A decision method and system for the entire process of mechanical ventilation weaning

Abstract

The application provides a decision-making method for the whole process of mechanical ventilation weaning, characterized by comprising the following steps: acquiring various index data of a patient at a first frequency to construct a patient index database; identifying an offline window period through key index data in the offline window period in the patient index database and AI captured dynamic behavior signals; in the offline window period, starting and performing a spontaneous breathing test, acquiring key index data of the offline window at a second frequency, generating a spontaneous breathing test tolerance score, and pushing a suggestion to medical staff on whether to perform weaning based on the spontaneous breathing test tolerance score; after the medical staff wean, monitoring key outcome index data at a third frequency, evaluating the recovery of the patient to obtain an evaluation result, and judging whether to intervene according to the evaluation result. In the whole weaning process, different frequencies are used to acquire corresponding index data at different stages, ensuring accurate grasp of the patient's condition.

Description

Technical Field

[0001] This invention relates to the field of medical decision-making technology, specifically to a decision-making method and system for the entire process of weaning from mechanical ventilation. Background Technology

[0002] Weaning from mechanical ventilation refers to the gradual reduction of mechanical ventilation support after a patient's condition improves, eventually weaning the patient off the ventilator. During weaning, the patient's physiological state undergoes complex changes, requiring a comprehensive consideration of multiple factors to make a sound decision. Inappropriate weaning decisions can lead to respiratory failure and other serious complications if made too early, increasing patient suffering and mortality; conversely, delayed weaning can result in respiratory muscle atrophy, lung infections, and other complications, prolonging hospital stays and increasing medical costs.

[0003] A search revealed that patent number CN103893866B discloses a method for weaning from ventilator use. This method involves first detecting changes in alveolar volume (ΔV), intrapleural pressure (ΔP), and the patient's spontaneous respiratory rate (f). Then, the patient's work of breathing (W) and shallow and rapid breathing index (RSBI) are calculated based on the detected values. Simultaneously, the patient's oxygen saturation (SpO2) and end-tidal carbon dioxide (EtCO2) are monitored. The calculated and monitored values ​​are compared with the ventilator weaning criteria. If the criteria are met, weaning is performed. If the criteria are not met, a second weaning assessment is conducted.

[0004] However, the aforementioned weaning methods have certain limitations. These methods primarily focus on detecting a few physiological indicators and comparing them to fixed weaning criteria, lacking a comprehensive consideration of the dynamic changes in the patient's physiological state throughout the weaning process. In actual mechanical ventilation weaning, the patient's physical condition is constantly changing, and different problems and risks may arise at different stages. Relying solely on a few fixed physiological indicators and a single weaning criterion is insufficient to accurately reflect the patient's true condition throughout the weaning process. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to provide a decision-making method and system for the entire process of weaning from mechanical ventilation, thereby overcoming the limitations of existing weaning methods. The decision-making method of the present invention comprehensively considers the dynamic changes in the patient's physiological state throughout the weaning process, combining multi-dimensional information for integrated analysis, thereby formulating a more scientific, reasonable, and personalized weaning plan.

[0006] The specific technical solution of this application is: a decision-making method for the entire process of mechanical ventilation and weaning, specifically including:

[0007] A patient indicator database was constructed by continuously collecting routine indicator data at the highest frequency.

[0008] The off-the-vir window period is identified using key off-the-vir window indicators from the patient indicator database and AI-assisted identification.

[0009] During the said weaning window, a spontaneous breathing test is initiated and performed. The key weaning index data are collected at a second frequency to generate a spontaneous breathing test tolerance score. Based on the spontaneous breathing test tolerance score, a weaning recommendation is pushed.

[0010] After weaning from ventilator, key outcome indicators are monitored at a third frequency to assess the patient's recovery and determine whether intervention is necessary.

[0011] Furthermore, the offline window period is identified through key offline window indicator data and AI-captured dynamic behavioral signals, including:

[0012] Real-time monitoring of key indicators in the disconnection window, including the first respiratory function indicators, which include inhaled oxygen concentration, positive end-expiratory pressure, and shallow and rapid breathing index. A basic warning is triggered when the inhaled oxygen concentration is ≤40%, the positive end-expiratory pressure is ≤5cmH2O, and the shallow and rapid breathing index is <105 breaths / (min・L).

[0013] The stability of spontaneous breathing is obtained by using AI visual analysis of the patient’s chest rise and fall frequency, limb range of motion, and audio analysis of breath sound characteristics.

[0014] When the basic early warning conditions are met and the spontaneous breathing stability reaches the set threshold, it is determined that the patient has entered the off-ventilator window period.

[0015] Furthermore, during the said weaning window, a spontaneous breathing test is initiated and performed, and the key weaning indicator data are collected at a second frequency to generate a spontaneous breathing test tolerance score, including:

[0016] During the weaning window, key weaning indicator data are collected at a set second frequency. The key weaning indicator data includes a second respiratory function indicator, a first blood gas indicator, and an assessment of the impact of medication. A spontaneous breathing tolerance score is generated based on the above key weaning indicator data.

[0017] Furthermore, the weaning suggestion based on the spontaneous breathing test tolerance score includes: when the spontaneous breathing tolerance score reaches a set threshold of 30-120 minutes, and the score remains stable above the threshold during this period, it is determined that the patient can tolerate the spontaneous breathing test well, and at this time, a weaning suggestion is pushed.

