Tracheal catheter with body temperature measuring function
Through the M-shaped airbag design and deep learning technology, combined with the heat exchange dynamic model, the problem of respiratory airflow interference and core body temperature estimation in the body temperature monitoring of tracheal catheter is solved, and high-precision and continuous body temperature monitoring are achieved.
Patent Information
- Application Number
- CN202411009452.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing tracheal catheters are difficult to effectively eliminate respiratory airflow interference in body temperature monitoring, and it is difficult to accurately estimate core body temperature, which affects the accuracy and reliability of measurements.
The M-shaped airbag design combines temperature sensors and vibration sensors. Through the heat exchange dynamic model and temperature gradient model, combined with deep learning technology, the temperature value in the patient's body is calculated and displayed in real time, reducing respiratory airflow interference and improving measurement accuracy.
It significantly improves the accuracy and reliability of body temperature measurements, and can provide personalized high-precision body temperature monitoring in different patients and clinical situations, reducing short-term fluctuations and providing a trend of temperature change.
Smart Images

Figure CN118925013A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of tracheal tubes, and in particular relates to a tracheal tube with a body temperature measurement function. Background Art
[0002] With the continuous advancement of medical technology and the improvement of patient safety awareness, temperature management during surgery has become an important part of anesthesia. Among the quality control indicators released by the National Anesthesia Quality Control Center in 2022, four directly involve temperature management, which fully reflects the important position of temperature monitoring in modern anesthesia practice. Among them, "temperature monitoring rate during general anesthesia" as a key indicator requires anesthesiologists to continuously monitor the patient's temperature during general anesthesia surgery.
[0003] Importance of core body temperature monitoring:
[0004] 1. Prevent surgery-related hypothermia: Studies have shown that the incidence of mild hypothermia (core body temperature <36°C) is as high as 70% in some surgeries. Hypothermia can lead to multiple complications such as coagulation disorders, delayed wound healing, and immune suppression.
[0005] 2. Identify malignant hyperthermia: Prompt detection of abnormally elevated body temperature can lead to early identification of the rare but life-threatening malignant hyperthermia syndrome.
[0006] 3. Optimize drug metabolism: Changes in body temperature significantly affect the metabolism and distribution of anesthetic drugs. Accurate body temperature monitoring helps to precisely control the depth of anesthesia.
[0007] 4. Guide temperature management strategies: Real-time and accurate body temperature data is the basis for developing individualized warming plans.
[0008] Limitations of existing core body temperature monitoring methods:
[0009] 1. Esophageal temperature probe: Although highly accurate, it may cause patient discomfort and is difficult to use in certain head and neck surgeries.
[0010] 2. Bladder temperature monitoring: requires a catheter, increases the risk of urinary tract infection, and is inaccurate when urine output is low.
[0011] 3. Nasopharyngeal temperature probe: easily affected by respiratory airflow and may interfere with certain craniofacial surgeries.
[0012] 4. Rectal temperature probe: accurate but highly invasive and difficult to use in abdominal surgery.
[0013] 5. Temporal artery temperature measurement: non-invasive, but the accuracy is greatly affected by the external environment.
[0014] In view of the above background, it is an urgent need for the anesthesia discipline to develop a core temperature monitoring method that can be widely used in various surgeries, is easy to operate, and has stable accuracy. As a necessary respiratory management device for patients under general anesthesia, if the endotracheal tube can also achieve accurate temperature monitoring, it will greatly improve clinical practicality and is expected to become an ideal solution to the current dilemma.
[0015] However, the existing tracheal tube temperature monitoring technology still has the following shortcomings: First, the temperature change of the airflow during breathing will interfere with the measurement results. The inhaled cold air and the exhaled hot air pass alternately, resulting in large temperature fluctuations measured by the sensor, which is difficult to reflect the true body temperature. Secondly, there is a difference between the temperature of the inner wall of the trachea and the core body temperature, and how to accurately calculate the core body temperature is a challenge. Furthermore, the patient's breathing pattern, individual differences in tracheal anatomical structure, and the setting of ventilator parameters will affect the accuracy of the measurement. In addition, existing technologies often ignore the time series characteristics of temperature data and lack effective analysis of long-term temperature change trends.
[0016] Therefore, there is an urgent need for a tracheal tube technology solution that can overcome the above problems, provide high-precision, continuous body temperature monitoring, effectively eliminate the interference of respiratory airflow, accurately estimate core body temperature, and provide medical staff with intuitive and reliable body temperature information. Summary of the invention
[0017] In view of this, the present invention provides a tracheal tube with a body temperature measurement function, which can solve the technical problem that the prior art is difficult to effectively eliminate the interference of respiratory airflow and accurately estimate the core body temperature.
[0018] The present invention is achieved in that:
[0019] The present invention provides an endotracheal tube with a body temperature measurement function, which comprises a hose, an airbag, a temperature sensor, a vibration sensor, a single-chip computer, and a display. The airbag is wrapped on the outer wall of the hose, and the portion of the hose wrapped by the airbag is provided with one or more through holes. The portion of the hose extending out of the airbag is a tube head. The outer wall of the airbag is an M-shape formed by rotating around the central axis of the hose. The two ridges of the M-shape are arc-shaped. The valley of the M-shape is provided with a temperature sensor and a vibration sensor. The temperature sensor and the vibration sensor are connected to the single-chip computer via a data line, and the data line is embedded in the outer wall of the hose. The single-chip computer is electrically connected to the display. The single-chip computer is provided with an internal body temperature precision calculation module for calculating the temperature value in the patient's body in real time according to the signals collected by the temperature sensor and the vibration sensor, and displaying it through the display.
[0020] Among them, when the airbag is inflated and fits into the trachea or bronchus of the human lungs, the M-shaped valley of the airbag forms a small space with the inner wall of the trachea or bronchus, the M-shaped ridge of the airbag is in close contact with the inner wall of the trachea or bronchus, and the temperature sensor and the vibration sensor are in contact with the inner wall of the trachea or bronchus through a layer of fixed membrane; the fixed membrane is attached to the M-shaped valley of the outer wall of the airbag.
[0021] Furthermore, when the airbag is inflated and fits in the trachea or bronchus of the human lung, the top of the M-shaped ridge of the airbag is 0.5 to 1.5 mm higher than the temperature sensor or the vibration sensor.
[0022] Furthermore, the outer diameter of the hose is 5 to 8 mm.
[0023] Furthermore, the M-shape of the airbag is expressed by the formula y=A×sin(B×x)+C×sin(D×x), wherein the size ranges of the variables in the specific formula include: A=0.5~1.5mm, B=2π / 3~2π / 5, C=0.2~0.8mm, D=4π / 3~4π / 5.
[0024] Furthermore, the opening of the hose away from the airbag is an air inlet for connecting to a ventilator; the opening of the tube head away from the airbag is an air outlet for delivering gas to the patient's lungs.
[0025] Furthermore, the size range of the M-shaped airbag includes: circumference 15-25 mm, height 1-3 mm.
[0026] Furthermore, the temperature sensor and the vibration sensor are arranged relative to each other.
[0027] Furthermore, the portion of the hose wrapped by the airbag is provided with four through holes.
[0028] Wherein, the body temperature accurate calculation module is used to perform the following steps:
[0029] S10, collecting raw data from the temperature sensor and the vibration sensor, and preprocessing them to remove noise, to obtain preprocessed temperature data and preprocessed vibration data; and simultaneously recording the airflow temperature and flow rate set by the ventilator;
[0030] S20, segmenting the preprocessed temperature data and the preprocessed vibration data using a sliding time window, wherein the window length of the sliding time window is optimized by a genetic algorithm;
[0031] S30, identifying the respiratory cycle based on the preprocessed vibration data, and marking each time window as an inhalation phase, an exhalation phase, or a transition phase;
[0032] S40, extracting features within each time window, including:
[0033] Temperature characteristics: average temperature, temperature change rate, maximum and minimum temperatures;
[0034] Vibration characteristics: vibration frequency, vibration amplitude, energy distribution;
[0035] Respiratory characteristics, which are the respiratory characteristics of the endotracheal tube with temperature measurement function set for the lungs of patients assisted by a ventilator, including: exhalation time, inhalation time, inhalation-exhalation ratio, and airflow velocity value;
[0036] S50, constructing a heat exchange dynamic model: calculating the heat excursion of the trachea wall in each time window according to the airflow velocity and airflow temperature, including heat loss and heat gain;
[0037] S60, constructing a temperature gradient model using the tracheal anatomical structure data and the computational fluid dynamics model, for obtaining temperature estimates at different locations in the trachea according to the respiratory phase, airflow velocity, and temperature sensor position, giving a higher weight to a time window marked as an exhalation phase, and calculating a temperature correction vector according to the estimated position of the temperature sensor in the trachea using the temperature gradient model;
[0038] S70, using a pre-trained temperature precision calculation model, inputting the following parameters: temperature characteristics and vibration characteristics of each time window, thermal offset calculated by a heat exchange dynamic model, and temperature correction vector calculated by a temperature gradient model, to obtain a precise calculation value of the patient's lung temperature in each time window;
[0039] S80, smoothing the precise calculated values of lung temperature in a plurality of consecutive time windows to reduce short-term fluctuations;
[0040] S90. Display the accurately calculated value of the patient's lung temperature in real time on a display, and display the confidence interval at the same time.
