Intelligent tanning electronic chest strap used after cardiothoracic surgery based on artificial intelligence

Abstract

The invention relates to the technical field of medical instruments, in particular to a tanning intelligent electronic chest strap used after a cardiothoracic surgery based on artificial intelligence, and a method comprises the following steps: based on a tanning intelligent electronic material, a sensor is embedded in the chest strap, and a sensor array detects biological data such as electrocardio, heartbeat, respiration and body temperature of a patient in real time; the embedded sensor array collects the biological signals in real time, and transmits the signals of the sensor sequence to the AI intelligent analysis platform through the Bluetooth device. The intelligent chest support has the advantages of high quality, comprehensive monitoring and intelligent adjustment, the tanning electronic material, namely the intelligent material with high flexibility, high air permeability and biodegradation resistance, is adopted, the comfort and safety of long-time wearing are ensured, the material has the self-adaptive adjustment characteristic, dynamic adjustment can be performed according to the postoperative recovery condition of a patient, the optimal chest support is ensured, and the intelligent chest support is suitable for popularization and application. The wearing comfort is improved, skin allergy and compression injury are reduced, the biocompatibility is enhanced, and the intelligent monitoring device is suitable for being worn after long-term postoperative rehabilitation and intelligently monitoring various vital sign changes after the operation.

Description

Technical Field

[0001] The present invention relates to the technical field of medical devices, and in particular to an artificial intelligence-based electronic chest belt for use after thoracic and cardiovascular surgery. Background Art

[0002] Medical devices are instruments, equipment, apparatus, in vitro diagnostic reagents and calibrators, materials, and other similar or related items, including required computer software, intended for direct or indirect use on the human body. Medical devices include both medical equipment and medical consumables that are primarily obtained through physical means, rather than pharmacological, immunological, or metabolic methods, or where these methods are involved but only in a supporting role.

[0003] Patients who undergo thoracic and cardiovascular surgery (including open-chest surgery and minimally invasive surgery) often face problems such as postoperative pain, limited respiratory function, difficulty expectoration, and reduced chest stability. First, traditional chest belts are made of relatively rigid materials and have poor breathability. Long-term wearing can easily lead to skin discomfort, compression injuries, and allergic reactions. They cannot adapt to the patient's postoperative chest changes and have limited fixation effects. Second, traditional chest belts only have mechanical fixation functions and cannot monitor the patient's postoperative biological signals. Existing single monitoring devices usually require additional electrodes or external devices, which is inconvenient for patients to wear daily after surgery. Finally, traditional chest belts cannot be intelligently adjusted according to the postoperative recovery of different patients. Manual adjustment of the fixation strength is required, which affects the rehabilitation effect. Postoperative complications (such as hypoventilation, sternal instability, and atelectasis) are difficult to warn early, and patients need to seek medical attention frequently.

[0004] Therefore, there is an urgent need for an artificial intelligence-based electronic chest belt for use after thoracic and cardiovascular surgery to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide an artificial intelligence-based electronic chest belt for use after thoracic and cardiovascular surgery, which has the advantages of high-quality materials, comprehensive monitoring, and intelligent adjustment, and solves the problems raised by the above-mentioned background technology.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solution: an artificial intelligence-based electronic chest belt for use after thoracic and cardiovascular surgery, wherein the method comprises the following steps:

[0007] S1: The sensor array embedded in the chest strap detects the patient's respiratory mechanics and vital signs in real time; the embedded sensor array collects 30-60 seconds of signals in real time, and transmits the sensor sequence signals to the AI ​​intelligent analysis platform through the Bluetooth device. The AI ​​data processing platform system comprehensively analyzes the wearer's physical indicators and obtains a real-time wearer's smart health report. The doctor monitors the wearer's physical condition by viewing the wearer's smart health report online, and the wearer can also view his or her own smart health report on the platform; the chest strap embedded sensor includes embedded biosensor monitoring (tannic pressure sensor, tannic stress sensor, respiratory detection sensor, tannic The AI ​​intelligent processing module has different processing strategies and methods for the signals transmitted by different sensors. For the sweat sensor detector, the sweat buffer layer in the tanned leather effectively completes the enrichment and discharge of sweat, and realizes the characteristic index detection of sweat sodium, potassium, chloride ions and pH value, and uploads the various ion indicators in sweat to the AI ​​intelligent analysis platform, and records the various indicators as clinical indicators for doctors' reference; the tanned ECG monitoring device uses functional leather to achieve better ECG (ECG) signal acquisition than commercial hydrogel electrodes, among which the heart rhythm, heart rate and ECG are detected. Figure 3 ECG signals, these three signals are analyzed by the ECG signal classification feedback module, and because the breathing amplitude and coughing amplitude will show the wearer's chest movement, the pressure sensor of the strap detects the coughing strength and breathing amplitude, and rapid breathing will cause changes in heart rate, so the AI ​​module will also judge the wearer's breathing condition based on the ECG signal, and perform mechanical assisted breathing, thermal insulation perception and intervention treatment on the patient based on the above signal AI analysis.

