Critical patient Pendelfluft automatic detection method based on electrical impedance tomography and artificial intelligence

Abstract

The invention belongs to the field of intelligent medical treatment, and particularly relates to an electrical impedance tomography and artificial intelligence-based critically ill patient Pendelluft automatic detection method. The method comprises the following steps: acquiring EIT data of a lung area of a patient within a period of monitoring time; dividing the lung region into a plurality of regions, and extracting EIT data of the plurality of regions to obtain an EIT signal of each region; obtaining a moisture resistance characteristic based on the EIT signal, wherein the moisture resistance characteristic comprises a moisture resistance amplitude difference in an area and a moisture resistance amplitude difference between adjacent areas; calculating an EIT ventilation characteristic based on the tidal resistance characteristic, wherein the EIT ventilation characteristic is swinging breathing; and based on the EIT ventilation characteristics, judging whether the patient has Pendelluft or not. According to the method, the indexes reflecting whether the Pendelluft exists or not and the serious condition of the Pendelluft are extracted by processing the time sequence data on the basis of the EIT time sequence data obtained in real time, and clinical doctors are assisted in clinical management of mechanical ventilation patients on the basis of real-time monitoring of the Pendelluft.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical care, and more specifically, to a method, device, medium and program product for automatic detection of Pendeluft in critically ill patients based on electrical impedance imaging and artificial intelligence. Background Art

[0002] Pendelluft is defined as asynchronous alveolar ventilation caused by variations in the time constants or dynamic pleural pressures between different lung regions. In patients with obstructive lung diseases (e.g., chronic obstructive pulmonary disease (COPD)), significant differences exist in the time constants of the respiratory system between different lung regions, leading to the pendulum phenomenon. In spontaneously breathing patients with lung injury (e.g., acute respiratory distress syndrome (ARDS)), the uneven transmission of pleural pressure generated by diaphragmatic contraction is the primary cause of the pendulum phenomenon. In extreme cases, such as flail chest, the pendulum volume can be as high as 12.5% of the total volume passing through the airways.

[0003] Because pendelluft may be potentially harmful by introducing local overdistension, tidal recruitment maneuvers, and inflammation, it is necessary to monitor pendelluft so that treatment or ventilation strategies can be adjusted accordingly. Pendelluft was observed during open hemithorax surgery in the early days when technical options were limited. In experimental settings, only ventilatory dyssynchrony between the left and right lungs can be detected. The extent of pendelluft can also be assessed by calculating correlated lung mechanics. Pathological changes in the lungs can be visualized using computed tomography and magnetic resonance imaging, which provide indirect information on regional time constants. In previous studies, positron emission tomography (PET) imaging was used to capture the clearance of the tracer nitrogen-13 to calculate pendulum. In a more recent study, a microphone array was used to detect pendulum. Dark-field microscopy and multispectral oximetry have also been used to assess the effects of sighing on pendulum in experimental settings.

[0004] However, these methods cannot be used to identify and continuously measure pendulums at the bedside. Electrical impedance tomography (EIT) is a novel imaging technique, and in recent years, an increasing number of studies have utilized EIT to detect pendelluft. Currently, EIT can be used to monitor regional lung ventilation in real time, but its clinical real-time monitoring relies heavily on expert interpretation, and clinical implementation is hampered by complex data interpretation.

[0005] Mechanical ventilation remains the cornerstone intervention for patients with acute respiratory failure. However, prolonged mechanical ventilation can lead to various complications and adverse outcomes. The timing and strategy of ventilator weaning are key factors affecting patient prognosis, with approximately 20-30% of patients experiencing difficulty weaning, resulting in prolonged mechanical ventilation and increased mortality. This challenge is particularly evident in patients with acute respiratory distress syndrome (ARDS), where mortality remains high (35-46%) despite advances in mechanical ventilation strategies3.

[0006] Among various respiratory parameters, the pendelluft phenomenon (the movement of air between different lung regions without changing total lung volume) has emerged as a key indicator of ventilation heterogeneity. Detection and quantification of pendelluft can provide valuable insights into regional lung mechanics and guide ventilation strategies. Summary of the Invention

[0007] In view of the above problems, the present invention provides an automatic detection method for Pendelluft in critically ill patients based on electrical impedance imaging and artificial intelligence. By extracting regional ventilation parameters from real-time monitored EIT data, the occurrence and severity of Pendelluft can be monitored in a timely manner.

[0008] The present application (first aspect) discloses a method for automatic detection of Pendelluft in critically ill patients based on electrical impedance imaging and artificial intelligence, the method comprising:

[0009] S101: Acquire EIT data of a patient's lung region during a monitoring period;

[0010] S102: Segmenting the lung region into multiple regions, extracting EIT data of the multiple regions, and obtaining EIT signals of each region;

[0011] S103: obtaining a moisture resistance feature based on the EIT signal, wherein the moisture resistance feature includes: a moisture resistance amplitude difference within a region and a moisture resistance amplitude difference between adjacent regions;

[0012] S104: Calculating an EIT ventilation feature based on the tidal resistance feature, where the EIT ventilation feature is pendulum breathing, and the pendulum breathing is a ratio of a maximum inter-region tidal resistance amplitude difference to an average inter-region tidal resistance amplitude difference;

[0013] S105: Determine whether the patient has Pendelluft based on the EIT ventilation characteristics.

[0014] Furthermore, the swing breathing is expressed as:

[0015]

[0016] Where PI represents pendulum respiration, ΔA i,j represents the difference in tidal resistance amplitude between adjacent regions i and j, max() represents the maximum value, mean(ΔA i,j ) represents the average value of the tidal resistance amplitude differences among all regions;

[0017] Optionally, if the patient has Pendelluft, the EIT ventilation characteristics are input into a grade discrimination classifier to determine the severity of the patient's Pendelluft.

[0018] Furthermore, after S102, the method further includes:

[0019] S102-1: Perform frequency domain transformation on the EIT signals of the respective regions to obtain spectrum signals of the respective regions;

[0020] S103 is replaced by S103': obtaining a spectrum feature based on the spectrum signal, the spectrum feature including: a phase difference between adjacent regions;

[0021] S104 is replaced by S104': obtaining an EIT ventilation feature by calculation based on the frequency spectrum feature, wherein the EIT ventilation feature is a ventilation cycle difference, and the ventilation cycle difference is an average of phase differences between adjacent regions.