[0041] Specifically, the step S10 specifically includes:
[0042] Step 101, collecting original temperature data of the temperature sensor and original vibration data of the vibration sensor;
[0043] Step 102: Perform median filtering on the original temperature data, and set the filter window size to 5 sampling points;
[0044] Step 103, low-pass filtering is performed on the temperature data after the median filtering, with a cutoff frequency of 10 Hz;
[0045] Step 104, performing wavelet transform denoising on the original vibration data, using db4 wavelet with 4 decomposition layers;
[0046] Step 105, performing bandpass filtering on the vibration data after wavelet transform denoising, with a passband of 0.1 to 10 Hz;
[0047] Step 106, recording the airflow temperature set by the ventilator with an accuracy of plus or minus 0.1 degrees Celsius;
[0048] Step 107, recording the airflow rate set by the ventilator with an accuracy of plus or minus 2%;
[0049] Step 108: Record the air flow temperature and the air flow velocity once per second.
[0050] Wherein, the step S20 specifically includes:
[0051] Step 201, define the chromosome encoding method, use 8-bit binary code to represent the window length, and the window length ranges from 1 second to 4 seconds;
[0052] Step 202, designing a fitness function, including three factors: signal-to-noise ratio, root mean square error, and computational complexity;
[0053] Step 203, executing the genetic algorithm, including initializing the population, evaluating fitness, selection, crossover and mutation operations;
[0054] Step 204, repeating step 203 until 100 generations are reached or the fitness value is no longer significantly improved;
[0055] Step 205: Select the chromosome with the highest fitness and decode to obtain the optimal window length;
[0056] Step 206: segment the data using the optimal window length, starting from the data starting point, and moving half of the window length each time;
[0057] Step 207: Ensure that adjacent windows have 50% overlap to capture the gradual change characteristics of the signal.
[0058] Wherein, the step S30 specifically includes:
[0059] Step 301: Perform time domain analysis on the vibration signal in each time window and calculate the mean and standard deviation of the signal;
[0060] Step 302, detect the peak and valley values of the signal, and set the threshold to the mean plus or minus 1.5 times the standard deviation;
[0061] Step 303, calculating the time interval between adjacent peak values or valley values to obtain a preliminary respiratory cycle estimate;
[0062] Step 304, perform fast Fourier transform on the signal to find the main frequency in the range of 0.1 to 0.5 Hz;
[0063] Step 305: Calculate power spectrum density and analyze energy distribution;
[0064] Step 306: Combining the time domain and frequency domain features, use a support vector machine classifier to classify each time window;
[0065] Step 307: introduce a hidden Markov model for post-processing to improve the temporal coherence of classification;
[0066] Step 308: Smoothing the state sequence output by the hidden Markov model and setting a minimum duration threshold.
[0067] Wherein, the step S40 specifically includes:
[0068] Step 401, calculating the arithmetic mean of the temperature data as the average temperature;
[0069] Step 402, using a linear regression method to calculate the rate of change of temperature over time;
[0070] Step 403, finding the highest temperature and the lowest temperature in the temperature data;
[0071] Step 404: Estimate the main vibration frequency of the vibration data using the autocorrelation function and the peak detection method;
[0072] Step 405, calculating the root mean square value of the vibration data as a measure of the vibration amplitude;
[0073] Step 406: Divide the vibration signal into three frequency bands: low frequency, medium frequency and high frequency, and calculate the energy proportion of each frequency band;
[0074] Step 407: Calculate the duration of exhalation and inhalation in each complete cycle based on the respiratory cycle recognition result;
[0075] Step 408, calculating the ratio of the inhalation time to the exhalation time as the inhalation-exhalation ratio;
[0076] Step 409: extract the average airflow velocity of the corresponding time window from the data recorded by the ventilator.
[0077] Wherein, the step S50 specifically includes:
[0078] Step 501, establishing a heat conduction model between the tracheal wall and the airflow, assuming that the tracheal wall is cylindrical;
[0079] Step 502, using the Dittus-Boelter equation to calculate the Nusselt number, and then calculate the convective heat transfer coefficient;
[0080] Step 503: Establish a heat balance equation, taking into account the heat exchange between the tracheal wall and the airflow;
[0081] Step 504, using a finite difference method to solve the heat balance equation, using an adaptive time step;
[0082] Step 505, calculating the heat loss in each time window, when the air flow temperature is lower than the air pipe wall temperature;
[0083] Step 506, calculating the heat gain in each time window, when the air flow temperature is higher than the air pipe wall temperature;
[0084] Step 507: introducing a periodic weight function to consider the effect of the breathing cycle on heat exchange;
[0085] Step 508: Calculate the temperature change of the tracheal wall as the final result of the thermal excursion.
[0086] Wherein, the step S60 specifically includes:
[0087] Step 601: Establish a geometric model of the trachea, and simplify the trachea into a cylinder with a variable diameter;
[0088] Step 602: construct a computational fluid dynamics model and discretize the tracheal space using a finite volume method;
[0089] Step 603: setting the boundary conditions of the model, including inlet, outlet and wall conditions;
[0090] Step 604, using SIMPLE algorithm to solve the computational fluid dynamics model;
[0091] Step 605: extract the temperature distribution on the centerline of the trachea based on the solution of the computational fluid dynamics model;
[0092] Step 606: Use cubic spline interpolation to fit the temperature distribution curve to obtain a continuous temperature gradient function;
[0093] Step 607, adjusting the temperature gradient function for different breathing stages, taking into account the characteristics of the inhalation, exhalation and transition stages;
[0094] Step 608, calculating the temperature difference between the sensor position and the ideal measurement position according to the estimated position of the temperature sensor in the trachea;
[0095] Step 609: Give a higher weight to the time window marked as the exhalation phase, and calculate a weighted temperature correction vector.
[0096] Wherein, the step S70 specifically includes:
[0097] Step 701, constructing an input feature vector, including temperature features, vibration features, breathing features, thermal offset and temperature correction vector;
[0098] Step 702: Use a hybrid model of a long short-term memory network combined with a fully connected neural network to accurately calculate the temperature;
[0099] Step 703: design a long short-term memory network part, including two long short-term memory layers, for capturing time series features;
[0100] Step 704, designing a fully connected network part, including two fully connected layers and an output layer, for integrating the long short-term memory network output and non-temporal features;
[0101] Step 705: Collect a large amount of patient data, including samples of different ages, genders, and medical conditions;
[0102] Step 706: Use time series data enhancement techniques, such as time warping, amplitude scaling, etc., to increase the diversity of training data;
[0103] Step 707: Use the root mean square error as the loss function and use the Adam optimizer to perform model training;
[0104] Step 708: Use K-fold cross validation to evaluate model performance and use Bayesian optimization to adjust hyperparameters;
[0105] Step 709: Convert the trained model to TensorFlow Lite format to adapt to the embedded system;
[0106] Step 710: Use the Monte Carlo dropout technique to estimate the uncertainty of the prediction and calculate the 95% confidence interval.
[0107] Wherein, the step S80 specifically includes:
[0108] Step 801, using exponentially weighted moving average as the main smoothing method;
[0109] Step 802: dynamically adjust the smoothing factor using an adaptive method according to the temperature change rate;
[0110] Step 803: For each new time window, calculate the difference between the current temperature estimate and the previous smoothed value;
[0111] Step 804, adjusting the smoothing factor according to the temperature difference, increasing the smoothing factor when the difference is large, and decreasing the smoothing factor when the difference is small;
[0112] Step 805: Calculate a new smoothing value using the adjusted smoothing factor;
[0113] Step 806: Introduce the median absolute deviation method to detect and process outliers;
[0114] Step 807, calculating the median and median absolute deviation of the smoothed values in the most recent period;
[0115] Step 808: determine whether the current smoothed value is an abnormal value, and if it is an abnormal value, use an interpolation method to replace it;
[0116] Step 809: introduce a trend protection mechanism to calculate the linear regression slope of the original temperature estimate in the most recent period of time;
[0117] Step 810: If the absolute value of the slope is large and the smoothed value is significantly different from the predicted value of the regression line, the smoothed value is adjusted;
[0118] Step 811: further adjust the smoothed temperature value based on the confidence interval;
[0119] Step 812: rounding the final smoothed value, and determining the number of decimal places to be retained according to the temperature range.
[0120] Wherein, the step S90 specifically includes:
[0121] Step 901, designing a display interface, including a main display area, a trend graph area, a confidence interval display, an alarm indication area, and a status information area;
[0122] Step 902: Use a color coding system to encode body temperature values using different colors;
[0123] Step 903: Update the temperature value and color of the main display area every 5 seconds, and update the trend chart every 30 seconds;
[0124] Step 904, calculate the temperature change rate for 30 minutes, and use arrows on the trend graph to indicate the temperature change direction and speed;
[0125] Step 905: setting a multi-level alarm threshold, and triggering an alarm when the temperature exceeds a normal range;
[0126] Step 906: display the data reliability level according to the width of the confidence interval;
[0127] Step 907: providing interactive functions, allowing viewing detailed historical data and switching temperature units;
[0128] Step 908: Provide data export and sharing functions, and support exporting to CSV or JSON format;
[0129] Step 909: monitor the sensor connection status and battery power in real time, and display the system running time and the next calibration time;
[0130] Step 910: Set up an abnormal situation handling mechanism, including temperature mutation detection and signal loss handling.