[0008] S2: The real-time signal received by the sensor is transmitted to the specially designed ECG-ai monitoring module.

[0009] S3: The ECG signal denoising module is designed to address the ECG signal contamination problem caused by the trunk activities of the smart harness wearer in daily life.

[0010] S4: The wavelet transform convolution block uses two sets of convolutional layers instead of traditional filters. The input ECG signal is first calculated by the first convolution layer. Subsequently, another set of convolutional layers with exactly the same structure is orthogonalized by inverting the convolution kernel and taking the interval negation. In addition, large convolution kernels are generated by inserting zeros into small convolution kernels, similar to the effect of dilated convolution. Only one set of convolution kernels in the entire wavelet transform convolution block needs to be trained, and the rest of the convolution kernels are generated by this set of convolution kernels, which greatly reduces the number of convolution parameters.

[0011] S5: Obtain the denoised ECG signal and use the R-peak location method based on Shannon energy to determine the R-peak position in the signal. By detecting the R-wave peak, determine the start and end times of the heartbeat and calculate the heart rate. If the heart rate is higher than the standard value, the strap will sound an alarm to notify the doctor. At the same time, by analyzing the changes in the RR interval, the respiratory rate is calculated.

[0012] S6: With the R peak as the center, the signal is divided into data segments with a length of 360, and the R peak interval is added as auxiliary information. Then, the three consecutive signal segments are used as inputs of the classification module respectively, ensuring that each input contains a complete heartbeat cycle and fully utilizes the contextual information of the ECG signal.

[0013] S7: The input of the rhythm classification feedback module is three consecutive ECG segments centered on the R peak: input1, input2, and input3. First, the CNN block is used to perform preliminary feature extraction on the signal to shorten the signal length. Next, a bidirectional Transformer (BiTrans) is used to capture the temporal relationship in the features and use the information of the waveforms of the previous and next ECG segments to learn the temporal dependency between the signals. An inversion mechanism is introduced in the feature map to generate an additional input containing an inverse sequence of spatial information. This enables the model to consider the contextual information of the previous and next ECG segments when predicting a specific position. By combining the outputs of these two directions, richer contextual understanding is achieved, effectively capturing the complex relationships in the input sequence. The three input signals are linearly projected into the query, key, and value matrices of each head:

[0014]

[0015] in, and is the query, key, and value matrix of the h-th head. Similarly, and is the learnable weight parameter matrix of the h-th head, and then the self-attention mechanism of each head is applied as follows:

[0016]

[0017] where d k for The multi-head attention mechanism is obtained by concatenating the operation results of all h heads and applying another linear mapping to obtain the final multi-head attention output

[0018] MHA(f input1 , f input2 , f input3 )=Concat(A 1 ,..,AH )·W o (3)

[0020] Among them, MHA(f input1 , f input2 , f input3 ) represents the multi-head self-attention output, W O The weight parameter matrix representing the output of the multi-head attention is then inverted to consider both past and future tokens, implementing a bidirectional Transformer as follows:

[0021] f rev [i,j]=x[i,dj-1] (4)

[0023] Among them, the range of i is 0, 1, ..., N-1, the range of j is 0, 1, ..., d-1;

[0024] The output of the bidirectional Transformer layer is as follows:

[0025]

[0026] The encoder is composed of a stack of identical layers, each with two sublayers, namely the multi-head self-attention mechanism and the fully connected feed-forward network, which processes the input data f in sequence. That is, the i-th layer includes the operations expressed by equations (5) to (9), capturing increasingly abstract and contextual representations;