[0018] Furthermore, after weaning from ventilator, key outcome data are monitored at a third frequency to assess the patient's recovery and determine whether intervention is necessary. This includes: real-time monitoring of key outcome data after weaning, including the third respiratory function index, which includes peak cough flow (PCF) and Kubota water swallowing test results. When PCF < 60 L / min or Kubota water swallowing test ≥ 3, an intervention warning is triggered and a corresponding intervention plan is recommended.

[0019] Furthermore, the second frequency is greater than the first frequency and the third frequency.

[0020] Furthermore, the first respiratory function index, the second respiratory function index, and the third respiratory function index are obtained by capturing the ventilator display interface in real time through a binocular camera deployed at the foot of the bed and by using AI vision to identify the ventilator mode and ventilator parameters in the ventilator interface.

[0021] The blood gas parameters are captured in real time by a binocular camera deployed at the foot of the bed by a blood gas analyzer.

[0022] The interface displays various blood and qi data.

[0023] The beneficial effect of this invention lies in that, through comprehensive and dynamic monitoring and analysis of the patient's physiological state throughout the entire weaning process, the decision-making method and system of this invention can effectively overcome the limitations of existing weaning methods. During the entire weaning process, corresponding indicator data are collected at different frequencies at different stages, ensuring an accurate grasp of the patient's condition.

[0024] During the phase of collecting routine indicator data to build a patient indicator database, data was continuously collected at the highest frequency, providing fundamental data support for subsequent analysis. Through real-time monitoring of key indicators within the weaning window and with AI assistance, the weaning window period could be accurately identified, a crucial node in the weaning process. Within the weaning window, a spontaneous breathing test was initiated, and data was collected at a second frequency to generate a spontaneous breathing test tolerance score, providing a scientific basis for pushing weaning recommendations.

[0025] After weaning from ventilator therapy, monitoring key outcome indicators at a third frequency allows for timely assessment of the patient's recovery. If any key outcome indicator shows abnormalities, an intervention alert is triggered, and a corresponding intervention plan is recommended, ensuring the patient's safety and recovery after weaning. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the method described in this application;

[0027] Figure 2-5 This is a schematic diagram of the specific application interface of the method in this application;

[0028] Figure 6 This is the overall discrimination capability curve of the model in this application;

[0029] Figure 7 This is the model confusion matrix of this application.

[0030] Figure 8 This is the model prediction confidence distribution of this application.

[0031] Figure 9 This is a global feature importance distribution map of the model in this application;

[0032] Figure 10 This is a graph showing the global feature importance ratio of the model in this application;

[0033] Figure 11 It is an individual predictive attribution waterfall plot

[0034] Figure 12 This is a system diagram of this application; Detailed Implementation

[0035] Weaning from mechanical ventilation refers to the gradual reduction of mechanical ventilation support after a patient's condition improves, eventually weaning the patient off the ventilator. During weaning, the patient's physiological state undergoes complex changes, requiring a comprehensive consideration of multiple factors to make a sound decision. Inappropriate weaning decisions can lead to respiratory failure and other serious complications if made too early, increasing patient suffering and mortality; conversely, delayed weaning can result in respiratory muscle atrophy, lung infections, and other complications, prolonging hospital stays and increasing medical costs.

[0036] A search revealed that patent number CN103893866B discloses a method for weaning from ventilator use. This method involves first detecting changes in alveolar volume (ΔV), intrapleural pressure (ΔP), and the patient's spontaneous respiratory rate (f). Then, the patient's work of breathing (W) and shallow and rapid breathing index (RSBI) are calculated based on the detected values. Simultaneously, the patient's oxygen saturation (SpO2) and end-tidal carbon dioxide (EtCO2) are monitored. The calculated and monitored values ​​are compared with the ventilator weaning criteria. If the criteria are met, weaning is performed. If the criteria are not met, a second weaning assessment is conducted.

[0037] However, the aforementioned weaning methods have certain limitations. These methods primarily focus on detecting a few physiological indicators and comparing them to fixed weaning criteria, lacking a comprehensive consideration of the dynamic changes in the patient's physiological state throughout the weaning process. In actual mechanical ventilation weaning, the patient's physical condition is constantly changing, and different problems and risks may arise at different stages. Relying solely on a few fixed physiological indicators and a single weaning criterion is insufficient to accurately reflect the patient's true condition throughout the weaning process.

[0038] like Figure 1 As shown, in some embodiments, a decision-making method for the entire process of mechanical ventilation and weaning is disclosed, specifically including:

[0039] A patient indicator database was constructed by continuously collecting routine indicator data at the highest frequency.

[0040] For example, routine indicator data include body mass index, gender, age, diagnosis name, diagnosis code, surgery name, surgery code, duration of mechanical ventilation, etc. These data are all from the hospital's HIS system. Through the real-time data interface with the HIS system, the above basic information and dynamically updated data of the patient since admission can be automatically captured and stored, providing comprehensive patient background information for subsequent weaning decisions.

[0041] The off-the-vir window period is identified using key off-the-vir window indicators from the patient indicator database and AI-assisted identification.

[0042] During the said weaning window, a spontaneous breathing test is initiated and performed. The key weaning index data are collected at a second frequency to generate a spontaneous breathing test tolerance score. Based on the spontaneous breathing test tolerance score, a weaning recommendation is pushed.

[0043] After weaning from ventilator, key outcome indicators are monitored at a third frequency to assess the patient's recovery and determine whether intervention is necessary.