[0131] Wherein, the step S70 further includes:
[0132] Step 711: compress and quantize the model to reduce the model size and improve the reasoning speed;
[0133] Step 712: Implementing the incremental learning function of the model, allowing the model to be updated when new data is available;
[0134] Step 713: Introduce the attention mechanism to enable the model to dynamically focus on important features in different time periods;
[0135] Step 714: Implement the model explanation function and use methods such as SHAP value or LIME to explain the model prediction results;
[0136] Step 715: Establish a model performance monitoring system to regularly evaluate the performance of the model in actual use and trigger a retraining process when the performance degrades.
[0137] Wherein, the step S80 further includes:
[0138] Step 813, implement multi-scale smoothing, and smooth at different time scales simultaneously to capture short-term and long-term temperature changes;
[0139] Step 814: Introduce an adaptive threshold mechanism to dynamically adjust the abnormal value judgment criteria based on individual patient characteristics and historical data;
[0140] Step 815: Use wavelet transform to perform signal denoising to better retain important features of the temperature signal;
[0141] Step 816, implement predictive smoothing, use the time series prediction method to estimate the temperature value at the next time point, and incorporate it into the smoothing process;
[0142] Step 817: Introduce a feedback mechanism to adjust the smoothing parameters according to the feedback from medical staff to improve the adaptability of the system.
[0143] Wherein, the step S60 further includes:
[0144] Step 610: Use a computational fluid dynamics model to simulate airflow dynamics characteristics under different breathing modes;
[0145] Step 611, establishing a tracheal wall elastic deformation model, taking into account the dynamic changes of the trachea during breathing;
[0146] Step 612: Introduce humidity factor to simulate the effect of airway humidity on temperature distribution;
[0147] Step 613, construct a multi-layer heat conduction model, taking into account the different thermal characteristics of the tracheal mucosa, submucosal layer and cartilage layer;
[0148] Step 614: Implement adaptive mesh refinement, increase mesh density in areas with large temperature gradients, and improve calculation accuracy.
[0149] Compared with the prior art, the beneficial effects of the endotracheal tube with a body temperature measurement function provided by the present invention are as follows: the endotracheal tube with a body temperature measurement function proposed by the present invention effectively solves the problems existing in the prior art through innovative structural design and advanced data processing algorithms, and significantly improves the accuracy and reliability of body temperature measurement. Its main technical effects are reflected in the following aspects:
[0150] First, the invention adopts an M-shaped airbag design, which can not only effectively fix the position of the temperature sensor and the vibration sensor, but also form a small space in the trachea to reduce the direct impact of the respiratory airflow on the sensor. This innovative design greatly improves the stability and accuracy of the measurement.
[0151] Secondly, by introducing the vibration sensor and combining the data from the temperature sensor, the present invention can accurately identify the respiratory cycle and distinguish between the inhalation and exhalation phases. This enables the system to perform more accurate temperature sampling at the end of exhalation (closest to the core body temperature), significantly improving the accuracy of the measurement.
[0152] Furthermore, the present invention uses a complex data processing algorithm, including a heat exchange dynamic model and a temperature gradient model, which can effectively eliminate the influence of the ventilator airflow temperature on the measurement results and accurately estimate the core body temperature based on the temperature distribution in the trachea. This method not only improves the measurement accuracy, but also adapts to the individual differences of different patients and various clinical conditions.
[0153] In summary, the technical solution of the present invention solves the technical problem that it is difficult in the prior art to effectively eliminate the interference of respiratory airflow and accurately estimate core body temperature. BRIEF DESCRIPTION OF THE DRAWINGS
[0154] Figure 1 A schematic diagram of the structure of a tracheal tube with a body temperature measurement function provided by the present invention;
[0155] Figure 2 A flowchart of the steps executed by the module for accurate temperature calculation;
[0156] In the accompanying drawings, the components represented by the reference numerals are listed as follows:
[0157] 10. Hose; 11. Tube head; 12. Airbag; 13. Temperature sensor; 131. First data line; 14. Vibration sensor; 141. Second data line; 20. Single chip microcomputer; 21. Display. DETAILED DESCRIPTION
[0158] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0159] like Figure 1 As shown, the present invention provides an endotracheal tube with a body temperature measurement function, which includes a hose 10, an airbag 12, a temperature sensor 13, a vibration sensor 14, a single-chip computer 20, and a display 21. The airbag is wrapped on the outer wall of the hose, and the part of the hose wrapped by the airbag is provided with one or more through holes. The part of the hose extending out of the airbag is a tube head 11. The outer wall of the airbag is an M-shaped body formed by rotating around the central axis of the hose. The two ridges of the M-shaped body are arc-shaped. The valley of the M-shaped body is provided with a temperature sensor and a vibration sensor. The temperature sensor and the vibration sensor are connected to the single-chip computer through a first data line 131 and a second data line 141 respectively. The first data line 131 and the second data line 141 are embedded in the outer wall of the hose. The single-chip computer is electrically connected to the display. The single-chip computer is provided with an internal body temperature precision calculation module for calculating the temperature value in the patient's body in real time according to the signals collected by the temperature sensor and the vibration sensor, and displaying it through the display.
[0160] Among them, when the airbag is inflated and fits into the trachea or bronchus of the human lungs, the M-shaped valley of the airbag forms a small space with the inner wall of the trachea or bronchus, the M-shaped ridge of the airbag is in close contact with the inner wall of the trachea or bronchus, and the temperature sensor and the vibration sensor are in contact with the inner wall of the trachea or bronchus through a layer of fixed membrane; the fixed membrane is attached to the M-shaped valley of the outer wall of the airbag.
[0161] Furthermore, when the airbag is inflated and fits in the trachea or bronchus of the human lung, the top of the M-shaped ridge of the airbag is 0.5 to 1.5 mm higher than the temperature sensor or the vibration sensor.
[0162] Furthermore, the outer diameter of the hose is 5 to 8 mm.
[0163] Furthermore, the M-shape of the airbag is expressed by the formula y=A×sin(B×x)+C×sin(D×x), wherein the size ranges of the variables in the specific formula include: A=0.5~1.5mm, B=2π / 3~2π / 5, C=0.2~0.8mm, D=4π / 3~4π / 5.
[0164] Furthermore, the opening of the hose away from the airbag is an air inlet for connecting to a ventilator; the opening of the tube head away from the airbag is an air outlet for delivering gas to the patient's lungs.
[0165] Furthermore, the size range of the M-shaped airbag includes: circumference 15-25 mm, height 1-3 mm.
[0166] like Figure 2 As shown, the body temperature accurate calculation module is used to perform the following steps:
[0167] S10, collecting raw data from the temperature sensor and the vibration sensor, and preprocessing them to remove noise, to obtain preprocessed temperature data and preprocessed vibration data; and simultaneously recording the airflow temperature and flow rate set by the ventilator;
[0168] S20, segmenting the preprocessed temperature data and the preprocessed vibration data using a sliding time window, wherein the window length of the sliding time window is optimized by a genetic algorithm;
[0169] S30, identifying the respiratory cycle based on the preprocessed vibration data, and marking each time window as an inhalation phase, an exhalation phase, or a transition phase;
[0170] S40, extracting features within each time window, including:
[0171] Temperature characteristics: average temperature, temperature change rate, maximum and minimum temperatures;
[0172] Vibration characteristics: vibration frequency, vibration amplitude, energy distribution;
[0173] Respiratory characteristics, which are the respiratory characteristics of the endotracheal tube with temperature measurement function set for the lungs of patients assisted by a ventilator, including: exhalation time, inhalation time, inhalation-exhalation ratio, and airflow velocity value;
[0174] S50, constructing a heat exchange dynamic model: calculating the heat excursion of the trachea wall in each time window according to the airflow velocity and airflow temperature, including heat loss and heat gain;
[0175] S60, constructing a temperature gradient model using the tracheal anatomical structure data and the computational fluid dynamics model, for obtaining temperature estimates at different locations in the trachea according to the respiratory phase, airflow velocity, and temperature sensor position, giving a higher weight to a time window marked as an exhalation phase, and calculating a temperature correction vector according to the estimated position of the temperature sensor in the trachea using the temperature gradient model;
[0176] S70, using a pre-trained temperature precision calculation model, inputting the following parameters: temperature characteristics and vibration characteristics of each time window, thermal offset calculated by a heat exchange dynamic model, and temperature correction vector calculated by a temperature gradient model, to obtain a precise calculation value of the patient's lung temperature in each time window;
[0177] S80, smoothing the precise calculated values of lung temperature in a plurality of consecutive time windows to reduce short-term fluctuations;
[0178] S90. Display the accurately calculated value of the patient's lung temperature in real time on a display, and display the confidence interval at the same time.