[0027] The output of each layer in the bidirectional Transformer is used as the input of the next layer, so that complex patterns can be gradually extracted. The final output of the bidirectional Transformer network is represented as BiTrans(f);

[0028] BiTrans(f)=Layer L (Layer L-1 (...(Layer1(f)…)) (6)

[0030] S8: After the bidirectional Transformer output, the obtained feature vector is put into the classifier for classification. Because the cardiac cycle waveform lacks information about the R peak interval, which plays an important role in determining the ECG type, the four R peak intervals closest to the middle R peak are normalized into features, concatenated with the features extracted in the cardiac cycle, and input into the fully connected layer. The final classification result is output through the softmax function, and the ECG data is classified into five common heart rhythms: normal sinus rhythm, atrial tachycardia, bigeminy, tripeminy, and ventricular premature beats.

[0031] S9: The loss function used by the denoising module needs to retain the peak values ​​of different ECG waveforms as much as possible during the denoising process. Therefore, a peak coefficient is added to the traditional covariance MSE loss function to give higher weights to points closer to the peak value. The peak coefficient C p And the improved MSE loss function L d As shown below:

[0032] C p =1+abs(X c -median(X c )) (7)

[0034]

[0035] Where, ° represents the element-by-element multiplication of the matrix, N represents the signal length, X c Indicates a clean signal, Xd

[0036] represents the denoised signal.

[0037] S10: The loss function of the classification module is to deal with the problem of imbalanced ECG signal sample categories. The focus loss function is used, and its formula is as follows

[0038] L c =-a(1-y p ) γ log(y p ) (9)

[0040] Among them, y p is the probability of correct prediction, γ is (1-y p ) enhancement factor, the more difficult the sample is to distinguish, (1-y p )γ is larger, α is the weight of small samples;

[0041] If the classification module detects abnormal ECG, heartbeat, respiration and body temperature signals, it will issue an alarm and notify the doctor in time.

[0042] Furthermore, as a preferred embodiment of the present invention, in step S2, the ECG-ai monitoring module consists of two parts: an ECG signal denoising module and a heart rhythm classification feedback module.

[0043] Furthermore, as a preferred embodiment of the present invention, in step S3, the ECG signal denoising module adopts a design similar to a U-shaped network structure, and introduces a stationary wavelet transform of a convolution structure in the denoising process, which is called a wavelet transform convolution block.

[0044] Furthermore, as a preferred embodiment of the present invention, in step S5, if the patient's breathing rate is abnormal, the smart harness will automatically strengthen the expectoration mode and issue an alarm to notify the doctor.

[0045] Furthermore, as a preferred embodiment of the present invention, in step S7, input1, input2 and input3 are provided by a denoising module, which utilizes the waveform correlation of each cardiac cycle and combines the contextual information between cardiac cycles to improve the accuracy of disease diagnosis.

[0046] Beneficial effects: The technical solution of this application has the following technical effects: the present invention has the advantages of high-quality materials, comprehensive monitoring, and intelligent adjustment. It adopts tanned electronic materials, that is, highly flexible, highly breathable, and biodegradable intelligent materials to ensure comfort and safety for long-term wearing. The material has adaptive adjustment characteristics and can be dynamically adjusted according to the patient's postoperative recovery to ensure optimal chest support, improve wearing comfort, reduce skin allergies and compression injuries, enhance biocompatibility, and is suitable for long-term rehabilitation wear. It intelligently adapts to postoperative changes and provides support that better meets the patient's needs; secondly, a flexible biosensor is integrated inside the chest strap, which can non-invasively measure key physiological parameters such as respiratory rate, chest expansion, heart rate, blood oxygen, and skin temperature. A flexible strain sensor is used to detect the patient's chest movement in real time, analyze postoperative recovery, and transmit data wirelessly. It can be synchronized to the doctor's or patient's app for remote monitoring, real-time monitoring of postoperative physiological indicators, and improved accuracy of postoperative management. The flexible design fits comfortably on the skin, reducing the discomfort of traditional electrodes. The wireless data connection allows doctors to remotely monitor the patient's recovery status. Finally, the AI ​​algorithm combines the patient's real-time physiological data (such as breathing depth and chest expansion) to automatically adjust the support strength of the chest strap to optimize the rehabilitation experience. Based on big data analysis, AI identifies postoperative abnormalities (such as hypoventilation, pneumothorax, and postoperative atelectasis) and issues an alert to the doctor or patient. Based on the patient's recovery progress, AI analyzes the above signals and implements mechanical assisted breathing, thermal insulation perception, and intervention treatment. AI can recommend different wearing modes (such as nighttime tightness mode and exercise mode) to improve rehabilitation efficiency. It intelligently adapts to the needs of different patients and reduces the tediousness of manual adjustments. Postoperative risk warnings improve patient safety, personalized recovery management, and shorten the rehabilitation cycle. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0048] Figure 1 is a process flow chart of the leather of the present invention;

