ​​denoted as... .

[0197] Calculate the normalized matrix Any two features and Pearson correlation coefficient between them:

[0198]

[0199] in

[0200] The correlation coefficients among all features form a symmetric matrix. .

[0201] For the The conflict of a feature is defined as follows:

[0202]

[0203] The larger the value, the lower the correlation between the feature and other features, the stronger its independence, and the more independent information it contains.

[0204] No. Information content of each feature By its standard deviation With conflict Joint decision:

[0205]

[0206]

[0207] Objective weights of each feature Its proportion of the total information content:

[0208]

[0209] And satisfy

[0210] Finally, the weight vector is obtained. .

[0211] S403. This step aims to calculate the comprehensive intervention effect score corresponding to each score by weighted summing of the multidimensional feature change rates generated after each subject plays each set of test scores, based on the objective weights of each feature determined in S402. Finally, the test score with the highest score is selected for each subject, and its unique label is used as the subject's personalized training sample label for subsequent training and recommendation of machine learning models.

[0212] For the The subject, in the performance of the first... After the group tested the sheet music, there were a total of The eigenvector is composed of the rate of change of each characteristic:

[0213]

[0214] in Indicates the first The first characteristic in the performance Rate of change after setting the score.

[0215] Let the feature weight vector determined by the CRITIC method be:

[0216]

[0217] Then the first The subject played the first... Comprehensive intervention scoring effect of music scores The calculation formula is as follows:

[0218]

[0219] This score reflects the overall quantitative evaluation of the effect of the musical score on the cardiovascular health intervention of the subjects; the higher the score, the better the intervention effect.

[0220] For the The test subjects were selected, and all the test scores they had played were reviewed (let there be a total of 10 subjects). (Group), the sheet music with the highest comprehensive intervention effect score was selected as its personalized best training sheet music:

[0221]

[0222] The unique label of the best score was used as the sample label for that subject. :

[0223]

[0224] S5. Personalized Music Score Recommendation: Construct a machine learning model, using the feature parameters corresponding to the subject's resting baseline pulse wave signal as the model input, and the highest-rated music score label as the model output target. After training the machine learning model, personalized training music score recommendation is achieved for new subjects.

[0225] Furthermore, step S5 is as follows:

[0226] S501. Construct the model training dataset: Integrate the data of all subjects, and use the standardized morphological features and pulse rate variability (PRV) features obtained by processing the resting baseline pulse wave signal of each subject through steps S2 and S3 as the input feature vector of the model; use the unique label corresponding to the best intervention effect score of the subject determined in step S4 as the output target label of the model to form a structured "feature-label" dataset.

[0227] S502. A subject-based partitioning strategy is adopted, randomly dividing all subjects into training set subjects and validation set subjects. In practice, the partitioning ratio between the training and validation sets is first determined; then, all subjects are numbered and randomly sorted to ensure the randomness and unbiasedness of the partitioning process. The partitioning strictly adheres to the "subject independence" principle, meaning that all data samples from the same subject (including their resting baseline characteristics and corresponding best musical notation labels) must belong entirely to either the training set or the validation set, and cannot span between the two sets. This strategy effectively avoids model overfitting and evaluation bias caused by the repeated occurrence of the same subject's physiological characteristics during the training and validation phases, thus more realistically reflecting the model's classification and generalization ability when facing entirely new subjects, and improving the practical applicability and robustness of the recommendation system.

[0228] S503. Train a random forest classification model using the training set data, and use grid search combined with five-fold cross-validation to systematically tune the key hyperparameters of the model. Use the cross-validation accuracy as the evaluation index to select the optimal parameter combination.

[0229] Finally, the hyperparameter combination with the highest cross-validation accuracy was selected as the optimal model parameters, and the random forest classifier was retrained based on the full training set data to obtain the final optimized personalized music score recommendation model. This process effectively avoids model overfitting and improves the predictive stability and classification accuracy for new subjects.

[0230] S504. For new subjects, pulse wave signals are collected at rest. After preprocessing and feature extraction in steps S2 and S3, standardized morphological features and pulse rate variability feature vectors are obtained and used as input to the model. The model automatically predicts and outputs the corresponding optimal musical notation based on these input features, realizing intelligent recommendation of personalized playing training scores.