[0179] Step S60 has the following operation points:
[0180] 1. "Tracheal anatomical structure data" provides the geometric shape and size information of the trachea, which is crucial for accurately simulating temperature distribution.
[0181] 2. The "Computational Fluid Dynamics Model" can simulate the flow and heat transfer process of gas in the trachea, taking into account the air flow velocity, temperature and heat exchange of the trachea wall.
[0182] 3. The temperature gradient model is constructed based on these data and models, which can predict the temperature distribution at different locations in the trachea.
[0183] 4. The model uses the breathing phase (inhalation, exhalation, or transition), the current airflow velocity, and the location of the temperature sensor as input parameters to obtain a more accurate temperature estimate.
[0184] 5. A higher weight is given to the exhalation phase because the temperature of the exhaled gas is closer to the core body temperature and provides more reliable temperature information.
[0185] 6. The calculation of the temperature correction vector takes into account the difference between the actual position of the sensor and the ideal measurement position (such as core temperature), thereby improving the accuracy of temperature measurement.
[0186] This step combines anatomical knowledge, fluid dynamics and real-time measurement data to create a complex but accurate temperature estimation system that can significantly improve the accuracy of body temperature measurement.
[0187] The specific implementation methods of the above steps are described in detail below:
[0188] The specific implementation of step S10 is: first, collect the raw data of the temperature sensor and the vibration sensor. The temperature sensor adopts the principle of thermal resistance temperature measurement and obtains temperature data by measuring the change of resistance value. The vibration sensor adopts a piezoelectric acceleration sensor and obtains vibration data by measuring the charge generated by the piezoelectric element. The sampling frequency is set to 100Hz to ensure that the rapid changes in the breathing process are captured.
[0189] Next, the raw data was preprocessed to remove noise. For the temperature data, a median filter algorithm was used to remove sudden noise, and the filter window size was set to 5 sampling points. Then a low-pass filter was used with a cutoff frequency of 10 Hz to remove high-frequency noise. For the vibration data, wavelet transform was first used for denoising, using the db4 wavelet with 4 decomposition layers. Then a bandpass filter was applied with a passband of 0.1-10 Hz to retain the frequency components related to breathing.
[0190] At the same time, record the airflow temperature and flow rate set by the ventilator. The airflow temperature is obtained through the temperature sensor built into the ventilator with an accuracy of ±0.1°C. The airflow velocity is obtained through the flow sensor of the ventilator with an accuracy of ±2%. These parameters are recorded once per second for subsequent heat exchange model calculations.
[0191] The purpose of this step is to obtain high-quality basic data to lay the foundation for subsequent analysis and calculation. Through denoising, the signal-to-noise ratio of temperature and vibration signals is improved, which is conducive to extracting more accurate features. At the same time, recording ventilator parameters provides necessary input data for the heat exchange model.
[0192] The specific implementation of step S20 is: segmenting the pre-processed temperature data and vibration data using a sliding time window. The selection of the window length is crucial for subsequent analysis, so a genetic algorithm is used for optimization.
[0193] First, define the chromosome encoding method: use 8-bit binary encoding to represent the window length, ranging from 1 second to 4 seconds (100-400 sampling points). For example, "00000000" represents 1 second, and "11111111" represents 4 seconds.
[0194] Then, design the fitness function:
[0195] F = w1*SNR+w2*(1 / RMSE)+w3*(1 / computational complexity)
[0196] Among them, SNR is the signal-to-noise ratio, RMSE is the root mean square error (relative to the manually annotated breathing cycle), and the computational complexity is proportional to the window length. The weights w1, w2, and w3 are set to 0.4, 0.4, and 0.2 respectively.
[0197] Next, execute the genetic algorithm:
[0198] 1. Initialize the population: randomly generate 50 chromosomes.
[0199] 2. Evaluate fitness: Calculate the fitness function value for each chromosome.
[0200] 3. Selection: Use roulette wheel selection method to select outstanding individuals.
[0201] 4. Crossover: Use single-point crossover and set the crossover probability to 0.8.
[0202] 5. Mutation: Bit flip mutation is used, and the mutation probability is set to 0.01.
[0203] 6. Repeat steps 2-5 until 100 generations are reached or the fitness value no longer increases significantly.
[0204] Finally, the chromosome with the highest fitness is selected and decoded to obtain the optimal window length.
[0205] The data were segmented using the optimal window length: starting from the data start point, the window length was moved by half each time, ensuring that adjacent windows had a 50% overlap to capture the gradual changes in the signal.
[0206] The purpose of this step is to determine the optimal time scale for data analysis. Through genetic algorithm optimization, a balance can be achieved between signal quality, respiratory cycle identification accuracy, and computational efficiency. The use of sliding windows allows continuous analysis, which is beneficial for capturing dynamic changes during breathing.
[0207] The specific implementation of step S30 is: identifying the respiratory cycle based on the preprocessed vibration data, marking each time window as an inhalation phase, an exhalation phase or a transition phase. This step uses a multi-level signal analysis method, combining time domain and frequency domain features to improve the accuracy of identification.
[0208] First, the vibration signal in each time window is analyzed in the time domain:
[0209] 1. Calculate the mean μ and standard deviation σ of the signal.
[0210] 2. Detect the peak and valley values of the signal and set the threshold to μ±1.5σ.
[0211] 3. Calculate the time interval between adjacent peaks or valleys to obtain a preliminary estimate of the respiratory cycle.
[0212] Next, perform frequency domain analysis:
[0213] 1. Perform a Fast Fourier Transform (FFT) on the signal.
[0214] 2. Find the dominant frequency in the range of 0.1-0.5 Hz (corresponding to 6-30 breaths per minute).
[0215] 3. Calculate the power spectral density (PSD) and analyze the energy distribution.
[0216] Then, the support vector machine (SVM) classifier is used to classify each time window by combining the time domain and frequency domain features:
[0217] 1. The characteristic vectors include: peak amplitude, valley amplitude, peak-valley time difference, main frequency, and PSD maximum value.
[0218] 2. Use the RBF kernel function with kernel parameter γ=0.1 and penalty parameter C=1.
[0219] 3. Use a three-category strategy: inhalation, exhalation, and transition phase.
[0220] 4. Optimize SVM parameters using cross validation.
[0221] Optionally, in order to improve the temporal coherence of the classification, a Hidden Markov Model (HMM) is introduced for post-processing:
[0222] 1. Define the state space: S = {inhale, exhale, transition}.
[0223] 2. Initial probability distribution: π = [0.4, 0.4, 0.2].
[0224] 3. Transition probability matrix:
[0225]
[0226] 4. The emission probability is based on the classification result of SVM.
[0227] 5. Use the Viterbi algorithm to solve the most likely state sequence.
[0228] Finally, the state sequence output by HMM is smoothed:
[0229] 1. Set minimum duration thresholds: no less than 0.5 seconds for the inhalation and exhalation phases, and no less than 0.1 seconds for the transition phase.
[0230] 2. Merge states with too short duration into adjacent states.
[0231] The purpose of this step is to accurately identify the various stages of the respiratory cycle and provide a time reference for subsequent temperature analysis. By combining multiple signal processing and machine learning techniques, the robustness and accuracy of recognition are improved. The introduction of HMM takes into account the time dependency of the respiratory state, further improving the coherence of the classification.
[0232] The specific implementation of step S40 is: extracting features in each time window, including temperature features, vibration features and breathing features. This step aims to capture key information in the breathing process from multiple dimensions and provide comprehensive input data for subsequent accurate calculation of body temperature.
[0233] Temperature feature extraction:
[0234] 1. Average temperature: Calculate the arithmetic mean of the temperature data within the window.
[0235]
[0236] Where N is the number of data points in the window, T i is the ith temperature data point.
[0237] 2. Temperature change rate: Use linear regression method to calculate the rate of change of temperature over time.
[0238]
[0239] Among them, t i is the timestamp of the i-th data point.
[0240] 3. Maximum and minimum temperature:
[0241]
[0242] Vibration feature extraction:
[0243] 1. Vibration frequency: The main vibration frequency is estimated using the autocorrelation function and peak detection method.
[0244]
[0245] Among them, x i is the vibration signal, τ is the time delay. The main frequency f is R xx The frequency corresponding to the first non-zero peak of (τ).
[0246] 2. Vibration Amplitude: Calculate the root mean square (RMS) value as a measure of the vibration amplitude.
[0247]
[0248] 3. Energy distribution: Calculate the energy proportion of different frequency bands.
[0249] The signal is divided into three frequency bands: low frequency (0-1Hz), medium frequency (1-5Hz) and high frequency (5-10Hz).
[0250] Calculate the energy share for each frequency band j:
[0251]
[0252] Where X(f) is the Fourier transform of the signal.
[0253] Respiration feature extraction:
[0254] 1. Exhalation duration and inhalation duration: Based on the respiratory cycle identification result of step S30, the exhalation and inhalation duration in each complete cycle is calculated.
[0255] 2. Inspiratory-expiratory ratio: Calculate the ratio of inhalation duration to exhalation duration.
[0256]
[0257] 3. Airflow velocity value: Extract the average airflow velocity in the corresponding time window from the data recorded by the ventilator.
[0258]
[0259] Among them, V i is the i-th airflow velocity data point recorded by the ventilator.