[0049] Figure 2 This is a diagram of the preparation process of the electrocardiogram and pressure sensor device of the present invention;

[0050] Figure 3 A diagram showing the process of preparing a sample of the heating device of the present invention;

[0051] Figure 4 This is a block diagram of the structure and working mechanism of the ECG signal processing module of the AI ​​algorithm of the strap of the present invention;

[0052] Figure 5 This is a schematic diagram of the electronic chest strap of the present invention. DETAILED DESCRIPTION

[0053] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. In order to better understand the technical content of the present invention, specific embodiments are cited and explained in conjunction with the drawings as follows. Various aspects of the present invention are described in this disclosure with reference to the drawings, which show many illustrative embodiments. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0054] Example 1:

[0055] Tanning electronic materials: Leather process (see attached Figure 1 )

[0056] Based on the porous structure, flexibility and good biocompatibility of leather, functional materials are added to the leather's fiber network to give it good electrical properties while retaining the leather's original physical properties. According to different functional requirements, the materials and reaction conditions are adjusted, and different coating layers, connection circuits and back-end systems are combined to achieve functions such as heating, physiological signal acquisition, and sensing.

[0057] Flexible sensor system: flexible pressure sensor, flexible stress sensor, breathing detection sensor, flexible thermal insulation device, flexible ECG monitoring, sweat sensor detector, etc. (see attached Figure 2 ); the chest cavity breathing movement amplitude and the chest cavity amplitude change during coughing are sensed by the stress sensor; the pressure sensor in the precordial area directly senses the strength of the heartbeat movement.

[0058] Tanzhi conductive leather is selected to prepare ECG dry electrodes, and the Tanzhi conductive leather is combined with a chest strap. Taking advantage of the leather's cuttability and sewability, the two are sewn together to obtain a Tanzhi leather ECG strap. At the same time, Tanzhi conductive leather cut to set specifications is sewn together with a chest strap and connected to a power supply through a wire to prepare a leather heating chest strap.

[0059] First-layer suede sheepskin is selected and cut into samples of a certain size. Conductive ink is printed onto the leather using an inkjet printer, thereby drawing an interdigitated electrode array on the leather surface. The array is then combined with tanned conductive leather material to prepare a flexible pressure sensor.

[0060] The sensing electrode patches prepared from Tanzhi electronic materials are scientifically and rationally arranged according to specific needs, and are used in the monitoring of physiological signals such as electrocardiogram, heartbeat, respiration, body temperature, myoelectricity, pressure, and thermal therapy. Combining the device with a chest strap can accurately capture the human body's electrocardiogram, heartbeat, respiration, body temperature, and myoelectricity activities. The leather electrode patches used not only have excellent high sensitivity and outstanding biocompatibility, but can also be seamlessly integrated with the chest strap, thereby significantly improving the monitoring accuracy while greatly enhancing the convenience and comfort of wearing.

[0061] The structure and working mechanism of the ECG signal processing module of the strap AI algorithm (algorithms of other sensors) (see Example 2 and Appendix Figure 3 )

[0062] Example 2: An artificial intelligence-based electronic chest belt for use after cardiothoracic surgery, the method comprising the following steps:

[0063] S1: The sensor array embedded in the chest strap detects the patient's respiratory mechanics and vital signs in real time; the embedded sensor array collects 30-60 seconds of signals in real time, and transmits the sensor sequence signals to the AI ​​intelligent analysis platform through the Bluetooth device. The AI ​​data processing platform system comprehensively analyzes the wearer's physical indicators and obtains a real-time wearer's smart health report. The doctor monitors the wearer's physical condition by viewing the wearer's smart health report online, and the wearer can also view his or her own smart health report on the platform; the chest strap embedded sensor includes embedded biosensor monitoring (tannic pressure sensor, tannic stress sensor, respiratory detection sensor, tannic The AI ​​intelligent processing module has different processing strategies and methods for the signals transmitted by different sensors. For the sweat sensor detector, the sweat buffer layer in the tanned leather effectively completes the enrichment and discharge of sweat, and realizes the characteristic index detection of sweat sodium, potassium, chloride ions and pH value, and uploads the various ion indicators in sweat to the AI ​​intelligent analysis platform, and records the various indicators as clinical indicators for doctors' reference; the tanned ECG monitoring device uses functional leather to achieve better ECG (ECG) signal acquisition than commercial hydrogel electrodes, among which the heart rhythm, heart rate and ECG are detected. Figure 3ECG signals, these three signals are analyzed by the ECG signal classification feedback module, and because the breathing amplitude and coughing amplitude will show the wearer's chest movement, the pressure sensor of the strap detects the coughing strength and breathing amplitude. At the same time, rapid breathing will cause changes in heart rate, so the AI ​​module will also judge the wearer's breathing condition based on the ECG signal. Based on the above signals, AI analysis will perform mechanical assisted breathing, thermal insulation perception and intervention treatment on the patient.

[0064] S2: The real-time signal received by the sensor is transmitted to the specially designed ECG-ai monitoring module.

[0065] Furthermore, in step S2, the ECG-ai monitoring module consists of two parts: an ECG signal denoising module and a heart rhythm classification feedback module.

[0066] S3: The ECG signal denoising module is designed to address the ECG signal contamination problem caused by the torso activities of the smart harness wearer in daily life.

[0067] Furthermore, in step S3, the ECG signal denoising module adopts a design similar to a U-shaped network structure, and introduces a stationary wavelet transform of a convolutional structure in the denoising process, which is called a wavelet transform convolution block.

[0068] S4: The wavelet transform convolution block uses two sets of convolutional layers instead of traditional filters. The input ECG signal is first calculated by the first convolution layer. Subsequently, another set of convolutional layers with exactly the same structure is orthogonalized by inverting the convolution kernel and taking the interval negation. In addition, large convolution kernels are generated by inserting zeros into small convolution kernels, similar to the effect of dilated convolution. Only one set of convolution kernels in the entire wavelet transform convolution block needs to be trained, and the rest of the convolution kernels are generated by this set of convolution kernels, which greatly reduces the number of convolution parameters.

[0069] S5: Obtain the denoised ECG signal and use the R-peak location method based on Shannon energy to determine the R-peak position in the signal. By detecting the R-wave peak, determine the start and end times of the heartbeat and calculate the heart rate. If the heart rate is higher than the standard value, the strap will sound an alarm to notify the doctor. At the same time, by analyzing the changes in the RR interval, the respiratory rate is calculated.

[0070] Furthermore, in step S5, if the patient's breathing rate is abnormal, the smart harness will automatically enhance the expectoration mode and sound an alarm to notify the doctor.

[0071] S6: With the R peak as the center, the signal is divided into data segments with a length of 360, and the R peak interval is added as auxiliary information. Then, the three consecutive signal segments are used as inputs of the classification module respectively, ensuring that each input contains a complete heartbeat cycle and fully utilizes the contextual information of the ECG signal.

[0072] S7: The input of the rhythm classification feedback module is three consecutive ECG segments centered on the R peak: input1, input2, and input3. First, the CNN block is used to perform preliminary feature extraction on the signal to shorten the signal length. Next, a bidirectional Transformer (BiTrans) is used to capture the temporal relationship in the features and use the information of the waveforms of the previous and next ECG segments to learn the temporal dependency between the signals. An inversion mechanism is introduced in the feature map to generate an additional input containing an inverse sequence of spatial information. This enables the model to consider the contextual information of the previous and next ECG segments when predicting a specific position. By combining the outputs of these two directions, richer contextual understanding is achieved, effectively capturing the complex relationships in the input sequence. The three input signals are linearly projected into the query, key, and value matrices of each head:

[0073]

[0074] in, and is the query, key, and value matrix of the h-th head. Similarly, and is the learnable weight parameter matrix of the h-th head, and then the self-attention mechanism of each head is applied as follows:

[0075]

[0076] where d k for The multi-head attention mechanism is obtained by concatenating the results of all h heads and applying another linear mapping to obtain the final multi-head attention output.