[0044] In this embodiment, the setting of the first frequency is based on extensive clinical practice and research. It can comprehensively collect patients' routine indicator data without increasing the consumption of excessive medical resources. This routine indicator data covers various aspects such as the patient's basic vital signs, medical history, and current treatment status, providing a rich and reliable foundation for subsequent analysis. The patient indicator database constructed through the organization and analysis of this data provides strong support for the entire weaning decision-making process.

[0045] In identifying the weaning window, real-time monitoring of key indicators is a crucial first step. The three primary respiratory function indicators—inspired oxygen concentration, positive end-expiratory pressure, and rapid breathing index—trigger a basic warning when they simultaneously meet specific conditions, indicating a potential weaning opportunity. Furthermore, AI visual analysis of the patient's chest rise and fall frequency, limb range of motion, and spontaneous breathing stability obtained from audio analysis of breath sound characteristics further confirm whether the patient truly meets the criteria for entering the weaning window. Only when both the basic warning conditions and spontaneous breathing stability are met is the patient deemed to have entered the weaning window; this dual-judgment mechanism significantly improves the accuracy of weaning window identification.

[0046] Specifically, when acquiring respiratory function indicators, the AI ​​vision system uses a binocular camera deployed at the foot of the bed to capture the ventilator display interface in real time. Image recognition algorithms and deep learning models are then used to accurately identify the ventilator mode and parameters within the interface. These algorithms, trained on a large amount of clinical data, can quickly and accurately extract key information from complex interface images, ensuring the accuracy and reliability of the acquired data.

[0047] Specifically, after a patient is admitted to the department, the system automatically connects to the core dimensions of offline indicators and synchronizes with the hospital information system to achieve data integration with zero manual data entry.

[0048] Medical records and history: Completely extract key information from the document, including underlying diseases (diabetes, hypertension, COPD, cerebral infarction, ARDS, renal insufficiency, etc.), surgical history, difficult airway, airway inflammation, etc., to clarify the patient's underlying risks;

[0049] Initial assessment upon admission: Automatically extract assessment indicators from documents (BMI, age, gender, disease diagnosis, name of surgery, APACHE II score, GCS score, NRS-2002 nutritional risk screening, SOFA score, CPIS pulmonary infection score) to establish an initial patient profile;

[0050] Dynamic data synchronization: Continuously collect subsequent examination and test data (blood gas analysis, complete blood count, PCT, CRP, etc.), medication records (vasoactive drugs, sedatives and analgesics, neuromuscular blocking agents, glucocorticoids, etc.), vital signs data (heart rate, blood pressure, respiratory rate, SpO2, etc.) and ventilator parameters (PS, PC, PEEP, FiO2, tidal volume, etc.), forming a dynamic database covering all indicators of the documents.

[0051] Based on the "key threshold before offline operation" in the offline metrics, and combined with AI vision / audio technology, intelligent early warning for SBT initiation is achieved:

[0052] Indicator monitoring: Real-time tracking of respiratory function indicators (RSBI shallow and rapid breathing index, MIP maximum inspiratory pressure, peak expiratory flow rate, etc.), vital signs and ventilator parameters. When indicators such as FiO2≤40%, PEEP≤5cmH2O, and RSBI<105 breaths / (min・L) are met, basic warning conditions are triggered.

[0053] AI-assisted judgment: AI visually recognizes patients' spontaneous breathing movements (such as chest rise and fall frequency and limb range of motion) and analyzes respiratory sounds (such as the presence or absence of wheezing and respiratory rhythm stability) to supplement the "dynamic behavioral signals" not covered by the document indicators.

[0054] Healthcare professionals are reminded that when "document metrics meet the standards and AI behavior judgments are correct", an SBT assessment reminder will be automatically pushed to healthcare professionals to avoid missing the early offline opportunity due to human error in overlooking key document metrics.

[0055] During the weaning window, a spontaneous breathing trial is initiated, with key weaning indicator data collected at a second frequency. This second frequency is higher than the first and third frequencies because more frequent patient information is needed during this crucial phase of the spontaneous breathing trial to promptly and accurately assess patient tolerance. The second respiratory function indicator, the first blood gas indicator, and the medication impact assessment among the key weaning indicator data reflect the patient's physical condition during the spontaneous breathing trial from different perspectives. The spontaneous breathing tolerance score generated based on this data provides a scientific and intuitive basis for pushing weaning recommendations. When the score reaches a set threshold and remains stably above the threshold for a certain period, the patient is considered to have good tolerance for the spontaneous breathing trial, and a weaning recommendation is pushed out.

[0056] After entering the spontaneous breathing trial phase, the system compares the dynamic indicators of the weaning indicators with those of the ventilator, and combines AI multimodal analysis to achieve "real-time monitoring + risk warning + tolerance assessment":

[0057] Dynamic indicator tracking: Continuously collect respiratory function indicators (RSBI changes, airway resistance Raw, lung compliance Cdyn, etc.), blood gas indicators (PaO2 / FiO2 ratio, pH value, lactate Lac, etc.), and vital signs. If abnormalities such as heart rate >140 beats / min, systolic blood pressure >180 mmHg or <90 mmHg, or SpO2 <90% occur, a risk alarm will be triggered immediately.

[0058] Medication impact assessment: Based on medication records, analyze the inhibitory effects of sedatives and analgesics (such as propofol and remifentanil) on the cough muscle and the changes in sputum viscosity caused by anticholinergic drugs (such as atropine) (sputum viscosity grading), and quantify the impact of drugs on weaning from mechanical ventilation.