[0022] Furthermore, the ventilation cycle difference is expressed as:

[0023]

[0024] Among them, TDI represents the ventilation cycle difference, represents the phase difference between adjacent regions i and j, and N represents the total number of phase differences;

[0025] Optionally, the lung area can be divided into four regions: anterior, posterior, left, and right;

[0026] Optionally, the ventilation cycle difference is expressed as:

[0027]

[0028] Among them, TDI represents the average ventilation cycle difference of 4 adjacent areas. Indicates the phase difference between 4 adjacent areas

[0029] Optionally, the phase difference between the adjacent regions is expressed as:

[0030]

[0031] Among them, X i(f) represents the frequency domain signal of the i-th region; Indicates the phase value corresponding to the main frequency of the spectrum;

[0032] Optionally, if the patient has Pendelluft, the EIT ventilation characteristics are input into a grade discrimination classifier to determine the severity of the patient's Pendelluft.

[0033] Furthermore, the method comprises:

[0034] After acquiring the EIT signals of each region, performing frequency domain transformation on the EIT signals of each region to obtain spectrum signals of each region;

[0035] Obtaining a tidal resistance feature based on the EIT signal, the tidal resistance feature including: a ventilation volume difference between adjacent regions and a time delay between adjacent regions;

[0036] Obtaining spectrum features based on the spectrum signal, the spectrum features including: phase differences between adjacent regions;

[0037] An EIT ventilation feature is calculated based on the spectral feature and the tidal resistance feature, where the EIT ventilation feature is a ventilation synchronization heterogeneity index, and the ventilation synchronization heterogeneity index is obtained based on a standard deviation of a phase difference between adjacent regions, a standard deviation of a ventilation volume difference between adjacent regions, and a mean value of a time delay between adjacent regions;

[0038] Determining whether the patient has Pendeluft based on the ventilation synchronization heterogeneity index;

[0039] Optionally, the ventilation synchronization heterogeneity index is the ratio of the product of the standard deviation of the phase difference between adjacent regions and the standard deviation of the ventilation volume difference between adjacent regions to the mean of the time delay between adjacent regions;

[0040] Optionally, the ventilation synchronization heterogeneity index is calculated as follows:

[0041]

[0042] Wherein, VSHI represents the ventilation synchronization heterogeneity index, represents the standard deviation of the phase difference between adjacent regions, σ V represents the standard deviation of the ventilation differences between adjacent areas, μ t represents the mean of the time delay between adjacent regions;

[0043] Optionally, the phase difference standard deviation is calculated as:

[0044] σ φ =std(Δφ 1,2 ,Δφ 1.3 ,Δφ 2,4 ,Δφ 3,4)

[0045] Among them, σ φ represents the standard deviation of the phase difference between adjacent regions, Δφ 1,2 ,Δφ 1.3 ,Δφ 2,4 ,Δφ 3,4 Represent the phase differences between adjacent regions 1 and 2, 1 and 3, 2 and 4, and 3 and 4, respectively;

[0046] Optionally, the standard deviation of the ventilation differences can be calculated as:

[0047] σ V =std(ΔV 1,2 ,ΔV 1,3 ,ΔV 2,4 ,ΔV 3,4 )

[0048] Among them, σ V represents the standard deviation of the ventilation difference between adjacent areas, ΔV 1,2 ,ΔV 1,3 ,ΔV 2,4 ,ΔV 3,4 represent the ventilation differences between adjacent areas 1 and 2, 1 and 3, 2 and 4, and 3 and 4, respectively;

[0049] Optionally, the mean of the time delays is calculated as:

[0050]

[0051] μ t represents the mean time delay between adjacent regions, Δt 1,2 , Δt 1,3 , Δt 2,4 , Δt 3,4 represent the time delays between adjacent regions 1 and 2, 1 and 3, 2 and 4, and 3 and 4, respectively;

[0052] Optionally, the regional time delay is expressed as:

[0053]

[0054] Where Δt i,j Indicates the regional time delay, represents the time domain ventilation signal after preprocessing in the i-th region, and argmax() represents the weighted average of the peak time of each respiratory cycle in the entire monitoring period;

[0055] Optionally, if the patient has Pendelluft, the EIT ventilation characteristics are input into a grade discrimination classifier to determine the severity of the patient's Pendelluft.

[0056] Furthermore, the method comprises:

[0057] Segment the lung region into anterior and posterior regions;

[0058] Extract the EIT signals of the front and rear areas, and obtain the spectrum signals of the front and rear areas after frequency domain transformation;

[0059] A tidal resistance feature is obtained based on the EIT signal, wherein the tidal resistance feature includes: ventilation volume;

[0060] Obtaining spectrum features based on the spectrum signal, the spectrum features including: phase differences between adjacent regions;

[0061] An EIT ventilation feature is calculated based on the spectrum feature and the tidal resistance feature, wherein the EIT ventilation feature is an optimized phase variation index.

[0062] Optionally, the optimized phase variation index is expressed as:

[0063]

[0064] Where PVI represents the optimized phase variation index, VT represents ventilation volume, and f s is the respiratory rate, Indicates the phase difference of the kth time window, constant 10 3 is the dimension adjustment factor;

[0065] Optional,

[0066]

[0067] arg(X {ant} (f d ,k)) indicates that in the kth time window, the front area is at the dominant frequency f d The complex spectrum value at ;

[0068] arg(X {post} (f d ,k)) indicates that in the kth time window, the front area is at the dominant frequency f d The complex spectrum value at .

[0069] Furthermore, the method comprises:

[0070] The dynamic airflow redistribution coefficient is calculated based on the spectrum characteristics and the moisture resistance characteristics,

[0071] Determining whether the patient has Pendeluft based on the dynamic airflow redistribution coefficient;

[0072] Optionally, the dynamic airflow redistribution coefficient is expressed as:

[0073]

[0074] Among them, the lung area is divided into the front area and the back area, VT ant (k) represents the tidal volume of the front area of the kth time window, VT post (k) represents the tidal volume in the kth time window, represents the phase difference between the front and back areas of the kth time window; M represents the number of selected observation time windows;

[0075] Optionally, one time window is one respiratory cycle, and M is 3.

[0076] The second aspect of the present application discloses a real-time PENDELLUFT monitoring system for mechanically ventilated patients based on EIT, comprising:

[0077] Acquisition module 201: used to acquire EIT data of the patient's lung area during a monitoring period;

[0078] The region segmentation module 202 is used to segment the lung region into multiple regions, extract EIT data of the multiple regions, and obtain EIT signals of each region;

[0079] The data preprocessing module 203 is configured to obtain a moisture resistance feature based on the EIT signal. The moisture resistance feature includes:

[0080] The amplitude difference of moisture resistance within a region and the amplitude difference of moisture resistance between adjacent regions;

[0081] Feature extraction module 204: configured to calculate an EIT ventilation feature based on the tidal resistance feature, wherein the EIT ventilation feature is swing breathing, and the swing breathing is a ratio of a maximum inter-region tidal resistance amplitude difference to an average inter-region tidal resistance amplitude difference;

[0082] Prediction module 205: used to determine whether the patient has Pendeluft based on the EIT ventilation characteristics.