[0077] MHA(f input1， f input2 , f input3 )=Concat(A 1 ...,A H )·W o (3)

[0079] Among them, MHA(f input1 , f input2 , f input3 ) represents the multi-head self-attention output, W O The weight parameter matrix representing the output of the multi-head attention is then inverted to consider both past and future tokens, implementing a bidirectional Transformer as follows:

[0080] f rev[i,j]=x[i,dj-1] (4)

[0082] Among them, the range of i is 0, 1, ..., N-1, the range of j is 0, 1, ..., d-1;

[0083] The output of the bidirectional Transformer layer is as follows:

[0084]

[0085] The encoder is composed of a stack of identical layers, each with two sublayers, namely the multi-head self-attention mechanism and the fully connected feed-forward network, which processes the input data f in sequence. That is, the i-th layer includes the operations expressed by equations (5) to (9), capturing increasingly abstract and contextual representations;

[0086] The output of each layer in the bidirectional Transformer is used as the input of the next layer, so that complex patterns can be gradually extracted. The final output of the bidirectional Transformer network is represented as BiTrans(f);

[0087] BiTrans(f)=Layer L (Layer L-1 (...(Layer1(f))...)) (6)

[0089] Furthermore, in step S7, input1, input2, and input3 are provided by the denoising module, which utilizes the waveform correlation of each cardiac cycle and combines the contextual information between cardiac cycles to improve the accuracy of disease diagnosis.

[0090] S8: After the bidirectional Transformer output, the obtained feature vector is put into the classifier for classification. Because the cardiac cycle waveform lacks information about the R peak interval, which plays an important role in determining the ECG type, the four R peak intervals closest to the middle R peak are normalized into features, concatenated with the features extracted in the cardiac cycle, and input into the fully connected layer. The final classification result is output through the softmax function, and the ECG data is classified into five common heart rhythms: normal sinus rhythm, atrial tachycardia, bigeminy, tripeminy, and ventricular premature beats.

[0091] S9: The loss function used by the denoising module needs to retain the peak values ​​of different ECG waveforms as much as possible during the denoising process. Therefore, a peak coefficient is added to the traditional covariance MSE loss function to give higher weights to points closer to the peak value. The peak coefficient C p And the improved MSE loss function L d As shown below:

[0092] C p =1+abs(X c -median(X c )) (7)

[0094]

[0095] Where, ° represents the element-by-element multiplication of the matrix, N represents the signal length, X c Indicates a clean signal, X d

[0096] represents the denoised signal.

[0097] S10: The loss function of the classification module is to deal with the problem of imbalanced ECG signal sample categories. The focus loss function is used, and its formula is as follows

[0098] L c =-a(1-y p ) γ log(y p ) (9)

[0100] Among them, y p is the probability of correct prediction, γ is (1-y p ), the more difficult the sample is to distinguish, the larger (1-yp)γ is, and α is the weight of the small sample;

[0101] If the classification module detects abnormal ECG, heartbeat, respiration, or body temperature signals, it will issue an alarm and notify the doctor in a timely manner.

[0102] Depend on Figure 5 It can be seen that

[0103] The structural design of the present invention includes: two pieces of tanned leather in the front and back + a tanning sensor, three elastic bands on each side, one side is fixed, and the other side is fastened to the leather with Velcro, and the two shoulder strap elastic bands are also fastened to the leather with Velcro. The elastic band is a mechanically retractable elastic band controlled by AI, which can be contracted and tightened according to various biological signs to match the patient's optimal breathing movement pattern.

[0104] Material selection: Tanzhi electronic materials are flexible, breathable, and conductive, ensuring comfort for long-term wear and reducing skin irritation. They are made using professional leather processing methods. Tanning materials are perceptive and mainly rely on elastic bands to receive AI signals to apply mechanics.

[0105] Biological data monitoring and remote health management: embedded biosensor monitoring (physical pressure sensor, physical stress sensor, respiratory detection sensor, physical thermal therapy device, physical ECG monitoring, sweat sensor detector, etc.) uploads physiological data to the cloud medical platform in real time; the amplitude of chest breathing movement and the changes in chest amplitude during coughing are perceived through stress sensors; the pressure sensor in the precordial area directly senses the strength of the heartbeat.