[0260] To improve the stability and representativeness of the features, the mean and standard deviation of the above features were calculated for time windows spanning multiple respiratory cycles.
[0261] The purpose of this step is to comprehensively capture the temperature changes, mechanical vibrations, and respiratory dynamics during breathing. Temperature features reflect local heat exchange, vibration features provide mechanical information about respiratory activity, and respiratory features describe the overall breathing pattern. These multidimensional features provide rich input data for subsequent accurate calculation of body temperature, help distinguish temperature change patterns in different respiratory stages, and consider the impact of ventilator assistance on temperature measurement.
[0262] The specific implementation of step S50 is: construct a heat exchange dynamic model, and calculate the heat excursion of the trachea wall in each time window according to the airflow velocity and airflow temperature, including heat loss and heat gain. This step uses heat conduction theory and computational fluid dynamics principles to establish a simplified but effective heat exchange model.
[0263] First, establish a heat conduction model between the tracheal wall and the airflow:
[0264] 1. Assume that the tracheal wall is cylindrical with an inner diameter of d and a wall thickness of δ.
[0265] 2. Air flow temperature is T f , the initial temperature of the tracheal wall is T w , the core temperature is T c .
[0266] 3. The air velocity is v, the density is ρ, and the specific heat capacity is c p .
[0267] Next, calculate the convective heat transfer coefficient h:
[0268] Calculate the Nusselt number using the Dittus-Boelter equation:
[0269] Nu=0.023·Re 0.8 ·Pr 0.4
[0270] Where Re is the Reynolds number and Pr is the Prandtl number.
[0271] μ is the dynamic viscosity of the gas.
[0272] k is the thermal conductivity of the gas.
[0273] Convective heat transfer coefficient:
[0274] Then, the heat balance equation is established:
[0275]
[0276] Among them, ρ w 、c p,w 、V w are the density, specific heat capacity and volume of the tracheal wall, A is the inner surface area, A c is the cross-sectional area, k w is the thermal conductivity of the tracheal wall.
[0277] Solve the heat balance equation using the finite difference method:
[0278] Discretize time with a step size of Δt, then:
[0279]
[0280] Here, the superscript n represents the time step.
[0281] In order to improve computational efficiency, an adaptive time step is used:
[0282] 1. The initial time step is set to 0.01 seconds.
[0283] 2. If the temperature change between two adjacent steps exceeds 0.1°C, the time step is halved.
[0284] 3. If the temperature change is less than 0.01°C for 10 consecutive steps, double the time step, but not more than 0.1 seconds.
[0285] Calculate the thermal excursion within each time window:
[0286] 1. Heat loss: When T f <T w hour,
[0287] 2. Heat gain: When T f >T w hour,
[0288] Among them, t1 and t2 are the start and end time of the time window.
[0289] In order to consider the effect of the breathing cycle on heat exchange, a periodic weight function is introduced:
[0290] w(t)=0.5+0.5sin(2πft)
[0291] Wherein, f is the respiratory frequency, which is obtained from the respiratory cycle recognition result in step S30.
[0292] Corrected thermal offset calculation:
[0293]
[0294] Finally, calculate the temperature change of the tracheal wall:
[0295]
[0296] The purpose of this step is to accurately simulate the dynamics of heat exchange between the tracheal wall and the airflow during breathing. By considering the effects of convective heat transfer, heat conduction, and the respiratory cycle, the model can accurately estimate the temperature changes of the tracheal wall and provide an important correction factor for the subsequent accurate calculation of body temperature. The introduction of the model's adaptability and periodic weights enables it to better adapt to different breathing patterns and temperature changes.
[0297] The specific implementation of step S60 is: constructing a temperature gradient model using tracheal anatomical structure data and a computational fluid dynamics model to obtain temperature estimates at different locations in the trachea according to the respiratory phase, airflow velocity, and temperature sensor location. This step combines physiological knowledge and numerical simulation technology to create an accurate tracheal temperature distribution model.
[0298] First, establish the geometric model of the trachea:
[0299] 1. Based on anatomical data, the trachea is simplified into a cylinder with variable diameter.
[0300] 2. Define the trachea length L, starting diameter D0, and ending diameter D L .
[0301] 3. Function of tracheal diameter changing with length: D(x)=D0+(D L -D0)·(x / L) 1 / 3 , where x is the distance from the tracheal entrance.
[0302] Then, a computational fluid dynamics (CFD) model was constructed:
[0303] 1. Discretize the tracheal space using the finite volume method.
[0304] 2. Grid generation: Use structured grid, which is divided into 100 units in the axial direction and 20 units in the radial direction.
[0305] 3. Control equations:
[0306] Continuity equation:
[0307] Momentum equation:
[0308] Energy equation:
[0309] Where ρ is the density, is the velocity vector, p is the pressure, τ is the stress tensor, is the gravitational acceleration, E is the total energy, k is the thermal conductivity, S h is the volume heat source term.
[0310] 4. Boundary conditions:
[0311] Inlet: Speed inlet, speed is provided by ventilator data.
[0312] Outlet: Pressure outlet, set to atmospheric pressure.
[0313] Wall: No-slip boundary condition, temperature set to body temperature (initial value 37°C).
[0314] 5. Solution method:
[0315] The SIMPLE (Semi-Implicit Method for Pressure Linked Equations) algorithm is used to solve the problem.
[0316] The time step is kept consistent with step S50 to ensure synchronization between models.
[0317] Next, build the temperature gradient model:
[0318] 1. Based on the solution results of the CFD model, extract the temperature distribution T(x) on the centerline of the trachea.
[0319] 2. Use cubic spline interpolation to fit the temperature distribution curve to obtain a continuous temperature gradient function.
[0320] 3. Define the temperature gradient function:
[0321] Where x is the position along the tracheal axis, v is the current airflow velocity, and φ represents the breathing phase (inhalation, exhalation, or transition).
[0322] 4. Adjust the temperature gradient function for different breathing stages:
[0323] Inspiratory phase:
[0324] Exhalation phase:
[0325] Transition phase:
[0326] Among them, t is the time after the start of the stage, α and β are adjustment parameters, and τ is the duration of the transition stage.
[0327] Then, calculate the temperature correction vector:
[0328] 1. Based on the estimated position x of the temperature sensor in the trachea s , calculate the temperature difference between the sensor location and the ideal measurement location (such as the core temperature measurement point):
[0329]
[0330] Among them, x c It is the ideal location for measuring the core temperature.
[0331] 2. Give higher weight to the time window marked as the exhalation phase:
[0332]
[0333] Among them, t exp is the exhalation duration, T cycle is the complete breathing cycle time, and γ is the weight factor (range 0.5-1.0).
[0334] 3. Calculate the weighted temperature correction vector:
[0335]
[0336] in, is the unit vector pointing from the sensor to the ideal measurement point.
[0337] Finally, in order to improve the accuracy and adaptability of the model, machine learning methods are introduced for model calibration:
[0338] 1. Collect real patient data, including temperature distribution under different breathing patterns.
[0339] 2. Use the random forest regression algorithm to train the calibration function of the temperature gradient model.
[0340] 3. Features include: airflow velocity, breathing phase, tracheal position, CFD model predicted temperature, etc.
[0341] 4. The target variable is the difference between the measured temperature and the temperature predicted by the model.
[0342] 5. Use cross-validation to optimize model parameters, such as the number of trees, maximum depth, etc.
[0343] The purpose of this step is to create a high-precision model of the temperature distribution in the trachea, taking into account the anatomical structure of the trachea, the fluid dynamics, and the effects of the respiratory cycle. By combining CFD simulation and machine learning techniques, the model can accurately estimate the temperature at different locations and provide a reliable temperature correction vector. This is crucial for inferring core temperature from local measurements, significantly improving the accuracy of body temperature measurements.
[0344] The specific implementation of step S70 is: using the pre-trained temperature precision calculation model, inputting various parameters, and obtaining the precise calculation value of the patient's lung temperature in each time window. This step integrates the outputs of the previous steps and uses advanced machine learning algorithms to achieve accurate body temperature estimation.
[0345] First, construct the input feature vector:
[0346] 1. Temperature characteristics: average temperature Temperature change rate dT / dt, maximum temperature T max , minimum temperature T min .
[0347] 2. Vibration characteristics: vibration frequency f, vibration amplitude A RMS 、Energy distribution E j (low, medium and high frequency bands).
[0348] 3. Breathing characteristics: Exhalation duration T exp , Inspiratory duration T ins , breathing ratio R I:E , air velocity value
[0349] 4. Thermal offset ΔT calculated by heat exchange dynamic model w .
[0350] 5. Temperature correction vector calculated by temperature gradient model
[0351] Then, use the deep learning model to accurately calculate the temperature:
[0352] 1. A hybrid model that uses a long short-term memory network (LSTM) combined with a fully connected neural network.