[0083] The third aspect of the present application discloses a computer device, which includes: a memory and a processor; the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, it is used to perform the steps of the above method.

[0084] In a fourth aspect, the present application discloses a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above-mentioned method when the computer program is executed by a processor.

[0085] In a fifth aspect, the present application discloses a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0086] This application has the following beneficial effects:

[0087] (1) The application process begins with the collection of continuous EIT data from bedside monitors. This standardized data is then processed by our system, which performs automated Pendelluft detection and analysis in less than 3 minutes. This streamlined workflow significantly reduces the time and expertise required for EIT data interpretation while maintaining high accuracy and reliability.

[0088] (2) This application is based on real-time acquired EIT time series data. By processing the time series data, ventilation cycle differences and swing breathing are extracted. These indicators reflect the presence and severity of Pendelluft. Based on real-time monitoring of Pendelluft, it assists clinicians in the clinical management of mechanically ventilated patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0090] Figure 1 This is a schematic diagram of the method flow provided by the first aspect of the embodiment of the present invention;

[0091] Figure 2 is a schematic diagram of a program product provided by the second aspect of an embodiment of the present invention;

[0092] Figure 3 is a schematic diagram of a computer device provided by an embodiment of the present invention;

[0093] Figure 4 is a schematic diagram of the architecture of an exemplary computing device provided by an embodiment of the present invention;

[0094] Figure 5 is a schematic diagram of a storage medium provided by an embodiment of the present invention;

[0095] Figure 6 This is a flow chart of analyzing EIT time series data, extracting features, and performing predictions according to an embodiment of the present invention;

[0096] Figure 7A schematic diagram of a process for analyzing EIT time series data, extracting features, fusing the features, and then performing predictions, provided by an embodiment of the present invention;

[0097] Figure 8 This is a performance diagram based on different feature prediction models provided by an embodiment of the present invention;

[0098] Figure 9 This is a schematic diagram of the effectiveness of Pendeluft classification based on a combination of multiple indicators provided by an embodiment of the present invention;

[0099] Figure 10 This is a schematic diagram showing a performance comparison of the indicators proposed in the present invention on Pendeluft predictions provided by an embodiment of the present invention;

[0100] Figure 11 is a schematic diagram of lung region division provided by an embodiment of the present invention;

[0101] Figure 12 4 is a schematic diagram of a calculation flow of a moisture resistance characteristic provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0102] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions 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.

[0103] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as S101, S102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit "first" and "second" to be different types.

[0104] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0105] Figure 1The present invention provides a flowchart of a method for automatically detecting Pendeluft in critically ill patients based on electrical impedance imaging and artificial intelligence. Specifically, the method includes the following steps:

[0106] S101: Acquire EIT data of a patient's lung region during a monitoring period;

[0107] S102: Segmenting the lung region into multiple regions, extracting EIT data of the multiple regions, and obtaining EIT signals of each region;

[0108] S103: obtaining a moisture resistance feature based on the EIT signal, wherein the moisture resistance feature includes: a moisture resistance amplitude difference within a region and a moisture resistance amplitude difference between adjacent regions;

[0109] S104: Calculating an EIT ventilation feature based on the tidal resistance feature, where the EIT ventilation feature is pendulum breathing, and the pendulum breathing is a ratio of a maximum inter-region tidal resistance amplitude difference to an average inter-region tidal resistance amplitude difference;

[0110] S105: Determine whether the patient has Pendelluft based on the EIT ventilation characteristics.

[0111] This application is based on a study: Study design and setting: A retrospective multicenter study conducted in three tertiary hospitals in China from January 2020 to December 2024.

[0112] Participants: Of 458 screened patients, 278 mechanically ventilated patients who met the inclusion criteria were enrolled.

[0113] Main outcomes and measures: Primary outcomes included diagnostic accuracy (sensitivity, specificity, and AUC-ROC) and time efficiency. Secondary outcomes included cost-effectiveness and the correlation between pendelluft severity and clinical outcomes.

[0114] The current multicenter study aimed to: (1) develop and validate an automated pendelluft detection and grading system using EIT data, (2) compare its diagnostic accuracy and cost-effectiveness with those of conventional machine learning methods and expert assessment, and (3) investigate the relationship between pendelluft patterns detected by ChatGPT and clinical outcomes in mechanically ventilated patients.

[0115] Methods: Study design and patient selection: A multicenter retrospective diagnostic study was conducted in three tertiary hospitals in China (Shanghai Ruijin Hospital, Peking Union Medical College Hospital, and Fujian Provincial Hospital in Fuzhou) between January 2020 and December 2024.

[0116] Methods: The STARD and STROBE guidelines were followed. Of the 458 patients initially screened, 278 adult patients (≥18 years) receiving mechanical ventilation in pressure support mode and monitored by EIT were enrolled. Patients were excluded if they had incomplete EIT recordings, poor signal quality (signal quality index <90%), chest wall deformity, severe hemodynamic instability (norepinephrine >0.5 μg / kg / min), or incomplete follow-up data. The study protocol was approved by the institutional review boards of all participating centers.

[0117] EIT data collection and processing were performed using PulmoVista500 equipment ( EIT measurements were performed at a 16-electrode strip positioned in the fourth intercostal space according to a pre-defined protocol. Data were recorded at 50 Hz during 10-minute intervals of stable ventilation and continuously monitored for quality. Preprocessing included removal of cardiac-related impedance variations, motion artifact correction, bandpass filtering (0.1–1.0 [ZZ1] Hz), and normalization to a 32 × 32 pixel matrix. Signal quality was strictly controlled by signal-to-noise ratio assessment (minimum threshold >10 dB) and electrode contact verification (requiring >90% good contact), as validated in previous studies.

[0118] We combine deep learning capabilities with specialized ventilation pattern analysis. The model architecture begins with an EIT image encoder that processes regional ventilation distribution, capturing both controlled and spontaneous breathing patterns. This initial processing stage quantifies sway amplitude by analyzing the percentage of tidal volume shifts between non-dependent and dependent lung regions. The core analysis component, employing a six-layer Transformer encoder and a specialized attention mechanism, enables comprehensive detection of regional ventilation asynchrony. This is complemented by a classification head that generates a pendelluft severity rating with an associated confidence score. Time series analysis capabilities are particularly important for distinguishing between different ventilation cycles and identifying subtle patterns of ventilation heterogeneity. The model was systematically trained and validated using a comprehensive dataset, and its performance was benchmarked against traditional methods. This integrated approach ensures reliable pendelluft detection while maintaining real-time analysis capabilities for clinical decision support.

