A wearable borborygmus monitoring method based on double body out-of-reference

Abstract

The present application relates to a kind of wearable intestinal sound monitoring methods based on double body surface reference, the patch of first, second, third microphone is fixed on the body of the subject, the former two are located in the side of patch towards human body corresponding right lower abdominal ileocecal region and left upper abdominal stomach area, the third one is located away from human side.Three original audio signals are obtained and time synchronized, the former two are de-noised with the third microphone signal as environmental noise reference, and double-site bowel sound signal is obtained.The sequence of bowel sound events is extracted respectively, then merged, and the one with higher confidence is taken when time overlaps.Based on the merged event sequence, total event rate, gas-liquid sound event rate, high-pitched metallic sound event rate, average event duration, average event interval, frequency range and 24 hour trend slope are extracted as event stream features, then continuous risk classification results are calculated and displayed.The present application realizes efficient de-noising by double body surface microphone plus extracorporeal reference microphone, avoids omission by double-site monitoring, improves accuracy by event fusion, and provides quantitative evaluation tools by multi-dimensional features.

Description

Technical Field

[0001] This invention relates to the field of wearable bowel sound monitoring, and in particular to a wearable bowel sound monitoring method based on dual-body external reference. Background Technology

[0002] Postoperative intestinal obstruction is one of the most common complications in this group. It is caused by the combined effects of prolonged bed rest, opioid analgesia, anesthesia, and age-related degeneration of the enteric nervous system. Early identification and timely intervention of decreased gastrointestinal motility can significantly improve prognosis. Therefore, continuous monitoring of gastrointestinal function has important clinical value.

[0003] Bowel sound auscultation has been used as a bedside assessment method for over a century. Previous studies have shown that parameters such as bowel sound event rate and acoustic energy are significantly correlated with postoperative gastrointestinal function recovery and can predict food tolerance. Compared to imaging methods, which involve radiation exposure, scheduling difficulties, and invasive procedures, bowel sound monitoring is non-invasive, radiation-free, and can, in principle, be performed continuously at the bedside, making it a natural vehicle for continuous postoperative gastrointestinal monitoring.

[0004] However, routine bowel sound monitoring still has the following significant limitations in clinical application: 1. Monitoring frequency is extremely low and coverage is severely inadequate. Due to limited medical staff, routine bedside auscultation is only 2-4 minutes per subject per day, and more than 99% of the postoperative period is without monitoring; intestinal obstruction often progresses silently at night, and by the time the best intervention window is missed during the morning rounds the next day.

[0005] 2. Severe noise in the ward degrades signal quality. Conversations between family members, paging announcements, and alarms from infusion pumps and monitors are constant in wards with two to six beds. Ordinary electronic stethoscopes only amplify the signal and cannot specifically eliminate environmental noise. Blind noise reduction methods that rely on the body's surface channels to estimate noise are also inapplicable in this scenario.

[0006] 3. Insufficient spatial coverage. A standard stethoscope or single-channel patch only collects signals from a single location in the abdomen. Event signals from the distal gastrointestinal region may be completely missed after being attenuated by the abdominal wall tissue. However, simply increasing the number of channels will significantly increase the complexity, weight, and wearing burden of the device, and also introduce redundancy between adjacent locations.

[0007] 4. Lack of age-friendly design: Existing commercial solutions, such as the FDA-approved AbStats single-channel patch, smart clothing and multi-channel abdominal sensors, are difficult to deploy in one step due to the constraints of fragile skin, passive lying position and tight nursing time for elderly subjects. Large-area application can also easily cause tear injuries.