[0353] 2. The LSTM part is used to capture time series features, and its structure is as follows:
[0354] -Input layer: the feature dimension is the same as the input vector
[0355] -LSTM layer 1: 128 units, dropout rate 0.2
[0356] -LSTM layer 2: 64 units, dropout rate 0.2
[0357] 3. The fully connected network part is used to integrate LSTM output and non-temporal features:
[0358] -Fully connected layer 1: 128 neurons, ReLU activation function
[0359] -Fully connected layer 2: 64 neurons, ReLU activation function
[0360] -Output layer: 1 neuron, linear activation function
[0361] Model training process:
[0362] 1. Data preparation: Collect a large amount of patient data, including samples of different ages, genders, and medical conditions.
[0363] 2. Data enhancement: Use time series data enhancement techniques such as time warping, amplitude scaling, etc.
[0364] 3. Loss function: Root mean square error (RMSE) is used as the main loss function.
[0365] 4. Optimizer: Adam optimizer is used, the initial value of the learning rate is set to 0.001, and the learning rate decay strategy is used.
[0366] 5. Regularization: In addition to dropout, L2 regularization is also used to prevent overfitting.
[0367] 6. Training strategy: Use small batch gradient descent with a batch size of 32.
[0368] 7. Early stopping strategy: Stop training when the validation set loss has not improved for 5 consecutive epochs.
[0369] Model evaluation and tuning:
[0370] 1. Use K-fold cross validation (K=5) to evaluate model performance.
[0371] 2. Use Bayesian optimization method to adjust hyperparameters, including the number of LSTM units, dropout rate, and number of neurons in the fully connected layer.
[0372] 3. Performance indicators: In addition to RMSE, mean absolute error (MAE) and R 2 Fraction.
[0373] Model deployment and real-time computing:
[0374] 1. Convert the trained model to TensorFlow Lite format to suit the embedded system.
[0375] 2. Perform real-time preprocessing and feature extraction on the data of each time window.
[0376] 3. Use the sliding window technique to input data from multiple consecutive time windows (such as the past 30 seconds) into the LSTM each time.
[0377] 4. The model outputs an accurate estimate of body temperature for the current time window.
[0378] Confidence interval calculation:
[0379] 1. Use Monte Carlo dropout techniques to estimate the uncertainty of the predictions.
[0380] 2. Keep the dropout activation during the inference phase and make multiple (e.g. 100) predictions on the same input.
[0381] 3. Calculate the mean and standard deviation of these predictions.
[0382] 4. Assuming the predictions follow a normal distribution, calculate the 95% confidence interval:
[0383] CI 95% =[μ-1.96σ,μ+1.96σ]
[0384] Among them, μ is the predicted mean and σ is the predicted standard deviation.
[0385] The purpose of this step is to integrate the various features and intermediate results obtained in the previous steps into a highly accurate temperature estimation model. By using deep learning techniques, especially LSTM networks, the model is able to capture complex time dependencies and nonlinear patterns. At the same time, by providing confidence intervals, the model not only gives point estimates, but also provides information about the reliability of the estimate, which is extremely important for clinical decision-making. The advantage of this method is that it can adapt to changes in different patients and breathing patterns, providing personalized and highly accurate temperature estimation.
[0386] The specific implementation of step S80 is: smoothing the precise calculated values of lung temperature in a plurality of consecutive time windows to reduce short-term fluctuations. This step is intended to provide a more stable and reliable body temperature estimate while retaining a meaningful temperature change trend.
[0387] First, choose an appropriate smoothing algorithm:
[0388] 1. Use exponentially weighted moving average (EWMA) as the primary smoothing method. The advantage of EWMA is that it gives higher weight to recent data while retaining historical information.
[0389] 2. The calculation formula of EWMA is as follows:
[0390] S t =α·T t +(1-α)·S t-1
[0391] Among them, S t is the smoothed value at time t, T t is the original temperature estimate at time t, and α is the smoothing factor (0<α≤1).
[0392] 3. Selection of smoothing factor α:
[0393] The initial value is set to 0.2, which means that a 20% weight is given to new data.
[0394] Use an adaptive method to dynamically adjust the alpha value:
[0395] When the temperature change rate exceeds a threshold value (eg, 0.1°C / minute), the α value is increased to 0.4 to respond to the temperature change more quickly.
[0396] When the temperature remains stable (rate of change less than 0.02°C / minute) for more than 5 minutes, gradually reduce the α value to 0.1 to provide a smoother output.
[0397] Next, the specific steps to achieve smoothing:
[0398] 1. Initialization: Use the temperature estimate of the first time window as the initial smoothing value S0.
[0399] 2. Iterative calculation: For each new time window t, perform the following operations:
[0400] a. Calculate the current temperature estimate T t With the previous smoothing value S t-1 The difference:
[0401] ΔT=|T t -S t-1 |
[0402] b. Adjust the smoothing factor α according to ΔT:
[0403] If ΔT>0.1℃, then α=min(0.4,α+0.05)
[0404] If ΔT<0.02°C, then α=max(0.1,α-0.01)
[0405] c. Calculate the new smoothing value using the adjusted α:
[0406] S t =α·T t +(1-α)·S t-1
[0407] 3. Outlier processing:
[0408] The median absolute deviation (MAD) method is introduced to detect and handle outliers:
[0409] a. Calculate the median M of the most recent n (e.g. n=20) smoothed values.
[0410] b. Calculate MAD: MAD = median(|S i -M|), where i = t-n+1,...,t.
[0411] c. If |S t -M|>3.5·MAD, then S t Considered as an outlier.
[0412] d. For outliers, use interpolation methods instead:
[0413] S t =0.5·(S t-1 +S t+1 )
[0414] 4. Trend protection mechanism:
[0415] In order to avoid missing important temperature change trends due to excessive smoothing, a trend protection mechanism is introduced:
[0416] a. Calculate the linear regression slope k of the most recent m (e.g., m=10) raw temperature estimates.
[0417] b. If |k|>0.05℃ / min, and the smoothing value S t If the difference from the predicted value of the regression line exceeds 0.1℃, the smoothing value is adjusted:
[0418] S t =0.7·S t +0.3·T t
[0419] 5. Confidence interval adjustment:
[0420] Based on the confidence interval calculated in step S70, the smoothed temperature value is further adjusted:
[0421] a. Calculate the smoothing value S t Compared with the original estimate T t The 95% confidence interval [CIlower ,CI upper ] of the overlapping part.
[0422] b. If the smoothed value falls within the overlapping interval, it remains unchanged.
[0423] c. If the smoothed value falls outside the overlapping interval, adjust it to the nearest interval boundary.
[0424] 6. Output processing:
[0425] In order to provide a more intuitive temperature display, the final smoothed value is rounded:
[0426] a. For normal body temperature range (36.5℃-37.5℃), keep one decimal place.
[0427] b. For abnormal body temperature (<36.5℃ or >37.5℃), keep two decimal places to monitor temperature changes more accurately.
[0428] The purpose of this step is to provide stable and reliable temperature estimation through intelligent smoothing algorithm while retaining important temperature change information. Through adaptive smoothing factor, outlier processing and trend protection mechanism, this method is able to achieve a good balance between smoothness and responsiveness. In addition, the adjustment of confidence interval is combined to ensure the statistical reliability of the smoothing result. This processing method not only improves the accuracy of temperature monitoring, but also provides medical staff with more interpretable and reliable temperature data, which helps to detect abnormal temperature in time and make appropriate clinical decisions.
[0429] The specific implementation of step S90 is: the processed accurate calculated value of the patient's lung temperature is displayed in real time on the display, and the confidence interval is displayed at the same time. This step is intended to present the body temperature data in an intuitive and informative way, providing comprehensive temperature monitoring information for medical staff.
[0430] 1. Data preparation:
[0431] a. Obtain the smoothed temperature value S from step S80 t .
[0432] b. Obtain the corresponding 95% confidence interval [CI lower ,CI upper ].
[0433] c. Calculate the temperature change trend in the last 30 minutes.
[0434] 2. Display interface design:
[0435] a. Main display area: large font displays the current body temperature value S t , accurate to one decimal place.
[0436] b. Trend chart area: displays the temperature change curve in the last 2 hours.
[0437] c. Confidence interval display: The confidence interval is represented by a light-colored band on the trend chart.
[0438] d. Alarm indication area: used to display abnormal temperature alarm.
[0439] e. Status information area: displays device working status, battery power, etc.
[0440] 3. Color Coding System:
[0441] Different colors are used to encode the body temperature value to intuitively reflect the body temperature status:
[0442] Low temperature (<36.0℃): light blue
[0443] Normal (36.0℃-37.5℃): Green
[0444] Mild fever (37.6℃-38.0℃): yellow
[0445] Moderate fever (38.1℃-39.0℃): orange
[0446] High fever (>39.0℃): red
[0447] 4. Real-time data update:
[0448] a. The temperature value and color of the main display area are updated every 5 seconds.
[0449] b. Update the trend chart every 30 seconds.
[0450] c. Confidence intervals are updated every minute.
[0451] 5. Trend analysis and display:
[0452] a. Calculate the temperature change rate in 30 minutes: ΔT 30 =(S t -S t-30min ) / 30
[0453] b. Use arrows on the trend graph to indicate the direction and speed of temperature change:
[0454] Rising arrow (red): ΔT 30 >0.05℃ / min
[0455] Downward arrow (blue): ΔT 30 <-0.05℃ / min
[0456] Horizontal arrow (green): |ΔT 30 |≤0.05℃ / min
[0457] 6. Alarm system:
[0458] Set multiple alarm thresholds to trigger an alarm when the temperature exceeds the normal range:
[0459] a. Mild alarm: When the temperature is between 37.6℃-38.0℃ or 35.5℃-35.9℃, a yellow warning message is displayed.