[0059] Feasibility assessment for weaning: The AI ​​integrates the above-mentioned indicator data to generate an SBT tolerance score. If no abnormalities are observed for 30-120 minutes, a "weaning recommendation" prompt is sent to medical staff; otherwise, it is recommended to suspend SBT and adjust the treatment plan.

[0060] After weaning, key outcome indicators were monitored at a third frequency. Although the third frequency is relatively low, it is still sufficient to promptly detect abnormalities in the patient's recovery process. Third respiratory function indicators among the key outcome indicators, such as peak cough flow (PCF) and the Kubota water swallowing test results, are important indicators for assessing the patient's recovery after weaning. When these indicators become abnormal, intervention alerts are triggered, and corresponding intervention plans are recommended, acting like a timely "first aid kit" to ensure the patient's safety and recovery outcome after weaning. The decision-making method for the entire mechanical ventilation weaning process, through multi-stage and multi-dimensional monitoring and analysis, provides patients with personalized and scientifically sound weaning plans, effectively overcoming the limitations of existing weaning methods.

[0061] After extubation, key outcome indicators are used as the core monitoring data to achieve early detection and early intervention of abnormal situations:

[0062] Real-time tracking of respiratory function indicators (PCF peak cough flow rate, Kubota water swallowing test results, sputum characteristics score and sputum volume grade), vital signs, and blood gas parameters; if PCF < 60 L / min (indicating insufficient sputum expectoration capacity) or Kubota water swallowing test ≥ 3 (indicating risk of aspiration), an intervention reminder will be triggered immediately.

[0063] Intervention matching: Based on the type of abnormal indicators, the system automatically recommends corresponding intervention plans (e.g., nebulized expectoration is recommended when sputum is viscous, and adjusting the feeding method is recommended when the risk of aspiration is high), and records the changes in indicators after intervention (e.g., sputum viscosity decreased from grade III to grade II after intervention), forming a closed-loop management system;

[0064] Outcome data recording: Automatically records special events (weaning failure / extubation failure), number of SBT trials, and oxygen therapy methods, accumulating complete data for subsequent weaning factor analysis.

[0065] In some embodiments, an automated detection and recording system for SBT mode is provided. The core of this system is to transform clinical judgment logic into quantifiable and executable rules—using a three-layer screening logic of "hard parameter conditions + process characteristics + clinical scenario verification," while simultaneously connecting to in-hospital system data (ventilators, monitors, electronic medical records) to ensure accurate judgment and eliminate the need for manual intervention. The specific solution is as follows:

[0066] The system captures ventilator data in real time using a binocular camera deployed at the foot of the ICU bed. A system is considered to have "potential SBT activation" if all of the following conditions are met:

[0067] Mode matching: The current mode of the ventilator is PS (pressure support) mode or CPAP (continuous positive airway pressure) mode (excluding high support modes such as SIMV and A / C).

[0068] Parameter thresholds (must be maintained stably for ≥30 minutes to avoid misjudgments due to temporary parameter tuning):

[0069] FiO2 (inhaled oxygen concentration) ≤50%;

[0070] PEEP (Positive End-Expiratory Pressure) ≤ 8 cmH2O;

[0071] Additional requirements for PS mode: PS (supported pressure) ≤ 8cmH2O (CPAP mode has no PS or PS=0);

[0072] Access criteria met (interconnection to monitors / electronic medical records, exclusion of contraindications):

[0073] Stable circulation: Heart rate (HR) 60-140 bpm, systolic blood pressure (SBP) 90-180 mmHg, and no need for high doses of vasoactive drugs (such as norepinephrine > 0.1 μg / kg / min).

[0074] Oxygenation target: SpO2 ≥ 90% (or PaO2 / FiO2 ≥ 150 mmHg, based on blood gas analyzer data);

[0075] Consciousness level: GCS score ≥ 8 (or RASS score ≥ -3, which is used to correlate with sedation scores to avoid misjudgment of comatose patients).

[0076] Second layer: Process feature verification (excluding "false positives" and confirming that it is SBT rather than routine ventilation)

[0077] After the basic conditions are met, the system further verifies the following process characteristics (at least two of them must be met) to avoid misclassifying "long-term low support ventilation" as SBT:

[0078] Mode switching logic: Within 1 hour before startup, the ventilator mode is switched from "high support mode" (such as SIMV, A / C, PS>10cmH2O) to the current low support mode (PS / CPAP).

[0079] Parameter stability: The number of parameter adjustments within 30 minutes is ≤1 (FiO2 / PEEP / PS will not be frequently modified during SBT, but may be repeatedly adjusted during routine ventilation).

[0080] No contraindications: No neuromuscular blocking agents (such as rocuronium bromide) or high-dose sedatives (such as propofol infusion rate > 4 mg / kg / h) were used after initiation (data from the drug administration system).

[0081] Third layer: Clinical scenario evidence (optional, to improve the accuracy of judgment)

[0082] If the system can be integrated with an electronic medical record / medical order system, simply add one of the following items to confirm:

[0083] The doctor issued orders for "Spontaneous Breathing Test (SBT)" and "Weaning Assessment" (keyword matching: SBT, Weaning Test, Weaning Assessment).

[0084] The low support mode was activated for the first time that day (in line with the clinical practice of SBT "once a day screening").