[0008] Although some existing technologies, such as Chinese patent CN 110141266 B, disclose a method for detecting bowel sounds using wearable body sound capture technology, the main steps are as follows: 1. Acquire abdominal dual-channel audio through a wearable device; 2. Divide the audio data stream of each channel into frames to obtain two sets of frame sequences; 3. Preprocess the two sets of frame sequences to find non-bowel sound data frames and use them as pre-labeled sequences; 4. Extract time-frequency domain features from each frame except the pre-labeled sequence, calculate short-time energy, zero-crossing rate, and Mel-frequency cepstral coefficients to obtain n-dimensional Mel-frequency cepstral coefficients, and obtain p-dimensional audio features for each frame; calculate the sum and difference of the two sets of features for the two channels to obtain 4. 5. Obtain the initial training set from the audio obtained in step 2 through manual listening and data labeling, and train the classifier; 6. Use the 4-dimensional feature sequence obtained in step 4... The method classifies and identifies bowel sounds using p-dimensional feature sequences. However, the pre-labeling method used in this approach relies on the assumption of differences in sound velocity between in vivo and in vitro, and cannot effectively distinguish environmental noises of the same source and velocity. Furthermore, it still faces challenges in one-step deployment due to the constraints of fragile skin, passive lying position, and limited time for nursing procedures in elderly subjects. Large-area labeling can also easily cause avulsion injuries. In addition, its features, such as the short-time energy, zero-crossing rate, and Mel-frequency cepstral coefficients of the underlying layer, contribute to its low accuracy. Summary of the Invention

[0009] The purpose of this invention is to provide a wearable bowel sound monitoring method based on dual-body extracellular reference.

[0010] The objective of this invention can be achieved through the following technical solutions: A wearable bowel sound monitoring method based on dual-body extracellular reference includes: Step S1: Fix the patch equipped with the first microphone, the second microphone and the third microphone to the subject. The first microphone and the second microphone are located on the side of the patch facing the subject, corresponding to the right lower abdominal ileocecal region and the left upper abdominal gastric region, respectively. The third microphone is located on the side of the patch away from the subject. Step S2: Acquire the raw audio signals collected by the first microphone, the second microphone and the third microphone, synchronize them on the time axis, and use the audio signal of the third microphone that has been synchronized on the time axis as the environmental noise reference signal. Reduce the noise of the audio signals of the first microphone and the second microphone that have been synchronized on the time axis to obtain the intestinal sound signal of the right lower abdomen ileocecal region and the intestinal sound signal of the left upper abdomen gastric region, respectively. Step S3: Based on the intestinal sound signals in the right lower quadrant ileocecal region and the gastric region in the left upper quadrant, respectively, the intestinal sound event sequence in the right lower quadrant ileocecal region and the intestinal sound event sequence in the left upper quadrant gastric region are obtained. The intestinal sound event sequence consists of multiple intestinal sound events, and each intestinal sound time includes the event type, event start time, event end time and event confidence. Step S4: Merge the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence to obtain a merged event sequence. If, at the same time, the event types of both the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence are not empty, then select the event type with higher confidence as the event type at that time. Step S5: Based on the merged event sequence, extract the total event rate, gas-liquid sound event rate, high-pitched metallic sound event rate, average event duration, average event interval, frequency range, and 24-hour trend slope as event flow features; Step S6: Calculate and display the continuous risk classification results based on the event flow characteristics.

[0011] The patch comprises a data acquisition layer, a fabric layer, a body surface microphone layer, and a hydrogel layer arranged sequentially. The data acquisition layer is positioned away from the subject and integrates a microcontroller, an I2S interface, and a data transmission module. The fabric layer is made of self-adhesive non-woven fabric, and its projection range and area on the subject's skin surface are consistent with the overall projection range and area of ​​the patch on the subject's skin surface. The body surface microphone layer is used to accommodate the first microphone and the second microphone; The hydrogel layer covers only the area of ​​the first and second microphones.

[0012] The patch also has a diamond-shaped alignment hole for aligning with the subject's navel.

[0013] Step S2 includes: Step S2-1: Acquire the raw audio signals collected by the first microphone, the second microphone, and the third microphone, and synchronize them on the timeline; Step S2-2: Use the time-synchronized audio signal corresponding to the third microphone as the environmental noise reference signal, and perform noise reduction on the time-synchronized audio signals corresponding to the first and second microphones to obtain the right lower abdominal ileocecal region intestinal sound signal and the left upper abdominal gastric region intestinal sound signal, respectively.