[0119] Pendelluft assessment and classification: According to a validated method, the pendelluft amplitude was quantified as the impedance difference between the sum of all regional impedance changes and the global impedance change. According to established criteria, patients were classified into four grades: Grade 0 (no pendelluft, <2.5% TIV), Grade 1 (mild, 2.5-5% TIV), Grade 2 (moderate, 5-10% TIV), and Grade 3 (severe, >10% TIV). This grading system was validated by consensus of clinical experts and correlation with clinical outcomes.

[0120] Statistical Analysis: Sample size calculations were based on comparisons of AUC values using the DeLong method. Statistical analyses were performed using R version 4.0.0 (R Foundation for Statistical Computing, Vienna, Austria). Model performance was assessed using a composite diagnostic accuracy metric, calibration assessment using the Hosmer-Lemeshow test, and decision curve analysis. Statistical significance was assessed using the McNemar test with a Bonferroni correction (adjusted significance threshold P < 0.01). Missing data were handled using multiple imputation techniques.

[0121] Results: Of 458 screened patients, 278 were recruited from three tertiary hospitals, of whom 245 (88.1%) had complete data sets suitable for final analysis. Baseline characteristics stratified by pendelluft severity included a median age of 62.4 years (IQR: 51-73), with 156 (63.7%) men. Compared with patients with low-grade pendelluft (n=156), patients with high-grade pendelluft (n=122) exhibited significantly worse baseline characteristics, including a higher APACHE II score (21.3±5.6 vs 17.5±4.2, P<0.001), a lower PaO2 / FiO2 ratio (180±42 vs 220±48, P<0.001), and a higher respiratory rate (28±6 vs 22±5 / min, P<0.001).

[0122] Clinical outcome analysis showed a significant association between pendelluft severity and patient prognosis. The 28-day mortality rate gradually increased from 9.0% for low-grade pendelluft to 25.4% for high-grade pendelluft (adjusted HR 2.8, 95% CI: 1.6-4.9, P = 0.001). Kaplan-Meier survival analysis showed that high-grade patients had significantly worse survival (log-rank P < 0.001). Early detection and intervention of 82 patients with grade 2-3 pendelluft improved oxygenation (mean P / F ratio improved by 52 ± 18 mmHg, P < 0.001) and shortened ventilation time (median reduction of 2.8 days, P = 0.002).

[0123] Compared with traditional methods, the system of the present application showed excellent diagnostic performance, with an overall accuracy of 89.6% (95% CI: 86.2-92.4%) for pendelluft detection, a sensitivity of 88.7% (95% CI: 84.9-91.8%), and a specificity of 90.2%.

[0124] (95% CI: 87.1-92.8%). The system's AUC-ROC was 0.91 (95% CI: 0.87-0.95), significantly outperforming traditional machine learning methods. 2 7, including CNN (AUC 0.85, P = 0.003), random forest (AUC 0.82, P < 0.001) and SVM (AUC 0.79, P < 0.001). Model calibration analysis showed good agreement between predicted and observed probabilities (Hosmer-Lemeshow test, P = 0.87). Performance analysis of specific centers showed that the model accuracy was consistent across hospitals, with AUCs ranging from 0.89 to 0.92 (P = 0.38 for inter-center comparison). The implementation of the system in this application significantly improved workflow efficiency, shortening the average analysis time from 12.5 minutes to 2.8 minutes and saving RMB 850 per patient (95% CI: RMB 620-1080).

[0125] The strong correlation between pendelluft severity and clinical outcomes goes beyond previous observations. While earlier studies have demonstrated an association between ventilation heterogeneity and mortality, our findings reveal a specific threshold effect. The presence of pendelluft during the T-piece trial was associated with significantly higher mortality in patients with difficult weaning (28-day mortality: 37.8% vs 11.1%, p = 0.014). Importantly, we demonstrate that early detection and intervention can alter this trajectory. This finding supports the growing evidence supporting personalized mechanical ventilation strategies, as highlighted by recent multicenter trials. A key advantage of our system is its robust performance across diverse clinical settings and patient populations. Due to differences in clinical practice and patient characteristics, previous AI implementations often exhibit significant performance degradation when applied across centers. Our model maintained consistent performance across three centers (AUC range, 0.89-0.92) and across different patient subgroups, likely due to our comprehensive training approach that incorporates center-specific variations, as recommended by recent AI validation guidelines. The system's integration into clinical workflow deserves particular attention. While previous studies have focused primarily on diagnostic accuracy, we demonstrated significant improvements in efficiency (analysis time reduced from 12.5 to 2.8 minutes) and cost-effectiveness (savings of RMB 850 per patient). This is consistent with recent health economic analyses of AI implementation in critical care, but importantly, our study provides detailed workflow integration metrics, including uptime (99.4%) and false alarm rate (3.2%), addressing key implementation issues raised by Wang et al.

[0126] The analysis process of monitoring pendulum breathing based on real-time EIT data collected in this application mainly includes the following key stages:

[0127] Phase 1: EIT data collection

[0128] A continuous data sequence is acquired through the EIT imaging device, which is represented as: I(1), I(2)…I(t), where I(y) represents the EIT data acquired at the t-th moment. During the acquisition of the EIT image data, the temporal continuity and quality of the image sequence are ensured.

[0129] In some embodiments, the starting moment of the time window is marked as 1, and the t-th moment is marked as t;

[0130] Phase II: Time series analysis of the collected EIT data ( Figure 12 )

[0131] 2.1 First, segment the lung area into regions 1, 2…, n,…, N;

[0132] In some embodiments, N=2, and the lung region is only divided into gravity-dependent and gravity-independent regions.

[0133] In some embodiments, N=2, and the lung region is divided only into anterior and posterior regions.

[0134] In some embodiments, N=4, the lung region is divided into front, back, left, and right regions ( Figure 11 shown).

[0135] 2.2 Extract the time series signal from the segmented area to obtain the EIT data series of each area.

[0136] Time signal series X i (1),X i (2)…X i (t) represents the EIT signal of the i-th region;

[0137] represents the EIT ventilation signal of the i-th region, which is calculated as follows:

[0138]

[0139] Among them, 0~T represents a ventilation cycle.