[0085] Automatically record content (SBT file is automatically generated after the system triggers a detection).

[0086] Once the above three-layer rules are met, the system will automatically record the following information without manual input: real-time capture of ventilator parameters (mode, FiO2, PEEP, PS, etc.) and monitor parameters (HR, SBP, SpO2, GCS / RASS scores) via AI vision.

[0087] Optional interfaces include electronic medical records (medical order keywords), medication systems (use of vasoactive drugs and sedatives), and blood gas analyzers (PaO2 / FiO2).

[0088] Configurable thresholds: Supports hospitals / departments to customize parameter thresholds (e.g., PEEP can be relaxed to 10cmH2O in some hospitals) and stabilization time (e.g., SBT in some departments only requires observation for 20 minutes), adapting to different clinical habits.

[0089] Exception handling logic: If the parameters exceed the threshold during SBT (e.g., FiO2 suddenly rises to 60%) or failure indicators appear (e.g., HR > 140 times / minute), the system will automatically determine "SBT termination" and record the termination time and reason;

[0090] If the parameters fluctuate repeatedly within 30 minutes and fail to reach the target level, the system will not trigger SBT recording, but will only mark it as "low support ventilation observation".

[0091] Medical staff confirmation process: After the system automatically generates the SBT record, it is pushed to the nurse station / doctor station. Medical staff can add remarks (such as "the patient tolerates it well, it is recommended to remove the tube the next day") to ensure the completeness of the record. At the same time, it supports manual correction of misjudgments (such as special cases that need to adjust the mode but are not SBT).

[0092] like Figure 2-5The diagram shows the software interface for implementing the above method. These interfaces intuitively display the operation and data presentation at each stage of the system. The interface clearly displays real-time updated ventilator data, monitor data, and electronic medical record data. For automatically detected potential SBT activation, the interface provides a clear indicator for prompt attention from medical staff. Furthermore, configurable information such as threshold parameters and stabilization duration can be adjusted on the interface, allowing different hospitals and departments to personalize settings according to their clinical practices.

[0093] During SBT (Self-Body-Tracking Therapy), the interface dynamically displays monitoring data curves, such as changes in heart rate, blood oxygen saturation, tidal volume, and respiratory rate, allowing medical staff to intuitively understand the patient's condition. If any parameter exceeds the threshold or a failure occurs, the interface will immediately pop up an alert box, displaying the termination time and reason, reminding medical staff to handle the situation promptly.

[0094] The interface provides detailed viewing and editing functions for automatically recorded SBT-related information. Healthcare professionals can view SBT initiation time, ventilator mode and parameters, baseline status at initiation, SBT duration, monitoring data curves, SBT outcomes, termination reasons, and other information. They can also add notes or manually correct misjudgments as needed.

[0095] The following concrete example will further explain the above content:

[0096] Basic information of a patient:

[0097] Body Mass Index: 23.5 kg / m²

[0098] Gender: Male

[0099] Age: 62

[0100] Diagnosis: Community-acquired pneumonia

[0101] Diagnostic code: J18.9

[0102] Surgical name: None

[0103] Surgical code: None

[0104] Mechanical ventilation duration: 48 hours

[0105] Scoring after admission to ICU

[0106] Acute Physiology and Chronic Health Evaluation II (APACHE II) score: 15 points

[0107] Glasgow Coma Scale score: 15 (conscious, not in a coma)

[0108] Nutritional risk screening score: 2 points (low nutritional risk)

[0109] Sequential Organ Failure Assessment (SOFA) score: 4 points (mild organ dysfunction).

[0110] Clinical lung infection score: 6 points (moderate lung infection)

[0111] In the method of this application, disease markers 0 = none and 1 = present are used to identify whether a patient has a specific underlying disease.

[0112] Diabetes: 1

[0113] High blood pressure: 1

[0114] Cerebral infarction: 0

[0115] Chronic obstructive pulmonary disease: 0

[0116] Coronary heart disease: 0

[0117] Acute respiratory distress syndrome: 0

[0118] Renal insufficiency: 0

[0119] Myasthenia gravis: 0

[0120] Respiratory failure: 0 (mild hypoxia, no need for high-concentration oxygen therapy)

[0121] Neck surgery history: 0

[0122] Difficult airway: 0

[0123] Airway inflammation: 1 (Pneumonia accompanied by mild airway inflammation)

[0124] This application uses medication markers 0 = not used and 1 = used to identify whether a patient is using a specific medication.

[0125] Vasopressor use: 0

[0126] Sedative and analgesic use: 1 (low dose of morphine to control mild pain)

[0127] Neuromuscular blocking agent usage: 0

[0128] Glucocorticoid use: 1 (low-dose methylprednisolone, for anti-inflammatory treatment)

[0129] Diuretic use: 0

[0130] Beta-blocker use: 1 (metoprolol, to control baseline hypertension)

[0131] Anticholinergic drug use: 0

[0132] Collected blood gas data (latest values)

[0133] Arterial oxygen partial pressure: 85 mmHg

[0134] Oxygenation index: 280 mmHg (30% inhaled oxygen concentration)

[0135] Arterial oxygen saturation: 95%

[0136] Arterial blood carbon dioxide partial pressure: 38 mmHg

[0137] pH: 7.42

[0138] Alveolar-arterial oxygen partial pressure difference: 45 mmHg

[0139] Bicarbonate: 24 mmol / L

[0140] Serum potassium: 3.8 mmol / L

[0141] Blood glucose: 7.8 mmol / L (for diabetic patients, well controlled)