[0231] Example 2

[0232] In a more specific embodiment, a recommendation for assessing the effects of sheet music playing training on cardiovascular health interventions is disclosed, with the following specific steps:

[0233] S1. Experimental Preparation Phase: Subjects were asked to sit quietly in a quiet environment for 5 minutes to achieve physiological stabilization. Then, their pulse wave signals were collected during a 1-minute resting state and used as a baseline reference signal. Next, subjects were guided to play multiple sets of prepared test scores sequentially. Behavioral data during each set of scores, as well as pulse wave signals before and after each performance, were recorded simultaneously. A 5-minute recovery period was set between each set of scores to eliminate cross-interference. The specific steps are as follows:

[0234] S101. The musical scores are classified as follows:

[0235] Beats per minute: 120-100, 100-80, 80-60;

[0236] Playing force: The playing air pressure range is 300 - 500 Pa, and the playing air pressure range is 800 - 1000 Pa;

[0237] Ventilation rate: approximately 5 times / minute, approximately 6 times / minute, approximately 7 times / minute;

[0238] S102. The subject is asked to sit quietly with his / her eyes closed for 5 minutes. After the countdown ends, the subject's pulse wave signal is collected for 1 minute using a photoplethysmography sensor.

[0239] S103. The subject sits in front of the screen and remains still for 5 minutes to eliminate cross-interference. After the countdown ends, the subject collects a pulse wave signal one minute before playing, following the on-screen prompts. After the signal is collected, the subject plays the designated score on a trombone, and a playing task guidance interface is displayed on the screen. The trombone is equipped with a VL53L0X-V2 laser rangefinder module to measure the trombone's position information and an RSCM17100KP010 air pressure sensor to measure the subject's playing pressure. This information is displayed on the screen in real time, guiding the subject to play and breathe according to the correct notes and rhythms. The subject's playing accuracy is calculated by comparing the reference air pressure range and reference position distance corresponding to the standard score. After playing, the subject collects a pulse wave signal one minute after playing, following the on-screen prompts. If the playing accuracy is below 70%, the playing data is discarded, and the score is played again. Figure 2 The overall experimental flowchart is shown.

[0240] S2. All pulse wave signals collected during the experiment were preprocessed. The preprocessing procedure included signal denoising, baseline correction, data segmentation and truncation, and signal quality assessment. Signals that did not meet the quality standards were removed, and valid signals were retained for subsequent analysis. The preprocessing process for the pulse wave signals is as follows: Figure 3 As shown:

[0241] S201. Suppose the collected discrete pulse wave signal sequence is... ,in This is the sampling point index. The output signal sequence after filtering is... The implementation of this digital filter is defined by the following linear constant-coefficient difference equation:

[0242]

[0243] Among them, coefficient and The design parameters of the 7th-order low-pass Butterworth filter are uniquely determined.

[0244] Sampling period Based on the physiological characteristics of human pulse wave signals, a cutoff frequency is set. This frequency range effectively preserves heart rate and respiratory harmonic components while suppressing high-frequency noise.

[0245] The original input signal sequence Substituting into the aforementioned difference equation, the output signal sequence after filtering out high-frequency noise can be obtained through recursive calculation. .

[0246] S202. Use cubic spline interpolation to process the filtered pulse wave signal. To perform baseline removal, the signal is first detected. For all trough points, construct a cubic spline interpolation function using the trough points as interpolation nodes. Then, the baseline estimate is calculated using a cubic spline function:

[0247]

[0248] Subtracting the estimated baseline from the original signal yields the baseline-removed signal:

[0249]

[0250] S203. The pulse wave signal after baseline drift elimination is segmented using the sliding window method. The 1-minute pulse wave signal is divided into 10-second windows with a step size of 5 seconds.

[0251] S204. The quality of the segmented pulse wave signal is evaluated using the attractor reconstruction method. The processed signal is as follows: Figure 4 As shown.

[0252] S3. Based on the preprocessed effective pulse wave signal, feature parameters related to cardiovascular health status are extracted. First, the starting point and peak point are detected using the method in S301. The detection results are as follows: Figure 5 As shown.

[0253] The peak systolic blood pressure was then extracted based on the morphological characteristics of the pulse wave. Peak diastolic blood pressure Double notch Pulse interval Augmentation Index Substitution Enhancement Index The ratio of the second beat notch to the peak contraction Negative relative growth index Peak duration of systole Double notch time diastolic peak time Features such as...