[0014] The event types are categorized as follows: no event, normal bowel sounds, gas-fluid sounds, and high-frequency metallic sounds.

[0015] The process of obtaining the intestinal sound event sequence based on a single intestinal sound signal in step S3 includes: Step S3-1: Divide the intestinal sound signal into multiple detection frames; Step S3-2: Calculate the 64-dimensional log-Mel spectrum for each detection frame; Step S3-3: Concatenate the 64-dimensional log-Mel spectra of all frames to form a log-Mel spectra sequence; Step S3-4: Input the log-Mel spectrum sequence into the HTS-AT backbone network to obtain the intestinal sound event sequence.

[0016] Step S4 includes: If, at the same time, the event types of both the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence are not empty, then the event type with higher confidence is selected as the event type at that time. If, at the same time, only one of the intestinal sound event sequences in the right lower quadrant ileocecal region and the left upper quadrant gastric region has a non-empty event type, then the non-empty event type is directly taken as the event type at that time.

[0017] The total event rate is the number of bowel sound events per minute; The gas-liquid sound event rate is the number of bowel sound events of type gas-liquid sound per minute. The high-pitched metallic sound event rate is the number of bowel sound events of metallic type per minute; The average event duration is the arithmetic mean of the durations of all bowel sound events; The average event interval is the arithmetic mean of the interval between any two bowel sound events; The frequency range is the effective bandwidth within the window for detecting bowel sound events; The slope of the 24-hour trend is: in: This represents the 24-hour trend slope. Indexed by hour. This is the average of the hourly indexes. for No. j Average total event rate over hours For the past 24 hours y j The overall mean.

[0018] A wearable bowel sound monitoring device based on dual-body extracellular reference includes a memory, a processor, and a program stored in the memory, wherein the processor executes the program to implement the method described above.

[0019] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. On the one hand, by setting two surface microphones on the side of the patch facing the human body corresponding to the right lower abdominal ileocecal region and the left upper abdominal gastric region, and setting an environmental microphone on the side away from the human body, effective reference and targeted noise reduction of environmental noise are achieved, which is significantly better than the traditional blind noise reduction method that relies on the assumption of the difference in sound velocity between the body and the outside. On the other hand, by simultaneously monitoring two key anatomical locations, the right lower abdominal ileocecal region and the left upper abdominal gastric region, the problem of missing distal gastrointestinal event signals that may be missed by single-point monitoring is avoided, thus improving the comprehensiveness of monitoring. Furthermore, the intestinal sound event sequences of the two sites are merged, and the result with higher confidence is automatically selected when there is temporal overlap, which ensures the accuracy and robustness of event identification. Multiple event flow features, including event rate, event type ratio, and temporal distribution characteristics, are extracted, and continuous risk classification results are generated accordingly, providing an intuitive and quantitative gastrointestinal function assessment tool for clinical use.

[0021] 2. Self-adhesive non-woven fabric is used as the fabric layer, which has good fit and is friendly to delicate skin; the hydrogel layer only covers the microphone area, avoiding skin tearing caused by large-area adhesion, making it particularly suitable for elderly subjects and long-term wear scenarios.

[0022] 3. By aligning the diamond-shaped alignment hole with the subject's umbilicus, nurses can quickly and accurately position the patch on the abdomen, ensuring that the first and second microphones are aligned with the right lower ileocecal region and the left upper gastric region, respectively, reducing the difficulty of operation and the reliance on the experience of the personnel.

[0023] 4. It emphasizes the premise of time axis synchronization, ensuring the timing consistency of the three microphone signals; then it explicitly uses the third microphone signal as the environmental noise reference signal, providing a high-quality reference input for subsequent noise reduction algorithms such as adaptive filtering or spectral subtraction, fundamentally improving the noise reduction effect.