[0460] b. Moderate alarm: When the temperature is between 38.1℃-39.0℃ or 35.0℃-35.4℃, an orange warning message is displayed and an intermittent beep sounds.
[0461] c. Serious alarm: When the temperature is >39.0℃ or <35.0℃, a red warning message will be displayed, a continuous beep will sound, and an emergency notification will be sent to medical staff.
[0462] 7. Data Reliability Indication:
[0463] Displays the level of confidence in the data based on the width of the confidence interval:
[0464] - High reliability: Confidence interval width < 0.3°C, displayed with a green check mark.
[0465] - Medium reliability: The confidence interval width is between 0.3℃-0.5℃, and a yellow exclamation mark is displayed.
[0466] - Low reliability: Confidence interval width > 0.5°C, a red question mark is displayed, indicating that the sensor position may need to be readjusted.
[0467] 8. Interactive functions:
[0468] a. Touch the trend graph to view detailed historical data.
[0469] b. Swipe the screen to view more statistical information, such as 24-hour average temperature, maximum / minimum temperature, etc.
[0470] 9. Abnormal situation handling:
[0471] a. When a sudden temperature change is detected (such as a change of more than 1°C within 5 minutes), a "Data Abnormal" warning will be displayed and the update of the main display will be suspended until the data is confirmed to be stable.
[0472] b. If no valid data can be obtained for 10 consecutive minutes, a "Signal Lost" warning will be displayed, and it is recommended to check the device connection.
[0473] The purpose of this step is to transform complex temperature data into intuitive and easy-to-understand visual information to help medical staff quickly and accurately understand the patient's temperature status. Through real-time updates, color coding, trend analysis, and a multi-level alarm system, this display solution can effectively present temperature changes and promptly alert abnormal conditions. At the same time, by displaying confidence intervals and data reliability indicators, additional reference information is provided for medical decision-making. This comprehensive and detailed display method not only improves the efficiency and accuracy of temperature monitoring, but also provides strong support for personalized patient care.
[0474] Specifically, the principle of the present invention is: the application of multi-dimensional data fusion and intelligent algorithm. Specifically, the technical principle of the present invention can be explained from the following aspects:
[0475] First of all, the design principle of the M-shaped airbag is based on fluid mechanics and heat conduction theory. The M-shaped structure can form a tiny space between the inner wall of the trachea and the airbag while ensuring the airtightness of the airbag. On the one hand, this design reduces the direct impact of the respiratory airflow on the sensor, and on the other hand, it can maintain sufficient heat exchange area. According to Fourier's law of heat conduction, the heat transfer rate is proportional to the temperature gradient and the contact area. Therefore, this design can not only ensure that the temperature sensor receives sufficient thermal signals, but also reduce the interference of the instantaneous airflow temperature.
[0476] Secondly, the introduction of vibration sensors is based on the principle of respiratory physiology. During breathing, airflow passing through the trachea will produce vibrations of specific frequency and amplitude. By analyzing these vibration signals, the different stages of the respiratory cycle can be accurately identified. The advantage of this method is that it is not affected by changes in airflow temperature, so it can provide stable and reliable information about the respiratory cycle.
[0477] Furthermore, the construction of the heat exchange dynamic model and the temperature gradient model is based on the principles of thermodynamics and computational fluid dynamics. By establishing a heat conduction model between the tracheal wall and the airflow, and combining computational fluid dynamics to simulate the flow and heat exchange process of the airflow in the trachea, the impact of the breathing process on the tracheal wall temperature can be accurately estimated. This method takes into account multiple factors such as airflow velocity, temperature, and humidity, and can effectively eliminate the interference of ventilator airflow on temperature measurement.
[0478] Deep learning technology, especially the application of long short-term memory (LSTM) networks, is based on time series analysis and machine learning principles. LSTM networks have the ability to remember long-term dependencies, making them ideal for processing time series data such as body temperature, which has long-term trends and short-term fluctuations. By training the LSTM model, the system can learn the complex patterns of body temperature changes, thereby providing more accurate body temperature estimates and predictions.
[0479] Finally, the design of the real-time data processing and display system is based on the principles of human-computer interaction and information visualization. Through reasonable data aggregation, smoothing and visualization technology, the system can transform complex temperature data into intuitive and easy-to-understand visual information. The design of the multi-level alarm mechanism is based on the principles of medical statistics, and by setting reasonable thresholds and rules, timely warning of abnormal body temperature can be achieved.
[0480] The following is the principle of using vibration data combined with ventilator airflow temperature to improve the accuracy of body temperature measurement:
[0481] 1. Respiratory cycle identification: Vibration sensors can accurately capture the rhythm of breathing. By analyzing the vibration pattern, the system can identify the inhalation and exhalation phases, which provides a time frame for understanding temperature changes.
[0482] 2. Assessment of airflow temperature impact: The temperature of the gas delivered by the ventilator is usually different from body temperature. During the inspiration phase, the cooler gas temporarily lowers the temperature of the tracheal wall. Vibration data can accurately mark these phases, allowing the system to distinguish between temperature fluctuations caused by ventilator airflow and true body temperature changes.
[0483] 3. Dynamic model of heat exchange: Combining vibration data (reflecting breathing rhythm) and temperature data, a dynamic model of heat exchange between the tracheal wall and breathing gas can be established. This model can predict the impact of breathing on local temperature, thereby more accurately estimating the true body temperature.
[0484] 4. Exhalation temperature analysis: The temperature during the exhalation phase is closer to the core body temperature. By accurately identifying the exhalation phase through vibration data, the system can perform more reliable temperature sampling at these time points.
[0485] 5. Airflow velocity estimation: The vibration intensity may be correlated with the airflow velocity. By analyzing the vibration amplitude, the system can estimate the airflow velocity and thus assess its cooling effect on the temperature sensor and compensate for it in the calculation.
[0486] 6. Temperature gradient analysis: There may be a temperature gradient along the length of the air pipe. Vibration data can help locate the exact position of the sensor, allowing the system to be corrected based on the known temperature distribution model.
[0487] 7. Adaptive filtering: Using vibration data as a reference, the system can implement adaptive filtering to effectively remove the periodic temperature fluctuations caused by the ventilator and extract a more stable baseline body temperature.
[0488] In general, the technical solution of the present invention builds a complete system from data acquisition, signal processing, model calculation to information presentation through the integration of multidisciplinary knowledge. This systematic solution can comprehensively cope with various challenges in endotracheal tube temperature measurement, thereby achieving high-precision and continuous temperature monitoring. Each link of the solution has a solid theoretical foundation and practical basis, which ensures the scientificity and effectiveness of the overall solution.
[0489] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is designed below: A 55-year-old male patient, Mr. Zhang, was admitted to the intensive care unit (ICU) of a tertiary hospital. Mr. Zhang needed mechanical ventilation treatment due to severe pneumonia complicated with acute respiratory distress syndrome (ARDS). The medical team decided to use the endotracheal tube with body temperature measurement function of the present invention to achieve accurate body temperature monitoring and respiratory management.
[0490] Equipment preparation:
[0491] 1. Tracheal tube: Use a hose with an outer diameter of 7.5 mm, suitable for adult male patients.
[0492] 2. Airbag: adopts M-shaped design, and its size parameters are as follows: circumference: 22mm, height: 2.5mm, M-shaped formula parameters: A=1.2mm, B=2π / 4, C=0.5mm, D=4π / 4;
[0493] 3. Temperature sensor: high-precision thermistor, measuring range 30-45℃, accuracy ±0.1℃
[0494] 4. Vibration sensor: piezoelectric sensor, frequency response range 0.1-100Hz
[0495] 5. Single chip microcomputer: 32-bit ARM Cortex-M4 core, main frequency 80MHz
[0496] 6. Display: 5-inch LCD touch screen, resolution 800x480 pixels
[0497] Implementation steps:
[0498] 1. Endotracheal intubation:
[0499] Under the guidance of a video laryngoscope, the anesthesiologist inserts the endotracheal tube into the patient's trachea through the mouth. The intubation depth is 22 cm (starting from the front teeth). After confirming that the tube is in the correct position, air is injected into the balloon until the balloon is completely attached to the inner wall of the trachea.
[0500] 2. Connect to the ventilator:
[0501] Connect the air inlet of the endotracheal tube to the ventilator. The ventilator settings are as follows: Mode: Pressure Controlled Ventilation (PCV); Inspiratory Pressure: 18 cmH2O; Positive End-Expiratory Pressure (PEEP): 8 cmH2O; Respiratory Rate: 16 times / min; Inspiratory to Expiratory Ratio: 1:2; Inspiratory Oxygen Concentration (FiO2): 60%; Humidifier Temperature: 37°C.
[0502] 3. System initialization: Turn on the MCU system, perform self-test and sensor calibration. The display shows the system is ready.