[0254] Then, based on the time domain, frequency domain, and nonlinear domain characteristics of PRV, the mean PP interval is extracted. ), standard deviation of normal PP interval ( ), root mean square of the difference between adjacent PP periods ( The ratio of intervals with a difference greater than 20ms between adjacent PP intervals to the total number of intervals. ), low frequency ( ),high frequency( ), the ratio of low-frequency power to high-frequency power ( ), sample entropy ( ), multi-scale entropy ( Approximate entropy ( Features such as )

[0255] S4. The extracted cardiovascular health-related feature parameters were weighted using the CRITIC method to obtain the intervention effect score corresponding to each test score. The test score number with the highest score was used as the sample label for the corresponding subject. The specific steps are as follows:

[0256] S401. For each subject, calculate the rate of change of each characteristic value relative to the pre-performance resting baseline after performing each set of test scores. Different formulas for calculating the rate of change are used based on the correlation between the characteristics and cardiovascular health:

[0257] If the characteristic is positively correlated with cardiovascular health, an increase in the characteristic value after performance is considered beneficial, and the formula for calculating the rate of change is:

[0258]

[0259] If a characteristic is negatively correlated with cardiovascular health, a decrease in characteristic value after performance is considered beneficial, and the formula for calculating the rate of change is:

[0260]

[0261] If the characteristic baseline value is zero or close to zero, then the absolute change is used:

[0262]

[0263] S402. Construct an evaluation matrix based on the characteristic change rate data of all subjects across all musical scores. ,in For the number of musical scores, The number of features is used. The matrix is ​​standardized by range to eliminate the influence of dimensions.

[0264] Calculate the standard deviation of each feature after standardization. and the Pearson correlation coefficient matrix between features According to the CRITIC method, the information content of each feature is calculated. :

[0265]

[0266] Objective weights of each feature The weight vector is ultimately determined by the proportion of its information content to the total information content. .

[0267] S403, The characteristic rate of change vector after each subject plays each set of musical scores. The scores are then weighted and summed to obtain the overall intervention effect score for the musical score.

[0268]

[0269] The test scores of the subject are traversed, and the score with the highest score is selected as the subject's personalized best training score. The unique number of the score is used as the subject's sample label for subsequent machine learning model training.

[0270] S5. Construct a machine learning model, using the subject's resting baseline pulse wave signal as the model input feature and the highest-rated score label determined in S4 as the model output target. Train the model and provide personalized training score recommendations. The specific steps are as follows:

[0271] S501. Construct the model training dataset: Integrate the data of all subjects, and use the standardized morphological features and pulse rate variability (PRV) features obtained by processing the resting baseline pulse wave signal of each subject through steps S2 and S3 as the input feature vector of the model; use the unique label corresponding to the best intervention effect score of the subject determined in step S4 as the output target label of the model to form a structured "feature-label" dataset.

[0272] S502. A subject-based partitioning strategy is adopted, randomly dividing all subjects into training set subjects and validation set subjects. Specifically, the ratio of training set to validation set is first determined to be 7:3; then, all subjects are numbered and randomly sorted to ensure the randomness and unbiasedness of the partitioning process. The partitioning strictly adheres to the "subject independence" principle, meaning that all data samples of the same subject (including their resting baseline characteristics and corresponding best musical score labels) must belong entirely to either the training set or the validation set, and cannot span between the two sets. This strategy effectively avoids model overfitting and evaluation bias caused by the repeated occurrence of the same subject's physiological characteristics during the training and validation phases, thus more realistically reflecting the model's classification generalization ability when facing entirely new subjects, and improving the practical applicability and robustness of the recommendation system.

[0273] S503. Train a random forest classification model using the training set data, and use grid search combined with five-fold cross-validation to systematically tune the key hyperparameters of the model. Use the cross-validation accuracy as the evaluation index to select the optimal parameter combination.

[0274] The training set data is randomly divided into 5 mutually exclusive subsets. Each time, 4 subsets are selected as training data, and the remaining subset is used as validation data. This process is repeated 5 times to ensure that each subset is used as the validation set at least once. For each hyperparameter combination, the average of the 5 validation accuracies is calculated as the cross-validation accuracy for that parameter combination.

[0275] Finally, the hyperparameter combination with the highest cross-validation accuracy was selected as the optimal model parameters, and the random forest classifier was retrained based on the full training set data to obtain the final optimized personalized music score recommendation model. This process effectively avoids model overfitting and improves the predictive stability and classification accuracy for new subjects.

[0276] S504. For new subjects, pulse wave signals are collected at rest. After preprocessing and feature extraction in steps S2 and S3, standardized morphological features and pulse rate variability feature vectors are obtained and used as input to the model. The model automatically predicts and outputs the corresponding optimal musical notation based on these input features, realizing intelligent recommendation of personalized playing training scores.