[0024] 5. A 64-dimensional log-Mel spectrum is used as the input feature and processed through the HTS-AT backbone network. Compared with traditional handcrafted features such as short-time energy, zero-crossing rate, and Mel-Cepstral coefficients, this method can automatically learn deeper and more discriminative acoustic patterns, significantly improving the accuracy of intestinal sound event recognition and overcoming the low accuracy problem mentioned in the background techniques.

[0025] 6. It not only specifies the selection rule when both sequences have events, but also clarifies the handling method when only one sequence has events, thus preserving effective monitoring information to the greatest extent and avoiding missed events due to weak signals from a certain channel.

[0026] 7. The defined total event rate directly corresponds to the frequency of pathological sounds of clinical concern, while the average event duration and average event interval reflect the rhythmicity of intestinal activity; the frequency range reflects the spectral characteristics of events; the key point is the introduction of a 24-hour trend slope, which quantifies the long-term trend of gastrointestinal function over time through linear regression, which helps to eliminate individual differences and enhance the generalization of the model. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the main steps of the method of the present invention. Detailed Implementation

[0028] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0029] A wearable bowel sound monitoring method based on dual-body external reference, such as Figure 1 As shown, it includes: Step S1: Fix the patch equipped with the first microphone, the second microphone and the third microphone to the subject. The first microphone and the second microphone are located on the side of the patch facing the subject, corresponding to the right lower abdominal ileocecal region and the left upper abdominal gastric region, respectively. The third microphone is located on the side of the patch away from the subject. The patch consists of a data acquisition layer, a fabric layer, a body surface microphone layer, and a hydrogel layer arranged sequentially. The data acquisition layer is located away from the subject setup and integrates a microcontroller, I2S interface, and data transmission module. The fabric layer is made of self-adhesive non-woven fabric, and its projection range and area on the subject's skin surface are consistent with the overall projection range and area of ​​the patch on the subject's skin surface, serving as a structural base and providing large-area gentle fixation.

[0030] The body surface microphone layer is used to house the first and second microphones; The hydrogel layer covers only the area of ​​the first and second microphones, providing acoustic impedance matching and localized reinforced adhesion.

[0031] The patch also features a diamond-shaped alignment hole for aligning with the subject's umbilicus. Specifically, a diamond-shaped alignment hole with a diagonal length of 10mm is located in the center of the patch, serving as an alignment anchor point. During application, the nurse aligns this hole with the subject's umbilicus. The rectangular shape provides a long-axis alignment clue, and both constrain the patch's translational and rotational degrees of freedom, ensuring that all sensing elements automatically fall into preset anatomical target positions: the first microphone is located in the lower right of the center, above the ileocecal junction in the lower right abdomen; the second microphone is located in the upper left of the center, above the proximal small intestine in the upper left abdomen; and the third microphone is located above the alignment hole in the center on the back of the patch, facing outwards to collect ambient sound. The entire application process can be completed by a single person within 20-30 seconds, without the need to measure anatomical landmarks, which is well-suited to the practical constraints of passive recumbency and limited nursing time after hip surgery in elderly patients.

[0032] Furthermore, since the projection range and area of ​​the nonwoven fabric on the subject's skin surface are consistent with the overall projection range and area of ​​the patch on the subject's skin surface, it provides weak adhesive distributed fixation across the entire patch area, dispersing the peeling force over a large area and reducing the risk of tearing on the fragile skin of the elderly. A medical hydrogel pad is placed directly below Mic-A and Mic-B, which not only compensates for the micro-adhesion force but also serves as an acoustic impedance matching layer between the skin and the MEMS diaphragm, avoiding sound reflection loss due to impedance mismatch. The total direct adhesive / gel-skin contact area is less than 100 cm². 2 This ensures acoustic coupling while controlling the risk of tear injuries.