[0140] 2.3 After processing the EIT data sequence using the Fourier analysis method, the frequency signals of each region are obtained: X 1(f) , X 2(f) ,…,X i(f) ,…,X N(f) , where X i(f) The frequency domain signal of the i-th region is represented by the frequency domain signal of the i-th region;

[0141] Since time signals and frequency signals are two different perspectives of the same signal, time signals show how a signal evolves over time t, while frequency signals show the composition of the signal at frequency.

[0142] 2.4 Quantifying ventilation cycle characteristics:

[0143] By analyzing the spectrum, we can:

[0144] ① Identify the main ventilation frequency: find the maximum peak in the power spectrum to obtain the patient's basic ventilation frequency

[0145] ②Calculate signal amplitude: Measure the difference between the peak and valley values of the signal to reflect the Impedance Tidal Variation Index (ITVI):

[0146]

[0147] Among them, max(X n [f vent ]) represents the maximum value of the moisture resistance signal in the nth area, min(X n [f vent ]) represents the minimum value of the moisture resistance signal in the nth area, max(X {global}} represents the maximum value of moisture resistance signal in all areas;

[0148] ③ Extract phase information: Analyze the phase angle of the signal to reflect the ventilation timing characteristics:

[0149] ④ Determine cycle duration: Calculate the duration of a complete ventilation cycle based on the main frequency:

[0150] ⑤VT represents tidal volume; Unit: mL

[0151] Where V(t) represents the real-time flow rate (measured by the ventilator or estimated by the derivative of the EIT impedance change); the integration time range is one basic unit: a single natural breathing cycle T cycle , from the start of the inspiratory phase to the start of the next inspiratory phase; achievement standard: 3 consecutive stable cycles (cycle variation coefficient <10%); the time window is expressed as [t0, t0+3T cycle ]

[0152] The tidal volume of the i-th region is expressed as:

[0153] 2.5 Calculate the differences between different regions:

[0154] In some embodiments, the lungs are divided into anterior-posterior and left-right regions, and the following are calculated separately:

[0155] ① Difference in tidal resistance amplitude between regions: Calculate the difference between the average tidal resistance amplitudes in different regions to reflect the uneven distribution of ventilation:

[0156] The difference between the peak values of the signals in the adjacent regions i and j is called the moisture resistance amplitude difference. represents the amplitude difference of the time domain signal in the i-th region, that is, the peak-to-valley value,

[0157] ②Inter-regional phase difference: Analyze the phase difference of ventilation waveforms in different regions to reflect ventilation synchronization

[0158]

[0159] Among them, X i(f) represents the frequency domain signal of the i-th region; Indicates the phase value corresponding to the main frequency of the spectrum;

[0160] ③ Time delay: Measure the time difference between different areas reaching ventilation peak, reflecting the inconsistency of ventilation timing:

[0161] The regional time delay is expressed as:

[0162]

[0163] in, represents the time-domain ventilation signal of the i-th region after preprocessing, where preprocessing includes denoising, signal enhancement and other preprocessing methods; argmax represents the weighted average of the peak time of each respiratory cycle in the entire monitoring period;

[0164] ④Different ventilation volume between regions:

[0165] ΔV i,j =VT i -VT j

[0166] VT i represents the ventilation volume of the i-th region within the respiratory cycle, VT j represents the ventilation volume of the jth region within the respiratory cycle;

[0167] Ventilation peak positioning method: The determination of ventilation peak must meet the following requirements:

[0168] Ⅰ) Preprocessing: Bandpass filtering of the original EIT signal (0.1-5 Hz to retain the respiratory-related frequency band)

[0169] II) Peak detection:

[0170] Use sliding windows (3s length, 50% overlap) to find local maxima

[0171] Ensure that the peak interval is consistent with the physiological respiratory cycle range (0.15-0.3Hz)

[0172] III) Main frequency calibration: When the main frequency in the frequency domain is f max When it is inconsistent with the main peak in the time domain, the peak value in the time domain shall prevail;

[0173] The third stage: feature extraction and fusion

[0174] 3.1 Extract key indicators:

[0175] ① The ventilation cycle difference index (TDI) is obtained by calculating the cumulative phase difference between adjacent lung regions. The specific calculation formula is:

[0176]

[0177] Where i and j represent adjacent areas. This indicator can effectively reflect the inconsistency of ventilation timing between different areas.

[0178] In some embodiments, the lungs are divided into four quadrants / regions, and the average phase difference between adjacent regions is calculated to obtain the ventilation cycle difference of the four regions:

[0179]

[0180] Lung regional divisions Figure 11 shown.

[0181] In some embodiments, the phase difference calculation steps include:

[0182] First, perform FFT transformation (1024 points, Hanning window) on the EIT signal of each ROI;

[0183] Secondly, identify the phase value corresponding to the spectrum peak (phase angle of the main frequency point);

[0184] Then, the absolute value of the difference between the main frequency phase angles of adjacent ROIs is calculated;

[0185] Finally, the TDI is calculated by averaging all adjacent pairs.

[0186] In some embodiments, the lung region is divided into four regions: front, back, left, and right. When calculating TDI, adjacent regions include:

[0187]

[0188] The TDI is then obtained by averaging the phase differences of the same pair.

[0189] ② The pendulum breathing index (PI) is obtained by calculating the maximum inter-region signal peak difference and the average of the peak-to-valley values of all regions, that is:

[0190] The pendulum breathing is expressed as:

[0191]

[0192] in, represents the difference in moisture resistance peak value between adjacent regions i and j, represents the amplitude difference of moisture resistance in the i-th region, that is, the peak-to-valley value, mean(ΔA i,j ) represents the average of the inter-regional amplitude peak-to-trough values of all regions;

[0193] ③ Furthermore, this application developed a ventilation synchronization heterogeneity index (VSHI), whose calculation model integrates the dual dimensions of timing asynchrony (Δφ / Δt) and amplitude heterogeneity (ΔV), and the formula is defined as:

[0194]

[0195] Wherein, VSHI represents the ventilation synchronization heterogeneity index,

[0196] Phase difference standard deviation:

[0197]

[0198] Standard deviation of ventilation difference:

[0199] σ V =std(ΔV 1,2 ,ΔV 1,3 ,ΔV 2,4 ,ΔV 3,4 )

[0200] Average delay:

[0201]

[0202] When divided into four quadrants (front, back, left, right), corresponding to Figure 11 The area in (1-F, 2-B, 3-R, 4-R); first, calculate all adjacent combinations:

[0203] -Front and left

[0204] -Front and right

[0205] -Back and left

[0206] -Back and right

[0207] In some embodiments, the calculated average phase difference is shown in the following table:

[0208] Regional pairing Anatomical proximity (yes / no) Average phase difference (rad) FL yes 0.32 FR yes 0.28 BL yes 0.41 BR yes 0.37 Facebook no N / A

[0209] 2.VSHI calculation rules include all parameter differences of these 4 pairs of combinations

[0210] In some embodiments, the calculation code of VSHI is as follows:

[0211]

[0212]

[0213] Clinical validation showed that in a cohort of 50 ARDS patients, the indicator's independent prediction AUC for the Pendeluft phenomenon reached 0.83 (p<0.001). When combined with traditional respiratory parameters, the AUC increased by 6 percentage points to 0.89.