[0277] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0278] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A recommended method for evaluating the intervention effect of sheet music playing training on cardiovascular health, characterized in that, Includes the following steps: S1. Have the subject sit quietly in a quiet environment until their physiological state is stable, and collect their pulse wave signal at rest as a baseline reference signal; then guide the subject to play multiple sets of preset test scores in sequence, and record the behavioral data during the performance of each set of scores, as well as the pulse wave signal before and after the performance. A recovery period is set between each set of scores to eliminate cross-interference. S2. Perform signal denoising, baseline correction, data segmentation and truncation and signal quality assessment on all collected pulse wave signals in sequence, remove signals with substandard quality and obtain effective pulse wave signals; S3. Based on the preprocessed effective pulse wave signal, extract feature parameters related to cardiovascular health status, including pulse wave morphological features and pulse rate variability features. S4. The extracted feature parameters are weighted using a weighting method based on the correlation of indicators to obtain the intervention effect score corresponding to each test score. The test score with the highest score is used as the sample label of the subject. S5. Construct a machine learning model, using the feature parameters corresponding to the resting baseline pulse wave signal of the subject as the model input, and the highest score musical score label as the model output target. After training the machine learning model, personalized training musical score recommendations are realized for new subjects.

2. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 1, characterized in that, The specific process of step S1 is as follows: S101. Classify the sheet music according to the number of beats per minute, playing dynamics, playing air pressure range, and breathing frequency to form multiple sets of test sheet music; S102. After the subject sat quietly with his / her eyes closed, the pulse wave signal in the resting state was collected by a photoplethysmography sensor as a baseline reference signal. S103. After the subject sits still to eliminate cross-interference, the pulse wave signal is collected before the performance. Then, the subject plays the designated score using a wind instrument. The pulse wave signal is collected immediately after the performance. This process is repeated until all test scores are played.

3. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 1, characterized in that, The specific process of step S2 is as follows: S201. A multi-order low-pass Butterworth filter is used to filter the pulse wave signal to remove high-frequency interference noise. S202. Use cubic spline interpolation to perform baseline correction on the filtered signal to remove baseline drift interference. S203. The baseline-removed signal is segmented using the sliding window method, and a uniform window size and step size are set. S204. The attractor reconstruction method is used to evaluate the quality of each segmented signal and select the effective signal segments that meet the preset quality threshold.

4. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 3, characterized in that, Step S204 is as follows: First, based on the segmented pulse wave fragments, a correlation signal is constructed by setting a time delay, and then a new variable is obtained through projection transformation. The specific variables are used as the horizontal and vertical coordinates respectively to form an image of the reconstructed attractor. Next, three key feature points are identified in the reconstructed image to construct an ideal triangle, and the discrete lines corresponding to the three sides of the triangle are obtained. Then, the trajectory of the attractor is divided into corresponding segments according to the three feature points mentioned above. The error between each trajectory segment and the corresponding side of the triangle is calculated. The maximum value of the error of all trajectory segments on each side is taken, and the three maximum values ​​are combined to obtain the global error of the attractor. Finally, a quality threshold is set. If the global error does not exceed the threshold, the quality of the pulse wave signal segment is deemed to meet the standard; if it exceeds the threshold, it is considered to be substandard and is removed.

5. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 1, characterized in that, The specific process of step S3 is as follows: S301. Detect the main wave peak of the screened pulse wave signal, and use the identified main wave peak as the reference point to locate a single complete pulse wave cycle. S302. Based on a single complete pulse wave cycle after localization, extract the pulse wave morphological features within that cycle; take the average of the morphological features of all valid pulse wave cycles as the standardized morphological features of the corresponding samples. S303. Based on the pulse wave signal after peak detection, calculate the pulse rate variability characteristics of the signal, wherein the pulse rate variability characteristics cover time domain characteristics, frequency domain characteristics and nonlinear domain characteristics.

6. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 5, characterized in that, In step S301, a first-order derivative peak detection algorithm with adaptive threshold is used to detect the main peak of the pulse wave signal, specifically as follows: First, the discrete signal is divided into multiple segments of equal length. A time window of a specific length is set at the beginning of each segment. The signal amplitude characteristics and instantaneous heart rate are estimated by finding local maxima, and the initial amplitude threshold and time interval threshold are determined. These two thresholds are dynamically adjusted according to the subsequently detected heartbeat characteristics. Next, the entire discrete sequence is scanned, and zero-crossing points are found in the difference sequence. The amplitude of the point is then checked in the original signal to see if it exceeds the current amplitude threshold. At the same time, it is confirmed that the interval between the point and the previous accepted peak point meets the time interval threshold requirement. Only points that meet both of these conditions will be determined as contraction peak points. Finally, for each pair of adjacent peak points, the signal segment between them is extracted, and the minimum point in the signal segment is located. This minimum point is used as the starting point of the corresponding pulse wave cycle. This starting point is usually located at the lowest point of each cycle, corresponding to the starting position of the signal waveform.

7. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 5, characterized in that, In step S303, the time-domain features include: Mean PP interval: It reflects the overall rhythm of the pulse wave by calculating the average of all pulse intervals; Normal PP interval standard deviation: By calculating the standard deviation of all pulse intervals, it reflects the overall variability of pulse interval over a longer time scale; Root mean square of the difference between adjacent pulse intervals: By calculating the root mean square of the difference between adjacent pulse intervals, it reflects the short-term fluctuation characteristics and is mainly used to assess the level of parasympathetic nerve activity. Percentage of adjacent pulse intervals with a difference greater than 20ms: Statistical analysis of the percentage of intervals with a difference greater than 20ms between adjacent pulse intervals out of the total number of intervals; The frequency domain features include: Low-frequency characteristic values: correspond to a specific frequency range, reflecting the combined regulatory function of the sympathetic and vagus nerves, and are related to the activity of the sympathetic nervous system; High-frequency characteristics: corresponding to another specific frequency range, reflecting the regulatory activity of the vagus nerve, and related to respiratory arrhythmias; The ratio of low-frequency to high-frequency power reflects the balance between the sympathetic nervous system and the vagus nervous system, and is an indicator of the overall balance of the autonomic nervous system. The nonlinear domain features include: Sample entropy: It quantifies the complexity of a signal by comparing the similarity probabilities of templates of different lengths in a time series. It has low dependence on data length, is computationally stable, and the larger the value, the more complex the signal. Multiscale entropy: As an extension of sample entropy, it calculates and integrates the sample entropy at each scale by transforming the original sequence at different scales, reflecting the overall trend over a long time scale and the detailed features of a short time series. Approximate entropy: It quantifies the regularity and complexity of a signal by statistically analyzing the probability of new patterns appearing in a sequence. The larger the value, the more difficult the signal is to predict and the weaker its regularity.

8. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 1, characterized in that, The specific process of step S4 is as follows: S401. Based on the extracted standardized pulse wave morphological features and pulse rate variability features, calculate the rate of change of the above features relative to the resting baseline before performance after the subject plays each set of test scores. S402. Based on the multi-dimensional feature change rate dataset, the weight of each feature parameter is calculated using a weight determination method based on indicator correlation. S403. Based on the determined objective weights of each feature, the multidimensional feature change rates generated by each subject after playing each set of test scores are weighted and summed to calculate the comprehensive intervention effect score corresponding to the score. Finally, the test score with the highest comprehensive intervention effect score is selected for each subject, and the unique label of the score is used as the subject's personalized training sample label.

9. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 8, characterized in that, In step S402, the weights of each feature parameter are calculated using a weight determination method based on indicator correlation, specifically as follows: First, an evaluation matrix is ​​constructed with the subjects playing different musical scores as rows and the rate of change of each feature as columns; Secondly, the evaluation matrix is ​​standardized to eliminate the differences in the dimensions of each feature; Next, the correlation coefficient matrix between each feature is calculated to quantify the conflict or contrast strength between features. Then, by combining the standard deviation of the standardized features with the conflict between features, the amount of information contained in each feature is calculated; Finally, based on the proportion of each feature's information content to the total information content, the final objective weight of each feature parameter is determined.

10. The method for evaluating and recommending the effect of musical score playing training on cardiovascular health intervention according to claim 1, characterized in that, Step S5 is as follows: S501. Integrate the data of all subjects, and use the standardized morphological features and pulse rate variability features obtained by processing the resting baseline pulse wave signal of each subject through steps S2 and S3 as the input feature vector of the machine learning model, and use the corresponding highest score musical score label as the output target label to form a structured "feature-label" dataset. S502. Divide the training set and validation set according to the subjects to ensure that the data of the same subject belongs to only one set; S503. Train a random forest classification model using the training set data, and fine-tune the hyperparameters using grid search combined with five-fold cross-validation. Select the parameter combination with the highest cross-validation accuracy to determine the optimal model. S504. Collect the resting pulse wave signal of new subjects, and after preprocessing and feature extraction, input it into the optimal model to output the corresponding best musical score number, thereby realizing personalized recommendations.

Images