[0033] The data acquisition module measures 50 mm × 45 mm × 2 mm and is controlled by an ESP32-S3 microcontroller. Its two independent I2S peripherals handle the sampling: I2S0 synchronously samples the first and second microphones using a shared SCK / WS clock, with each microphone occupying an SD0 data line to ensure time alignment between the two body surface channels; I2S1 independently samples the third microphone via the SD1 data line. A +5V DC input, regulated by a Buck converter, powers the ESP32-S3 and the three microphones. After reading the I2S frames, the microcontroller locally writes them to an SD card for redundancy, and simultaneously assembles UDP packets to push the raw audio to a bedside server in real time via Wi-Fi. The included 2000mAh lithium polymer battery supports continuous three-channel acquisition for more than 16 hours.

[0034] Step S2: Acquire the raw audio signals collected by the first, second, and third microphones, synchronize them along the time axis, and use the time-synchronized audio signal corresponding to the third microphone as the environmental noise reference signal. Perform noise reduction on the time-synchronized audio signals corresponding to the first and second microphones to obtain the intestinal sound signals from the right lower quadrant ileocecal region and the left upper quadrant gastric region, including: Step S2-1: Acquire the raw audio signals collected by the first microphone, the second microphone, and the third microphone, and synchronize them on the timeline; Step S2-2: Use the time-synchronized audio signal corresponding to the third microphone as the environmental noise reference signal, and perform noise reduction on the time-synchronized audio signals corresponding to the first and second microphones to obtain the right lower abdominal ileocecal region intestinal sound signal and the left upper abdominal gastric region intestinal sound signal, respectively.

[0035] Step S3: Based on the intestinal sound signals in the right lower quadrant ileocecal region and the gastric region in the left upper quadrant, respectively, the intestinal sound event sequence in the right lower quadrant ileocecal region and the intestinal sound event sequence in the left upper quadrant gastric region are obtained. The intestinal sound event sequence consists of multiple intestinal sound events, and each intestinal sound time includes the event type, event start time, event end time and event confidence. In this application, the event types are classified as no event, normal bowel sounds, gas-fluid sounds, and high-frequency metallic sounds.

[0036] Based on this, in this embodiment, the process of obtaining the intestinal sound event sequence based on a single intestinal sound signal in step S3 includes: Step S3-1: Divide the intestinal sound signal into multiple detection frames; Step S3-2: Calculate the 64-dimensional log-Mel spectrum for each detection frame; Step S3-3: Concatenate the 64-dimensional log-Mel spectra of all frames to form a log-Mel spectra sequence; Step S3-4: Input the log-Mel spectrum sequence into the HTS-AT backbone network to obtain the intestinal sound event sequence.

[0037] Step S4: Merge the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence to obtain a merged event sequence. If, at the same time, the event types of both the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence are not empty, then select the event type with higher confidence as the event type at that time. In this embodiment, step S4 includes: If, at the same time, the event types of both the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence are not empty, then the event type with higher confidence is selected as the event type at that time. If, at the same time, only one of the intestinal sound event sequences in the right lower quadrant ileocecal region and the left upper quadrant gastric region has a non-empty event type, then the non-empty event type is directly taken as the event type at that time.

[0038] Step S5: Based on the merged event sequence, extract the total event rate, gas-liquid sound event rate, high-pitched metallic sound event rate, average event duration, average event interval, frequency range, and 24-hour trend slope as event flow features; The total event rate is the number of bowel sound events per minute; The air-fluid sound event rate is the number of bowel sound events of type air-fluid sound per minute; The high-pitched metallic sound event rate is the number of bowel sound events of metallic type per minute; The average event duration is the arithmetic mean of the durations of all bowel sound events; The average event interval is the arithmetic mean of the intervals between any two bowel sound events; The frequency range is the effective bandwidth within the window for detecting bowel sound events; The 24-hour trend slope is: in: This represents the 24-hour trend slope. Indexed by hour. This is the average of the hourly indexes. for No. j Average total event rate over hours For the past 24 hours y j The overall mean.

[0039] The key point of this application is the introduction of a 24-hour trend slope, the function of which is as follows; 1. It can filter out short-term noise and focus on true trends. Specifically: The bowel sound event rate within a short time window is easily affected by various factors, such as brief changes in the subject's position, coughing, eating, or even nurses' rounds. These disturbances can cause significant fluctuations in the event rate, and relying solely on immediate values ​​for risk assessment can easily lead to false positives or false negatives.