[0142] Lactate: 1.2 mmol / L (without tissue hypoxia)

[0143] Creatinine: 89 μmol / L

[0144] Blood urea nitrogen: 6.5 mmol / L

[0145] Time series data (last 5 records, array format, unit: corresponding indicator unit, time interval: 1 minute)

[0146] Agitation Index: [0, 0, 0, 0, 0] (No physical movement throughout, stable)

[0147] Pain level: [1, 1, 1, 1, 1] (Mild pain throughout, stable)

[0148] Monitor data:

[0149] Heart rate: [88, 89, 88, 89, 88] (fluctuation < 2 beats / min, stable)

[0150] Pulse rate: [88, 89, 88, 89, 88]

[0151] Systolic blood pressure: [140, 142, 144, 146, 148] (5-minute increase of 5.7%, rapid rise, mild circulatory stress)

[0152] Diastolic blood pressure: [88, 89, 90, 91, 92] (5-minute increase of 4.5%, gradual rise)

[0153] Mean arterial pressure: [105, 107, 108, 110, 111]

[0154] Respiratory rate: [20, 21, 22, 21, 20] (no single trend, fluctuating)

[0155] Blood oxygen saturation: [95, 95, 96, 95, 96]

[0156] Body temperature: [37.3, 37.2, 37.3, 37.4, 37.3] (fluctuation < 0.2℃, stable)

[0157] Ventilator data:

[0158] Ventilation mode: SIMV (Synchronous Intermittent Mandatory Ventilation)

[0159] Tidal volume: [480, 485, 490, 485, 480] ml (stable)

[0160] Inspiratory tidal volume: [482, 486, 491, 486, 482] ml

[0161] Expiratory tidal volume: [478, 483, 488, 483, 478] ml

[0162] Ventilation rate on ventilator: [12, 12, 12, 12, 12] breaths / min (stable)

[0163] Inhaled oxygen concentration: [30%, 30%, 30%, 30%, 30%] (stable)

[0164] Positive end-expiratory pressure: [5, 5, 5, 5, 5] cmH2O (stable)

[0165] Peak pressure: [22, 23, 22, 23, 22] cmH2O (stable)

[0166] Platform pressure: [18, 18, 19, 18, 18] cmH2O (stable)

[0167] Mean airway pressure: [8, 8, 8, 8, 8] cmH2O (stable)

[0168] Inspiratory pressure: [20, 20, 21, 20, 20] cmH2O (stable)

[0169] Expiratory pressure: [5, 5, 5, 5, 5] cmH2O (stable)

[0170] Inhalation-to-exhalation ratio: [1:2, 1:2, 1:2, 1:2, 1:2] (stable)

[0171] Minute ventilation: [5.8, 5.9, 5.9, 5.9, 5.8] L / min (stable)

[0172] Inspiratory flow rate: [60, 60, 60, 60, 60] L / min (stable)

[0173] Expiratory flow rate: [55, 55, 56, 55, 55] L / min (stable)

[0174] Lung compliance: [85, 86, 85, 86, 85] ml / cmH2O (stable)

[0175] Airway resistance: [12, 12, 13, 12, 12] cmH2O / (L·s) (stable)

[0176] Laboratory data (latest values)

[0177] Hemoglobin: 130 g / L

[0178] Albumin: 35 g / L

[0179] Procalcitonin: 0.8 ng / ml (mildly elevated, suggesting infection)

[0180] C-reactive protein: 55 mg / L (moderately elevated, suggesting inflammation)

[0181] White blood cell count: 11.2 × 10⁻⁶ 9 / L (mildly elevated, infection-related)

[0182] Interleukin-6: 35 pg / ml (moderately elevated, indicating an inflammatory response)

[0183] D-dimer: 0.5 mg / L (normal range)

[0184] Activated partial thromboplastin time: 32 seconds (normal range)

[0185] Diagnostic markers in imaging (0 = none, 1 = present)

[0186] Lung infiltrates: 1 (typical manifestation of pneumonia)

[0187] Pleural effusion: 0

[0188] Pulmonary edema: 0

[0189] Atelectasis: 0

[0190] Pneumothorax: 0

[0191] Pulmonary consolidation: 0

[0192] Bronchiectasis: 0

[0193] Pulmonary embolism: 0

[0194] Lung nodules: 0

[0195] Longitudinal septum widening: 0

[0196] After obtaining the above parameters, in order to better assess the feasibility of weaning patients off ventilator therapy, it is necessary to select several important weaning predictors for comprehensive quantitative analysis and construct a scientific weaning prediction model. Specifically, the weight allocation (based on clinical priority, sum ≈ 1) is as follows:

[0197] Highest weighting (0.11~0.13): Oxygenation index (0.13), APACHE II score (0.11), SOFA score (0.12)

[0198] Medium to high weighting (0.07~0.09): Respiratory failure (0.09), use of vasopressors (0.08)

[0199] Medium weighting (0.03~0.06): Heart rate (0.05), respiratory rate (0.06), systolic blood pressure (0.05)

[0200] Low weighting (0.01~0.02): BMI (0.01), age (0.02), serum potassium (0.02)

[0201] Factor score mapping (-1 to 1 point, positive score = favorable for withdrawal, negative score = unfavorable for withdrawal)