[0214] ④ In addition, this application introduces the dynamic redistribution coefficient (DRC), which is constructed by multiplying the ratio of the tidal volume in the front area to the posterior area and the phase difference to form the expression:

[0215]

[0216] Among them, the lung area is divided into the front area and the back area, VT ant (k) represents the tidal volume in the kth time window, VT post (k) represents the tidal volume in the kth time window, Indicates the phase difference between the front and back areas;

[0217] In some embodiments, one time window is one respiratory cycle, M is the number of observation time windows, and M is 3.

[0218] ⑤ Improve the algorithm of phase variation index and propose the optimized phase variation index (PVI) in the form of time integral:

[0219]

[0220] PVI unit: mL, where VT represents tidal volume;

[0221] Where V(t) represents the real-time flow rate (measured by the ventilator or estimated by the derivative of the EIT impedance change); the integration time range is one basic unit: a single natural breathing cycle T cycle , from the starting point of the inhalation phase to the starting point of the next inhalation;

[0222] Achievement standard: 3 consecutive stable cycles (cycle variation coefficient <10%)

[0223] The time window is expressed as [t0,t0+3T cycle ]

[0224] It represents the phase difference of the kth time window, and its calculation is expressed as:

[0225] VT is tidal volume (unit: mL);

[0226] f s is the respiratory rate (times / minute);

[0227] Constant 10 3 is the dimension adjustment factor (rad·min / mL→standard unit)

[0228] in,

[0229] arg(X {ant} (f d ,k)) indicates that the front area is at the dominant frequency f d The complex spectrum value at (calculated by STFT);

[0230] f d The range is: 0.2-0.35Hz (normal adult respiratory rate range)

[0231] The calculation of CSI (ventilatory synchronization heterogeneity index) requires the combination of phase difference and amplitude heterogeneity.

[0232] The formula is:

[0233]

[0234] The parameters are:

[0235] α=0.6,β=0.4 (container weighting coefficient)

[0236] γ=1.2,k=0.8 (nonlinear conversion index)

[0237] Δφ ij =|φ i (f d )-φ j (f d )∣

[0238] ΔV ij =|V i (f d )-V j (f d )∣

[0239] Verification benchmark: Respiratory cycle variation must be eliminated within the time window integration (phase difference spectrum smoothing)

[0240] Time alignment is ensured by cubic spline interpolation

[0241] Implementation of early warning: When f d If the frequency falls within 0.05-0.2 Hz (heart rate band), the reset mechanism will be automatically triggered;

[0242] CSI calculation is disabled when the impedance signal-to-noise ratio is <15dB. This improved metric not only retains the phase difference metric but also captures the cumulative effect of timing through integration. In a comparative study of 45 cases, the new algorithm improved the positive predictive value by 27%, achieving a cross-validated AUC of 0.81 (p < 0.001).

[0243] The fourth stage is prediction based on extracted features

[0244] In some embodiments, the features obtained in the third stage are input into a machine learning model to predict the occurrence of Pendelluft, and a validation framework including multi-center clinical data is established, and the key performance indicators are shown in Table 1.

[0245] Table 1 Performance analysis of the indicators proposed in this application

[0246]

[0247]

[0248] As can be seen from the table, these indicators have passed rigorous validation and statistical testing and may have good efficacy in clinical applications. In this patent solution, the newly designed indicators demonstrated excellent predictive ability for the presence and severity of Pendelluft in the training set.

[0249] The four core indicators VSHI (AUC=0.83), DRC (AUC=0.78), TDI (AUC=0.77) and CSI (AUC=0.81) all passed rigorous validation with a sample size of 30 cases. The Delong test showed that the AUC of each indicator was significantly greater than the 0.7 benchmark.

[0250] After integrating the above indicators, the comprehensive model achieved AUC = 0.91 (95% CI: 0.87-0.95) in a multicenter validation of 278 cases, and its macro-average AUC reached 0.89 (Table 2&S4 of the research paper).

[0251] The graded validation showed that the AUC for differentiation between Grade 0 and Grade 3 groups was 0.94, and the AUC for differentiation between Grade 2 and Grade 3 groups was 0.82. All comparisons were statistically significant (ANOVA intergroup differences p < 0.001) ( Figure 9 ).

[0252] Clinical validation further confirmed that, when using the dual criteria of VSHI ≥ 10 and model confidence ≥ 0.5, the sensitivity for severe Pendelluft reached 89.6% (CI: 85.3-93.2%). Sample analysis confirmed good calibration (Brier score = 0.15) using the Hosmer-Lemeshow test. All validation processes were consistent with the hybrid ResNet-50 + BiLSTM architecture design.

[0253] In some embodiments, the predictive performance of the new indicators proposed in this application was independently verified, as shown in Table 2.

[0254] Table 2 Verification performance of indicators for the presence of Pendelluft

[0255]

[0256] In some embodiments, the significance of the new indicators proposed in this application in distinguishing between groups was determined through inter-group analysis of the proposed new indicators, as shown in Table 3.

[0257] Table 3 Verification of differences between groups (ANOVA test results)

[0258]

[0259] In some embodiments, the model performance of the new indicators proposed in this application for predicting the presence and severity of Pendelluft was verified, as shown in Table 4.

[0260] Table 4 Predictive validity verification of new indicators in the training set

[0261]

[0262] In some embodiments, the above features are fused and then predicted using a deep learning model, including: (1) constructing a feature matrix: organizing the ventilation cycle features and swing breathing features into a unified matrix form; (2) feature fusion processing. The system fuses the above basic indicators with other relevant features to construct a comprehensive feature vector F. This feature vector not only contains direct features such as TDI and PI, but also contains the interaction features between them and the pattern features that change over time. The dimension of the fused feature vector is N×1, where N represents the total number of features. To ensure the comparability of the features, the system will standardize the fused features. (3) High-order feature extraction. This step uses a deep learning method to further process the fused features. First, the fused feature vector F is input into a deep neural network, and the hidden layer features H=DNN(F) are obtained through multi-layer nonlinear transformation. Then, the system introduces the attention mechanism and weights and enhances the features through H'=Attention(H) to highlight the contribution of important features. Finally, the high-order features are mapped to the prediction probability space through the Softmax function to obtain the final prediction result Y=Softmax(H'). This process establishes an end-to-end mapping relationship from raw physiological signal features to predicted results.