[0040] The 24-hour trend slope is fitted with the average event rate of each hour over the past 24 hours using linear regression, which is equivalent to performing a "time smoothing" on the original signal. It filters out high-frequency random fluctuations and retains only the low-frequency trend components that reflect the long-term evolution of gastrointestinal function, thereby significantly improving the stability and accuracy of risk classification results.

[0041] 2. To achieve early warning and detect hidden deterioration, specifically: In clinical practice, intestinal obstruction often develops silently at night, when the subject may not have obvious symptoms, and a single-point auscultation may happen to fall during a period when the event rate is temporarily normal. However, if the trend slope of the past 24 hours is calculated, even if the event rate of the current hour is still within the normal range, as long as the slope shows a continuous downward trend, it can indicate in advance that gastrointestinal motility is weakening.

[0042] This ability to anticipate clinical symptoms is the core advantage of continuous monitoring over traditional intermittent auscultation. It helps healthcare professionals take action within the optimal intervention window, preventing delays in treatment.

[0043] 3. Eliminate individual differences and enhance model generalization, specifically: The baseline bowel sound levels vary greatly among different subjects, making it difficult to set a uniform threshold if the absolute event rate threshold is used directly for risk stratification.

[0044] The 24-hour trend slope is a relative quantity, describing the rate of change rather than an absolute value. Regardless of whether the subject's baseline is high or low, a significant negative deviation in the slope suggests potential functional impairment. This characteristic makes classification models based on this feature have better transferability and generalization accuracy across different populations.

[0045] 4. Provides intuitive clinical semantics to assist decision-making, specifically: For clinicians, trend information such as rising or declining event rates is far more instructive than a single isolated number. The sign and magnitude of the trend slope can be directly mapped to a qualitative judgment of improvement, stabilization, or deterioration, reducing the black box effect of machine learning model output and increasing doctors' trust in and willingness to adopt monitoring systems.

[0046] 5. Complementing other features to form a multi-dimensional characterization, specifically: The 24-hour trend slope does not operate in isolation; it works in conjunction with short-term features such as the total event rate, gas-liquid-sound event rate, and average event interval to form a complete three-dimensional feature space encompassing time, frequency, and events. For example: Short-term features are responsible for capturing what is happening at that moment, representing acute events; The trend slope is responsible for indicating the direction in which the whole is developing, representing a slow evolution.

[0047] The combination of these two approaches enables the model to respond quickly to emergencies while remaining vigilant against gradual deterioration, ultimately achieving a more refined and reliable risk stratification.

[0048] Step S6: Calculate and display the continuous risk classification results based on the event flow characteristics.

[0049] Specifically, in this embodiment, the XGBoost gradient boosting tree predictor is used to obtain continuous risk classification results. The XGBoost gradient boosting tree classifier is trained on the aforementioned features, and the continuous risk score ∈ [0,1] is output. The reason for choosing XGBoost is that its feature importance is interpretable, it is robust to heterogeneous feature dimensions, and it performs well on small to medium-sized tabular data. The model is trained independently of the event detection network, making the diagnostic module easy to iterate and update in clinical practice.

[0050] In this embodiment, a three-level color-coded alarm is used to map continuous risk scores to three levels of clinical alarms: 1. Low risk, displayed in green, specifically indicating a continuous risk score of less than 0.3, in which case no alarm will be pushed to mobile devices; 2. Medium risk, displayed in yellow, specifically indicating a continuous risk score between 0.3 and 0.7; 3. High risk, displayed in red, specifically indicating a continuous risk score greater than 0.7, and alarms will be pushed out in multiple ways.

[0051] When there is a 24-hour downward trend but the event rate rebounds briefly, the system avoids blindly downgrading by using trend characteristics in order to deal with the patient's P-L2-type non-monotonic recovery trajectory.