[0202] Table 1 Factor Score Mapping Table

[0203] Oxygenation index 280 mmHg Mild to moderate hypoxia, oxygenation is adequate +0.30 APACHE II rating 15 points Moderate condition, not critical -0.25 SOFA rating 4 points Mild organ dysfunction -0.20 respiratory failure 0 (None) Respiratory function is basically stable +0.22 Use of vasopressors 0 (Unused) Cyclic stability +0.25 Heart rate 88~89 beats / min (stable) Heart rate stable, no obvious stress -0.05 respiratory rate 20-22 beats / minute (fluctuation) Slight fluctuations at the normal upper limit. -0.08 systolic blood pressure 140→148 mmHg (rapid increase) Mild cyclic stress response -0.15 BMI 23.5 kg / m² Normal body type, no significant impact on machine removal 0 age 62 years old Elderly patients may experience slightly increased difficulty in weaning from ventilator use. -0.10 serum potassium 3.8 mmol / L Electrolytes are normal, with no disturbances. 0

[0204] Time-series trend assessment (based on 5 data points)

[0205] Heart rate: Stable (variation < 2%)

[0206] Systolic blood pressure: rapid rise (change range 5.7% > 5%)

[0207] Respiratory rate: fluctuating (no single trend)

[0208] Tidal volume, ventilator respiratory rate, etc.: stable (variation range <2%)

[0209] Total score calculation (weighted sum × 100, mapped to a percentage system)

[0210] Calculation process:

[0211] (0.30×0.13) + (-0.25×0.11) + (-0.20×0.12) + (0.22×0.09) + (0.25×0.08) + (-0.05×0.05) + (-0.08×0.06) + (-0.15×0.05) + (0×0.01) + (-0.10×0.02) + (0×0.02)

[0212] = 0.039 - 0.0275 - 0.024 + 0.0198 + 0.02 - 0.0025 - 0.0048 - 0.0075 -0.002≈ 0.011

[0213] Total score = 0.011 × 100 ≈ 11

[0214] 11 corresponds to 72 points on a 100-point scale (low risk, withdrawal assessment can be considered).

[0215] In the above data, the oxygenation index (280 mmHg): 280 indicates mild to moderate hypoxia, and the oxygenation function is still acceptable, which is a core favorable factor for weaning off the ventilator.

[0216] APACHE II score (15 points): A score of 15 points indicates that the patient has a moderate condition, is not critical, and has a slight adverse effect on weaning from mechanical ventilation.

[0217] Systolic blood pressure (140→148 mmHg): A rapid increase from 140 mmHg to 148 mmHg within 5 minutes indicates a mild circulatory stress response and slightly increases the risk of weaning from the machine, requiring close monitoring.

[0218] No vasopressors used: The absence of vasoactive drugs indicates stable circulatory function, which is an important favorable factor for weaning off ventilator.

[0219] Respiratory rate (20-22 breaths / min): The respiratory rate is at the upper limit of normal, with slight fluctuations, no obvious respiratory distress, and has little impact on weaning.

[0220] Total score (72 points): The patient has a low risk of weaning from ventilator, and the weaning assessment process can be initiated. During the assessment, the trend of systolic blood pressure changes and the control of lung infection need to be continuously monitored.

[0221] In some embodiments, the specific steps for processing this data are as follows:

[0222] Data acquisition and preprocessing module:

[0223] (1) Data integrity check: Read patient data and automatically check whether key fields exist. If they are missing, an error will be reported to ensure the validity of subsequent training data.

[0224] (2) Categorical variable encoding: Non-numerical clinical features are digitized. LabelEncoder is used to encode string fields.

[0225] • For example: the gender field (Sex) converts "male" to 0 and "female" to 1; the ventilator mode is also mapped accordingly.

[0226] For non-numerical discrete features (such as sex and ventilator mode), label encoding is used to map them into numerical vectors. Let a certain categorical feature be... Its value space is Encoding mapping function Defined as:

[0227]

[0228] Data cleaning: To maintain the validity of the model input, the system automatically removes sample data containing infinity (Inf, -Inf) or null values ​​(NaN). When facing problems, it is necessary to determine their importance and priority, and manage them in a hierarchical manner. Different levels of problems should be matched with different resources and methods. Small problems do not need to use large processes; large solutions will only lead to waste and inefficiency.

[0229] Feature engineering module:

[0230] This system selects six major categories of clinical features as input: basic features (age, BMI, APACHE II, etc.), comorbidities (hypertension, COPD, etc.), medication information (sedatives, etc.), laboratory tests (PaO2, lactate, etc.), imaging features, and vital signs, totaling 68 features.

[0231] Core Derivative Indicator Calculation: Rapid Shallow Breathing Index (RSBI) This invention specifically calculates RSBI using historical data as an important feature input for prediction.

[0232] Calculation formula:

[0233] in, Respiratory rate (breaths / min). Tidal volume (unit converted to liters).

[0234] Technical principle: This indicator reflects the patient's breathing pattern. If breathing is shallow and rapid (high RSBI), the prediction is more likely to fail; if breathing is deep and slow (low RSBI), the prediction is more likely to succeed. The model will automatically learn the non-linear relationship between this indicator and SBT results.

[0235] Model building and training module:

[0236] Algorithm Selection: XGBoost (Extreme Gradient Boosting) is adopted as the core algorithm. This model consists of Boosting (ensemble learning) and decision trees, essentially a collection of "clinical decision paths" (i.e., the superposition of If-Else logic from multiple trees). Scoring Mechanism: The final prediction result is the sum of the scores from multiple trees. For example, Tree1 scores based on respiratory indicators, Tree2 scores based on circulatory indicators, and the final score is mapped to a probability using a Sigmoid function.