[0263] This multi-level feature extraction and fusion approach offers significant advantages: it not only captures all aspects of pendulum breathing but also automatically learns the complex relationships between features through deep learning, improving prediction accuracy and reliability. Furthermore, the introduction of an attention mechanism enables the model to adaptively focus on the most relevant features, further enhancing system performance. This design provides an accurate and reliable technical solution for pendulum breathing assessment in clinical practice.

[0264] In some embodiments, the extracted indicators, the ventilation cycle difference index (TDI) and the pendulum breathing index (PI), are combined into a basic feature vector, and the interaction features between them (such as TDI×PI) are calculated to form an extended feature set. The features are then expanded in the spatiotemporal dimension, that is, the changing trends of the indicators are analyzed in the time dimension (such as calculating the statistical features within a short time window), and the correlation between regions is analyzed in the spatial dimension (such as calculating the feature correlation of adjacent regions). Finally, all features are weightedly fused using the learned weight matrix W to obtain the final fused feature vector F=W⊙[basic features, interaction features, spatiotemporal features], where the weight matrix W is optimized through model training.

[0265] The results of feature fusion are then input into a specially designed deep learning model, which uses a deep network structure consisting of input layer, convolution layer, pooling layer and fully connected layer, and can effectively process the complex relationship between temporal features and image features.

[0266] Once the model training is complete, the system enters the prediction and evaluation phase. In this phase, the model makes predictions on new data while also performing result verification and performance evaluation to ensure the reliability of the predictions.

[0267] Finally, the system presents the analysis results through various visualization methods, including heatmaps and time curves, and combines them with clinical data for comprehensive prognostic analysis, providing strong support for medical decision-making. This multi-level analysis approach not only improves the accuracy of pendulum breathing detection but also provides a more comprehensive basis for clinical decision-making.

[0268] Since this application is used for bedside monitoring and processes EIT data acquired in real time, the feature extraction in this application is often based on the extraction of the respiratory cycle, or for the sake of accuracy, a time window of three respiratory cycles is used for monitoring to balance the impact of noise. People in this field should understand that this application can process real-time EIT data for bedside monitoring.

[0269] Figure 3 is a schematic diagram of a computer device provided by an embodiment of the present invention, such as Figure 3As shown, the device 2000 may include: one or more processors 2010, and one or more memories 2020; wherein the memories store computer-readable codes, and when the computer-readable codes are run by the one or more processors, they may execute the method described above.

[0270] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. It can implement or execute the various methods, operations, and logic block diagrams disclosed in the embodiments of the present disclosure. The general-purpose processor can be a microprocessor or any conventional processor, etc., and can be an X86 architecture or an ARM architecture.

[0271] In general, various example embodiments of the present disclosure may be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. When various aspects of the disclosed embodiments are illustrated or described as block diagrams, flow charts, or using some other graphical representation, it will be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0272] For example, the method or apparatus according to the embodiment of the present disclosure may also be implemented by Figure 4 The architecture of the computing device 3000 shown in FIG. Figure 4 As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage device in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the method provided in the present disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 4 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 4 One or more components of a computing device are shown.

[0273] The embodiment of the present invention further provides a computer-readable storage medium, such as Figure 5As shown, it is a schematic diagram of a storage medium 4000 provided in an embodiment of the present invention, and computer-readable instructions 4010 are stored on the computer storage medium 4020. When the computer-readable instructions 4010 are executed by the processor, the method according to the embodiment of the present disclosure described with reference to the above figures can be executed. The computer-readable storage medium in the embodiment of the present disclosure can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. The non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory. It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0274] The present disclosure also provides a computer program product or a computer program, which implements the steps of the above method when executed by a processor, such as Figure 2 As shown, the computer program product or computer program includes:

[0275] Acquisition module 201: used to acquire EIT data of the patient's lung area during a monitoring period;

[0276] The region segmentation module 202 is used to segment the lung region into multiple regions, extract EIT data of the multiple regions, and obtain EIT signals of each region;

[0277] The data preprocessing module 203 is configured to obtain a moisture resistance feature based on the EIT signal. The moisture resistance feature includes:

[0278] The amplitude difference of moisture resistance within a region and the amplitude difference of moisture resistance between adjacent regions;

[0279] Feature extraction module 204: configured to calculate an EIT ventilation feature based on the tidal resistance feature, wherein the EIT ventilation feature is swing breathing, and the swing breathing is a ratio of a maximum inter-region tidal resistance amplitude difference to an average inter-region tidal resistance amplitude difference;

[0280] Prediction module 205: used to determine whether the patient has Pendeluft based on the EIT ventilation characteristics.

[0281] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0282] In general, various example embodiments of the present disclosure may be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. When various aspects of the disclosed embodiments are illustrated or described as block diagrams, flow charts, or using some other graphical representation, it will be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0283] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0284] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0285] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0286] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0287] The exemplary embodiments of the present disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will appreciate that various modifications and combinations may be made to these embodiments or their features without departing from the principles and spirit of the present disclosure, and such modifications should fall within the scope of the present disclosure.

Claims

1. A method for automatic detection of Pendeluft in critically ill patients based on electrical impedance imaging and artificial intelligence, characterized in that: The method comprises: S101: Acquire EIT data of a patient's lung region during a monitoring period; S102: Segmenting the lung region into multiple regions, extracting EIT data of the multiple regions, and obtaining EIT signals of each region; S103: obtaining a moisture resistance feature based on the EIT signal, wherein the moisture resistance feature includes: a moisture resistance amplitude difference within a region and a moisture resistance amplitude difference between adjacent regions; S104: Calculating an EIT ventilation feature based on the tidal resistance feature, where the EIT ventilation feature is pendulum breathing, and the pendulum breathing is a ratio of a maximum inter-region tidal resistance amplitude difference to an average inter-region tidal resistance amplitude difference; S105: Determine whether the patient has Pendelluft based on the EIT ventilation characteristics.

2. The automatic detection method for critically ill patients based on electrical impedance imaging and artificial intelligence according to claim 1 is characterized in that: The pendulum breathing is expressed as: Where PI represents pendulum respiration, ΔA i,j represents the difference in tidal resistance amplitude between adjacent regions i and j, max() represents the maximum value, mean(ΔA i,j ) represents the average value of the tidal resistance amplitude difference among all regions; Optionally, if the patient has Pendelluft, the EIT ventilation characteristics are input into a classifier to determine the severity of the patient's Pendelluft.