[0052] In addition, this application also designs a dual-role mobile user interface: 1. The Clinician View provides the following panels: (i) Current Risk Level – displayed with green / yellow / red badges; (ii) 24-Hour Event Rate Trend Chart – displays the hourly time series of the total event rate as a line chart, overlaid with risk threshold lines; (iii) Category Proportion Bar Chart – displays the proportion of normal / gurgling / high-pitched events; (iv) Event Playback Panel – allows for audio playback and spectrogram visualization of individual detected events, enabling physicians to quickly review them.

[0053] 2. Subject view, presented in a non-numerical, low-cognitive-load manner: simplified status indicators, such as "Your bowel has returned to normal" or "Please inform the nurse," and a one-click symptom reporting button, such as bloating or nausea. The reported content is fed back to the physician's panel as additional supporting evidence. This design is responsive to interview feedback from elderly patients who may not understand medical indicators but can distinguish colors and simple text.

[0054] 3. Privacy and alarm fatigue control: The system will only extract features from the original audio at the bedside and then discard it. Only the features exported, such as event counts, categories, and risk scores, will be persisted and transmitted. This reduces the risk of unintentionally collecting patients' private conversations from the source. Low-risk states will only be recorded and not pushed to alleviate nursing alarm fatigue.

[0055] Finally, a trial was conducted to verify this. Continuous bowel sound recordings were collected from 406 subjects aged 51-99 years in the post-orthopedic ward of a tertiary hospital, with a mean bowel sound of 68.4. 214 subjects were female. The trial was conducted with IRB approval. Two senior clinicians independently annotated 20 hours of recordings, obtaining 10,736 event-level labels (three categories: normal / gurgling / high-pitched). The annotation consistency was 0.84, and any disputes were resolved by a senior gastroenterologist.

[0056] At the patient level, labeled data were divided into 70% for training, 20% for validation, and 10% for testing to ensure that the same patient would not appear in both the training and testing sets simultaneously. Among them, 54 cases were longitudinal tracking data (51 cases covering up to postoperative day 7 and 3 cases extending to postoperative day 90) used to train and evaluate the XGBoost risk prediction module, employing Leave-One-Patient-Out cross-validation.

[0057] The patch terminal (ESP32-S3) is responsible for data acquisition and wireless transmission. The bedside terminal runs ANC, HTS-AT event detection, dual-channel fusion, and XGBoost risk prediction, and provides a dual-view web application for physicians and patients. Low-risk events do not trigger uplink alarms; only medium- and high-risk events are pushed to the ward nursing station. The whole system forms a closed loop of patch acquisition, bedside inference, and nursing station alarms.

[0058] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A wearable bowel sound monitoring method based on dual-body extracellular reference, characterized in that, include: Step S1: Fix the patch equipped with the first microphone, the second microphone and the third microphone to the subject. The first microphone and the second microphone are located on the side of the patch facing the subject, corresponding to the right lower abdominal ileocecal region and the left upper abdominal gastric region, respectively. The third microphone is located on the side of the patch away from the subject. Step S2: Acquire the raw audio signals collected by the first microphone, the second microphone and the third microphone, synchronize them on the time axis, and use the audio signal of the third microphone that has been synchronized on the time axis as the environmental noise reference signal. Reduce the noise of the audio signals of the first microphone and the second microphone that have been synchronized on the time axis to obtain the intestinal sound signal of the right lower abdomen ileocecal region and the intestinal sound signal of the left upper abdomen gastric region, respectively. Step S3: Based on the intestinal sound signals in the right lower quadrant ileocecal region and the gastric region in the left upper quadrant, respectively, the intestinal sound event sequence in the right lower quadrant ileocecal region and the intestinal sound event sequence in the left upper quadrant gastric region are obtained. The intestinal sound event sequence consists of multiple intestinal sound events, and each intestinal sound time includes the event type, event start time, event end time and event confidence. Step S4: Merge the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence to obtain a merged event sequence. If, at the same time, the event types of both the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence are not empty, then select the event type with higher confidence as the event type at that time. Step S5: Based on the merged event sequence, extract the total event rate, gas-liquid sound event rate, high-pitched metallic sound event rate, average event duration, average event interval, frequency range, and 24-hour trend slope as event flow features; Step S6: Calculate and display the continuous risk classification results based on the event flow characteristics.