[0106] Intelligent AI adjustment algorithm: Real-time monitoring of physiological data related to the patient's respiratory rate, calculation through AI algorithms on the cloud medical platform, condition assessment based on the patient's physiological status, further analysis of the risk of postoperative complications, such as prediction of atelectasis and postoperative pain scores, and early adjustment of treatment plans.

[0107] It should be noted that, in this document, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0108] While the present invention has been disclosed above with reference to preferred embodiments, this is not intended to limit the present invention. Persons skilled in the art will readily appreciate that various modifications and variations can be made without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the claims.

Claims

1. The AI-based electronic chest belt for post-cardiothoracic surgery is characterized by: The method comprises the following steps: S1: The sensor array embedded in the chest strap detects the patient's ECG, heart rate, respiration, body temperature and other biological data in real time; the embedded sensor array collects the above biological signals in real time, and transmits the sensor sequence signals to the AI ​​intelligent analysis platform through the Bluetooth device. The AI ​​data processing platform system comprehensively analyzes the wearer's physical indicators and obtains a real-time smart health report of the wearer. The doctor monitors the wearer's physical condition by viewing the wearer's smart health report online, and the wearer can also view his or her own smart health report on the platform; the chest strap embedded sensors include embedded biosensor monitoring (electrocardiogram monitoring, heart rate pressure sensor, respiratory stress sensor, temperature receptor, sweat sensor detector); and the AI ​​intelligent processing module has different processing strategies and methods for the signals transmitted by different sensors. For the sweat sensor detector, it uses the smart leather The sweat buffer layer in the sweat monitor effectively completes the enrichment and discharge of sweat, realizes the characteristic index detection of sweat sodium, potassium, chloride and other ions and pH value, and uploads the various ion indicators in sweat to the AI ​​intelligent analysis platform, and records the various indicators as clinical indicators for doctors' reference; the tanned ECG monitoring device uses functional leather to achieve better ECG signal acquisition than commercial hydrogel electrodes. Among them, for the three ECG signals of heart rhythm, heart rate and electrocardiogram, these three signals are analyzed by the ECG signal classification feedback module. Since the breathing amplitude and coughing amplitude will show the wearer's chest movement, the coughing strength and breathing amplitude are detected by the pressure sensor of the strap. At the same time, rapid breathing will cause changes in heart rhythm. Therefore, the AI ​​module will also judge the wearer's breathing condition based on the ECG signal. According to the above signal AI analysis, mechanical assisted breathing, thermal insulation perception and intervention treatment are performed on the patient; S2: The real-time signals received by the sensor are transmitted to a specially designed biological AI monitoring module; S3: ECG signal denoising module is designed to address the ECG signal contamination problem caused by the trunk movements of the smart harness wearer in daily life; S4: The wavelet transform convolution block uses two sets of convolutional layers instead of traditional filters. The input ECG signal is first calculated by the first convolution layer. Then, another set of convolutional layers with exactly the same structure is orthogonalized by inverting the convolution kernel and taking the interval negation. In addition, large convolution kernels are generated by inserting zeros into small convolution kernels, similar to the effect of dilated convolution. Only one set of convolution kernels in the entire wavelet transform convolution block needs to be trained, and the rest of the convolution kernels are generated by this set of convolution kernels, which greatly reduces the number of convolution parameters. S5: Obtain the denoised ECG signal and use the R-peak location method based on Shannon energy to determine the R-peak position in the signal. By detecting the R-wave peak, the start and end times of the heartbeat are determined and the heart rate is calculated. If the heart rate is higher than the standard value, the strap will sound an alarm to notify the doctor. At the same time, by analyzing the changes in the R-R interval, the respiratory rate is calculated. S6: Split the signal into data segments of length 360°, centered around the R peak, and add the R peak interval as auxiliary information. Then, the three consecutive signal segments are used as inputs for the classification module, ensuring that each input contains a complete heartbeat cycle and fully utilizes the contextual information of the ECG signal. S7: The input of the rhythm classification feedback module is three consecutive ECG segments centered on the R peak: input1, input2, and input3. First, the CNN block is used to perform preliminary feature extraction on the signal to shorten the signal length. Next, a bidirectional Transformer (BiTrans) is used