[0503] 4. Data collection and processing:
[0504] The system starts to continuously collect temperature and vibration data and executes each step of the temperature precision calculation module. The following details the specific implementation of each step:
[0505] S10. Data acquisition and preprocessing: Sampling frequency: temperature sensor 100Hz, vibration sensor 1000Hz; Preprocessing: Use low-pass filter to remove high-frequency noise, cut-off frequency is 50Hz; Record ventilator parameters: airflow temperature 37℃, flow rate 30L / min;
[0506] S20. Sliding time window segmentation: initial window length: 5 seconds; use genetic algorithm to optimize window length, and the fitness function considers temperature estimation accuracy and computational efficiency; optimized window length: 4.2 seconds;
[0507] S30. Respiratory cycle identification: Identify the respiratory cycle based on the frequency and amplitude changes of the vibration data. In this embodiment: Inhalation phase: lasts for 1.4 seconds, characterized by high-frequency and low-amplitude vibration; Exhalation phase: lasts for 2.8 seconds, characterized by low-frequency and high-amplitude vibration; Transition phase: 0.2 seconds each, with vibration characteristics between inhalation and exhalation;
[0508] S40. Feature extraction:
[0509] The following features are extracted for each 4.2 second time window:
[0510] Temperature characteristics: average temperature: 36.8℃; temperature change rate: 0.02℃ / s; maximum temperature: 37.1℃; minimum temperature: 36.5℃;
[0511] Vibration characteristics: Vibration frequency: 16Hz for inhalation, 8Hz for exhalation; Vibration amplitude: 0.5g for inhalation, 1.2g for exhalation
[0512] Energy distribution: calculated by fast Fourier transform (FFT);
[0513] Breathing characteristics: exhalation duration: 2.8 seconds; inhalation duration: 1.4 seconds; inhalation-exhalation ratio: 1:2; airflow velocity: inhalation 30L / min, exhalation 25L / min;
[0514] S50. Heat exchange dynamic model:
[0515] Based on the airflow velocity and temperature, the thermal excursion in each time window is calculated: including heat loss during the inspiration phase: 0.15°C and heat gain during the expiration phase: 0.10°C.
[0516] S60. Temperature gradient model:
[0517] The temperature gradient model is constructed using the pre-established tracheal anatomical structure data and CFD model. In this embodiment: the estimated position of the temperature sensor: 10 cm from the carina; the weight of the exhalation phase: 0.7; the weight of the inhalation phase: 0.3; the temperature correction vector: [0.2°C, 0.1°C, -0.1°C, -0.2°C] corresponds to the upper, front, back, and lower directions respectively;
[0518] S70. Temperature accurate calculation:
[0519] Using the pre-trained deep neural network model, the above features are input to obtain the accurate calculation value of lung temperature in each time window. The following are the calculation results of 5 consecutive time windows as shown in Table 1:
[0520] Table 1 Calculation results of 5 consecutive time windows
[0521] Time window Calculated temperature value (℃) 1 37.2 2 37.3 3 37.1 4 37.2 5 37.3
[0522] S80. Smoothing:
[0523] The temperature values of five consecutive time windows are smoothed by moving average to obtain the final lung temperature estimate: 37.2℃
[0524] S90. The results show:
[0525] Real-time update of patient lung temperature on the display: 37.2°C ± 0.1°C (95% confidence interval)
[0526] 5. Clinical application and observation:
[0527] The medical staff closely monitored the real-time temperature data on the monitor. During the first 4 hours after intubation, the following conditions were observed:
[0528] (1) Temperature change trend:
[0529] At the beginning of intubation, the patient's lung temperature was 37.2°C. In the following 2 hours, the temperature slowly rose to 37.5°C, which was related to the patient's inflammatory response and body stress. After that, the temperature tended to stabilize, fluctuating between 37.4-37.6°C.
[0530] (2) Breathing parameter adjustment:
[0531] Based on real-time temperature data, doctors fine-tuned the ventilator parameters: lowered the humidifier temperature from 37°C to 36.5°C to prevent airway overheating; slightly increased the tidal volume from 450ml to 480ml to improve alveolar ventilation; and adjusted the inspired oxygen concentration (FiO2) from 60% to 55% as the patient's oxygenation condition improved.
[0532] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A tracheal tube with a body temperature measurement function, characterized in that: It includes a hose, an airbag, a temperature sensor, a vibration sensor, a single-chip computer, and a display. The airbag is wrapped on the outer wall of the hose, and the part of the hose wrapped by the airbag is provided with one or more through holes. The part of the hose extending out of the airbag is a tube head. The outer wall of the airbag is an M-shape that is rotated around the central axis of the hose. The two ridges of the M-shape are arc-shaped. The valley of the M-shape is provided with a temperature sensor and a vibration sensor. The temperature sensor and the vibration sensor are connected to the single-chip computer through a data line, and the data line is embedded in the outer wall of the hose. The single-chip computer is electrically connected to the display. The single-chip computer is provided with an internal body temperature precision calculation module for calculating the temperature value in the patient's body in real time according to the signals collected by the temperature sensor and the vibration sensor, and displaying it through the display.
2. The endotracheal tube with body temperature measurement function according to claim 1, characterized in that: When the airbag is inflated and fits into the trachea or bronchus of the human lungs, the M-shaped valley of the airbag forms a tiny space with the inner wall of the trachea or bronchus, the M-shaped ridge of the airbag tightly abuts against the inner wall of the trachea or bronchus, and the temperature sensor and the vibration sensor abut against the inner wall of the trachea or bronchus through a layer of fixed membrane; the fixed membrane is attached to the M-shaped valley of the outer wall of the airbag.
3. The endotracheal tube with body temperature measurement function according to claim 2, characterized in that: When the airbag is inflated and fits in the trachea or bronchus of the human lung, the top of the M-shaped ridge of the airbag is 0.5 to 1.5 mm higher than the temperature sensor or the vibration sensor.
4. The endotracheal tube with body temperature measurement function according to claim 3, characterized in that: The outer diameter of the hose is 5 to 8 mm.
5. The endotracheal tube with body temperature measurement function according to claim 4, characterized in that: The M-shape of the airbag is expressed by the formula y=A×sin(B×x)+C×sin(D×x), wherein the size ranges of the variables in the specific formula include: A=0.5~1.5mm, B=2π / 3~2π / 5, C=0.2~0.8mm, D=4π / 3~4π / 5.
6. The endotracheal tube with body temperature measurement function according to claim 5, characterized in that: The opening of the hose away from the airbag is an air inlet for connecting to a ventilator; the opening of the tube head away from the airbag is an air outlet for delivering gas to the patient's lungs.
7. The endotracheal tube with body temperature measurement function according to claim 6, characterized in that: The size range of the M-shaped airbag includes: circumference 15-25mm, height 1-3mm.
8. The endotracheal tube with body temperature measurement function according to claim 7, characterized in that: The temperature sensor and the vibration sensor are arranged opposite to each other.
9. The endotracheal tube with body temperature measurement function according to claim 8, characterized in that: The portion of the hose wrapped by the airbag is provided with four through holes.
10. The endotracheal tube with body temperature measurement function according to claim 1, characterized in that: The body temperature accurate calculation module is used to perform the following steps: S10, collecting raw data from the temperature sensor and the vibration sensor, and preprocessing them to remove noise, to obtain preprocessed temperature data and preprocessed vibration data; and simultaneously recording the airflow temperature and flow rate set by the ventilator; S20, segmenting the preprocessed temperature data and the preprocessed vibration data using a sliding time window, wherein the window length of the sliding time window is optimized by a genetic algorithm; S30, identifying the respiratory cycle based on the preprocessed vibration data, and marking each time window as an inhalation phase, an exhalation phase, or a transition phase; S40, extracting features within each time window, including: Temperature characteristics: average temperature, temperature change rate, maximum and minimum temperatures; Vibration characteristics: vibration frequency, vibration amplitude, energy distribution; Respiratory characteristics, which are the respiratory characteristics of the endotracheal tube with temperature measurement function set for the lungs of patients assisted by a ventilator, including: exhalation time, inhalation time, inhalation-exhalation ratio, and airflow velocity value; S50, constructing a heat exchange dynamic model: calculating the heat excursion of the trachea wall in each time window according to the airflow velocity and airflow temperature, including heat loss and heat gain; S60, constructing a temperature gradient model using the tracheal anatomical structure data and the computational fluid dynamics model, for obtaining temperature estimates at different locations in the trachea according to the respiratory phase, airflow velocity, and temperature sensor position, giving a higher weight to a time window marked as an exhalation phase, and calculating a temperature correction vector according to the estimated position of the temperature sensor in the trachea using the temperature gradient model; S70, using a pre-trained temperature precision calculation model, inputting the following parameters: temperature characteristics and vibration characteristics of each time window, thermal offset calculated by a heat exchange dynamic model, and temperature correction vector calculated by a temperature gradient model, to obtain a precise calculation value of the patient's lung temperature in each time window; S80, smoothing the precise calculated values of lung temperature in a plurality of consecutive time windows to reduce short-term fluctuations; S90. Display the accurately calculated value of the patient's lung temperature in real time on a display, and display the confidence interval at the same time.
Citation Information
Cited By
Core body temperature measuring system and measuring method
CN120938363A