2. The wearable bowel sound monitoring method based on dual-body extracellular reference according to claim 1, characterized in that, The patch comprises a data acquisition layer, a fabric layer, a body surface microphone layer, and a hydrogel layer arranged sequentially. The data acquisition layer is positioned away from the subject and integrates a microcontroller, an I2S interface, and a data transmission module. The fabric layer is made of self-adhesive non-woven fabric, and its projection range and area on the subject's skin surface are consistent with the overall projection range and area of ​​the patch on the subject's skin surface. The body surface microphone layer is used to accommodate the first microphone and the second microphone; The hydrogel layer covers only the area of ​​the first and second microphones.

3. The wearable bowel sound monitoring method based on dual-body extracellular reference according to claim 2, characterized in that, The patch also has a diamond-shaped alignment hole for aligning with the subject's navel.

4. The wearable bowel sound monitoring method based on dual-body extracellular reference according to claim 1, characterized in that, Step S2 includes: Step S2-1: Acquire the raw audio signals collected by the first microphone, the second microphone, and the third microphone, and synchronize them on the timeline; Step S2-2: Use the time-synchronized audio signal corresponding to the third microphone as the environmental noise reference signal, and perform noise reduction on the time-synchronized audio signals corresponding to the first and second microphones to obtain the right lower abdominal ileocecal region intestinal sound signal and the left upper abdominal gastric region intestinal sound signal, respectively.

5. The wearable bowel sound monitoring method based on dual-body extracellular reference according to claim 1, characterized in that, The event types are categorized as follows: no event, normal bowel sounds, gas-fluid sounds, and high-frequency metallic sounds.

6. The wearable bowel sound monitoring method based on dual-body extracellular reference according to claim 5, characterized in that, The process of obtaining the intestinal sound event sequence based on a single intestinal sound signal in step S3 includes: Step S3-1: Divide the intestinal sound signal into multiple detection frames; Step S3-2: Calculate the 64-dimensional log-Mel spectrum for each detection frame; Step S3-3: Concatenate the 64-dimensional log-Mel spectra of all frames to form a log-Mel spectra sequence; Step S3-4: Input the log-Mel spectrum sequence into the HTS-AT backbone network to obtain the intestinal sound event sequence.

7. The wearable bowel sound monitoring method based on dual-body extracellular reference according to claim 1, characterized in that, Step S4 includes: If, at the same time, the event types of both the right lower quadrant ileocecal region bowel sound event sequence and the left upper quadrant gastric region bowel sound event sequence are not empty, then the event type with higher confidence is selected as the event type at that time. If, at the same time, only one of the intestinal sound event sequences in the right lower quadrant ileocecal region and the left upper quadrant gastric region has a non-empty event type, then the non-empty event type is directly taken as the event type at that time.

8. The wearable bowel sound monitoring method based on dual-body extracellular reference according to claim 1, characterized in that, The total event rate is the number of bowel sound events per minute; The gas-liquid sound event rate is the number of bowel sound events of type gas-liquid sound per minute. The high-pitched metallic sound event rate is the number of bowel sound events of metallic type per minute; The average event duration is the arithmetic mean of the durations of all bowel sound events; The average event interval is the arithmetic mean of the interval between any two bowel sound events; The frequency range is the effective bandwidth within the window for detecting bowel sound events; The slope of the 24-hour trend is: in: This represents the 24-hour trend slope. Indexed by hour. This is the average of the hourly indexes. for No. j Average total event rate over hours For the past 24 hours y j The overall mean.

9. A wearable bowel sound monitoring device based on dual-body extracellular reference, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-8.

10. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method as described in any one of claims 1-8.

Images