to capture the temporal relationship in the features and use the information of the waveforms of the previous and next ECG segments to learn the temporal dependency between the signals. An inversion mechanism is introduced in the feature map to generate an additional input containing an inverse sequence of spatial information. This enables the model to consider the contextual information of the previous and next ECG segments when predicting a specific position. By combining the outputs of these two directions, richer contextual understanding is achieved, effectively capturing the complex relationships in the input sequence. The three input signals are linearly projected into the query, key, and value matrices of each head: in, and is the query, key, and value matrix of the h-th head. Similarly, and is the learnable weight parameter matrix of the h-th head, and then the self-attention mechanism of each head is applied as follows: where d k for The multi-head attention mechanism is obtained by concatenating the results of all h heads and applying another linear mapping to obtain the final multi-head attention output. MHA(f input1 ,f input2 ,f input3 )=Concat(A 1 ,..,A H )·W O (3) Among them, MHA(f input1 , f input2 , f input3 ) represents the multi-head self-attention output, W O The weight parameter matrix representing the output of the multi-head attention is then inverted to consider both past and future tokens, implementing a bidirectional Transformer as follows: f rev [i,j]=x[i,d-j-1] (4) Among them, the range of i is 0, 1, ..., N-1, the range of j is 0, 1, ..., d-1; The output of the bidirectional Transformer layer is as follows: The encoder is composed of a stack of identical layers, each with two sublayers, namely the multi-head self-attention mechanism and the fully connected feed-forward network, which processes the input data f in sequence. That is, the i-th layer includes the operations expressed by equations (5) to (9), capturing increasingly abstract and contextual representations; The output of each layer in the bidirectional Transformer is used as the input of the next layer, so that complex patterns can be gradually extracted. The final output of the bidirectional Transformer network is represented as BiTrans(f); BiTrans(f)=Layer L (Layer L-1 (...(Layer1(f))...))(6) S8: After the bidirectional Transformer output, the obtained feature vector is put into the classifier for classification. Because the cardiac cycle waveform lacks information about the R peak interval, which plays an important role in determining the ECG type, the four R peak intervals closest to the middle R peak are normalized as features, concatenated with the features extracted from the cardiac cycle, and input into the fully connected layer. The final classification result is output through the softmax function, and the ECG data is classified into five common rhythms: normal sinus rhythm, atrial tachycardia, bigeminy, tripeminy, and ventricular premature beats; S9: The loss function used by the denoising module needs to retain the peak values ​​of different ECG waveforms as much as possible during the denoising process. Therefore, a peak coefficient is added to the traditional covariance MSE loss function to give higher weights to points closer to the peak value. The peak coefficient C p And the improved MSE loss function L d As shown below: C p =1+abs(X c -median(X c ))(7) Among them, represents the element-by-element multiplication of the matrix, N represents the signal length, X c Indicates a clean signal, X d represents the denoised signal; S10: The loss function of the classification module is to deal with the problem of imbalanced ECG signal sample categories. The focus loss function is used, and its formula is as follows L c =-a(1-y p ) γ log(y p )(9) Among them, y p is the probability of correct prediction, γ is (1-y p ), the more difficult the sample is to distinguish, the larger (1-yp)γ is, and α is the weight of the small sample; If the classification module detects abnormal ECG, heartbeat, respiration and body temperature signals, it will issue an alarm and notify the doctor in time.

2. The artificial intelligence-based electronic chest belt for post-cardiothoracic surgery according to claim 1 is characterized in that: In step S2, the ECG-ai monitoring module consists of two parts: an ECG signal denoising module and a heart rhythm classification feedback module.

3. The AI-based electronic chest belt for post-cardiothoracic surgery according to claim 1, characterized in that: In step S3, the ECG signal denoising module adopts a design similar to a U-shaped network structure, and introduces a stationary wavelet transform of a convolution structure into the denoising process, which is called a wavelet transform convolution block.

4. The AI-based electronic chest belt for post-cardiothoracic surgery according to claim 1, characterized in that: In step S5, if the patient's breathing rate is abnormal, the smart harness will automatically enhance the expectoration mode and send out an alarm to notify the doctor.

5. The artificial intelligence-based electronic chest belt for post-cardiothoracic surgery according to claim 1, characterized in that: In step S7, input1, input2 and input3 are provided by the denoising module, which utilizes the waveform correlation of each cardiac cycle and combines the contextual information between cardiac cycles to improve the accuracy of disease diagnosis.

Images