3. The automatic detection method for critically ill patients based on electrical impedance imaging and artificial intelligence according to claim 1 is characterized in that: The method further includes, after S102: S102-1: Perform frequency domain transformation on the EIT signals of the respective regions to obtain spectrum signals of the respective regions; S103 is replaced by S103': obtaining a spectrum feature based on the spectrum signal, the spectrum feature including: a phase difference between adjacent regions; S104 is replaced by S104': obtaining an EIT ventilation feature by calculation based on the frequency spectrum feature, wherein the EIT ventilation feature is a ventilation cycle difference, and the ventilation cycle difference is an average of phase differences between adjacent regions.

4. The automatic detection method for Pendeluft of critically ill patients based on electrical impedance imaging and artificial intelligence according to claim 3, characterized in that: The ventilation cycle difference is expressed as: Among them, TDI represents the ventilation cycle difference, represents the phase difference between adjacent regions i and j, and N represents the total number of phase differences; Optionally, the lung area can be divided into four regions: anterior, posterior, left, and right; Optionally, the ventilation cycle difference is expressed as: Among them, TDI represents the average ventilation cycle difference of 4 adjacent areas. Represents the phase difference of 4 adjacent areas Optionally, the phase difference between the adjacent regions is expressed as: in, represents the phase difference between adjacent regions i and j, X i(f) represents the frequency domain signal of the i-th region; Indicates the phase value corresponding to the main frequency of the spectrum; Optionally, if the patient has Pendelluft, the EIT ventilation characteristics are input into a classifier to determine the severity of the patient's Pendelluft.

5. The automatic Pendeluft detection method for critically ill patients based on electrical impedance imaging and artificial intelligence according to claim 1, characterized in that: The method comprises: After acquiring the EIT signals of each region, performing frequency domain transformation on the EIT signals of each region to obtain spectrum signals of each region; Obtaining a tidal resistance feature based on the EIT signal, the tidal resistance feature including: a ventilation volume difference between adjacent regions and a time delay between adjacent regions; Obtaining spectrum features based on the spectrum signal, the spectrum features including: phase differences between adjacent regions; An EIT ventilation feature is calculated based on the spectral feature and the tidal resistance feature, where the EIT ventilation feature is a ventilation synchronization heterogeneity index, and the ventilation synchronization heterogeneity index is obtained based on a standard deviation of a phase difference between adjacent regions, a standard deviation of a ventilation volume difference between adjacent regions, and a mean value of a time delay between adjacent regions; Determining whether the patient has Pendeluft based on the ventilation synchronization heterogeneity index; Optionally, the ventilation synchronization heterogeneity index is the ratio of the product of the standard deviation of the phase difference between adjacent regions and the standard deviation of the ventilation volume difference between adjacent regions to the mean of the time delay between adjacent regions; Optionally, the ventilation synchronization heterogeneity index is calculated as follows: Wherein, VSHI represents the ventilation synchronization heterogeneity index, represents the standard deviation of the phase difference between adjacent regions, σ V represents the standard deviation of the ventilation differences between adjacent areas, μ t represents the mean of the time delay between adjacent regions; Optionally, the phase difference standard deviation is calculated as: in, represents the standard deviation of the phase difference between adjacent regions, Represent the phase differences between adjacent regions 1 and 2, 1 and 3, 2 and 4, and 3 and 4, respectively; Optionally, the standard deviation of the ventilation differences can be calculated as: s V =std(ΔV 1,2 ,ΔV 1,3 ,ΔV 2,4 ,ΔV 3,4 ) Among them, σ V represents the standard deviation of the ventilation difference between adjacent areas, ΔV 1,2 ,ΔV 1,3 ,ΔV 2,4 ,ΔV 3,4 represent the ventilation differences between adjacent areas 1 and 2, 1 and 3, 2 and 4, and 3 and 4, respectively; Optionally, the mean of the time delays is calculated as: μ t represents the mean time delay between adjacent regions, Δt 1,2 , Δt 1,3 , Δt 2,4 , Δt 3,4 represent the time delays between adjacent regions 1 and 2, 1 and 3, 2 and 4, and 3 and 4, respectively; Optionally, the regional time delay is expressed as: Where, Δt i,j Indicates the regional time delay, represents the time domain ventilation signal after preprocessing in the i-th region, and argmax() represents the weighted average of the peak time of each respiratory cycle in the entire monitoring period; Optionally, if the patient has Pendelluft, the EIT ventilation characteristics are input into a classifier to determine the severity of the patient's Pendelluft.

6. The automatic Pendeluft detection method for critically ill patients based on electrical impedance imaging and artificial intelligence according to claim 5, characterized in that: The method comprises: Segment the lung region into anterior and posterior regions; Extract the EIT signals of the front and rear areas, and obtain the spectrum signals of the front and rear areas after frequency domain transformation; A tidal resistance feature is obtained based on the EIT signal, wherein the tidal resistance feature includes: ventilation volume; Obtaining spectrum features based on the spectrum signal, the spectrum features including: phase differences between adjacent regions; An EIT ventilation feature is calculated based on the spectrum feature and the tidal resistance feature, wherein the EIT ventilation feature is an optimized phase variation index. Optionally, the optimized phase variation index is expressed as: Where PVI represents the optimized phase variation index, VT represents ventilation volume, and f s is the respiratory rate, Indicates the phase difference of the kth time window, constant 10 3 is the dimension adjustment factor; Optional, arg(X {ant} (f d ,k)) indicates that in the kth time window, the front area is at the dominant frequency f d The complex spectrum value at ; arg(X {post} (f d ,k)) indicates that in the kth time window, the front area is at the dominant frequency f d The complex spectrum value at .

7. The automatic Pendeluft detection method for critically ill patients based on electrical impedance imaging and artificial intelligence according to claim 6, characterized in that: The method comprises: The dynamic airflow redistribution coefficient is calculated based on the spectrum characteristics and the moisture resistance characteristics, Determining whether the patient has Pendeluft based on the dynamic airflow redistribution coefficient; Optionally, the dynamic airflow redistribution coefficient is expressed as: Among them, the lung area is divided into the front area and the back area, VT ant (k) represents the tidal volume of the front area of the kth time window, VT post (k) represents the tidal volume in the kth time window, represents the phase difference between the front and back areas of the kth time window; M represents the number of selected observation time windows; Optionally, one time window is one respiratory cycle, and M is 3.

8. A computer device, characterized in that: The device comprises: a memory and a processor; the memory is used to store a computer program; and the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images