[0237] Let the dataset be The ensemble model includes Trees, Predicted Output Table As shown:

[0238]

[0239] Where F is the regression tree space.

[0240] To train the model, this invention minimizes the following objective function:

[0241]

[0242] First item For the loss function, the binary:logistic loss function is selected in this embodiment for the binary classification task.

[0243] Second item This is a regularization term used to control model complexity and prevent overfitting.

[0244] (2) Training set / test set partitioning: Stratified sampling (stratify=y) is adopted to ensure that the proportion of successful / failed SBT samples in the training set and the test set is consistent, so as to avoid model bias due to sample imbalance.

[0245] In some embodiments, in order to evaluate and validate the established model, such as Figure 6 As shown, this is the overall discrimination ability (ROC curve): The ROC curve shows the relationship between the true positive rate (TPR) and the false positive rate (FPR).

[0246] An AUC close to 1.0 indicates that the model can effectively distinguish between patients who succeeded and failed SBT at various thresholds, demonstrating extremely high diagnostic efficacy.

[0247] In some embodiments, such as Figure 7 As shown, this is the confusion matrix based on the test set. • TP (True Positive): 1134 cases (actual success, predicted success).

[0248] • TN (True Negative): 736 cases (actual failure, predicted failure).

[0249] • FP (False Positive): 116 cases (false positives).

[0250] • False negative (FN): 14 cases (missed diagnoses). Accuracy calculation formula:

[0251]

[0252] The results indicate that the model has extremely high overall accuracy and extremely low false negative rate (only 14 cases of FN).

[0253] like Figure 8 As shown, the prediction confidence distribution is as follows: blue represents actual failure samples, and orange represents actual success samples. The model mainly pushes the probability of failure samples to around 0 and the probability of success samples to around 1, with very little ambiguity in between, proving that the model's judgment is decisive and has high confidence.

[0254] This invention introduces SHAP (SHapley Additive exPlanations) values ​​to attribute model decisions.

[0255] Global feature importance such as Figure 9 and Figure 10 As shown, the model automatically filters out the features that have the greatest impact on the SBT results. According to the results in the figure, 1. peakPressure_4 (airway peak pressure): The SHAP value fluctuates the most and is the most significant factor affecting the results.

[0256] 2. pH (acidity / alkalinity) and SpO2_4 (blood oxygen saturation): followed closely by these.

[0257] 3. Lactate: High lactate levels significantly reduce the success rate.

[0258] In some embodiments, prediction results for a specific patient are shown. The detailed derivation process:

[0259] • Baseline value:

[0260] • Negative contributions (leading to prediction failure): peakPressure_4 (-3.57), lactate (-2.47), ph (-1.99). This indicates that the patient's airway pressure was too high and lactate levels were abnormal, which were the main reasons for the model's failure to determine SBT.

[0261] • Positive contributions (supporting successful prediction): spo2_4 (+2.56), pao2FiO2Ratio (+0.74). Although oxygenation is acceptable, it is insufficient to offset the aforementioned negative factors.

[0262] • Clinical value: Doctors can visually assess the patient's weaning success rate based on this chart. Prioritizing the treatment of airway hypertension and lactic acid buildup is crucial to improving the patient's weaning success rate.

[0263] In summary, this invention achieves accurate assessment of the SBT success rate in mechanically ventilated patients by constructing a weaning prediction model that integrates multi-dimensional clinical features. The model first ensures the integrity and validity of the input data through a data acquisition and preprocessing module, and labels and encodes categorical variables to meet algorithm requirements. The feature engineering module not only integrates 68 basic clinical features but also innovatively introduces core derived indicators such as the Rapid Shallow Breathing Index (RSBI), accurately quantifying the patient's breathing pattern through formulas. The model construction and training module uses the XGBoost algorithm, balancing model performance and complexity by minimizing the loss function and regularization term, and employs stratified sampling to ensure the consistency of sample distribution between the training and test sets. Evaluation results show that the model's overall accuracy is as high as 93.5%, with an AUC close to 1.0, demonstrating excellent discriminative ability. SHAP value attribution analysis further reveals key influencing factors such as peak airway pressure, pH, and lactate, providing interpretable evidence for clinical decision-making. For example, in predicting specific patients, the model can clearly demonstrate the positive and negative contributions of each factor, helping physicians intuitively identify clinical problems requiring priority intervention. This approach, which deeply integrates artificial intelligence technology with clinical practice, not only improves the scientific rigor and efficiency of weaning assessments but also provides strong support for the development of individualized treatment plans. It helps reduce unnecessary mechanical ventilation time, lowers the risk of complications, and ultimately improves patient outcomes.

[0264] In some embodiments, such as Figure 12 As shown, this application provides a decision-making system for the entire process of mechanical ventilation and weaning. The system includes at least one processor and a memory, in which instructions are stored. When the instructions are executed by at least one processor, the steps of the method described in any of the above-mentioned methods are implemented.

[0265] Furthermore, the interface also features data statistics and analysis functions, enabling the organization and analysis of data from multiple SBT trials, providing strong support for subsequent weaning factor analysis. Through in-depth analysis of this data, medical staff can summarize experiences, further optimize weaning plans, and improve the success rate of weaning and patient safety. The entire software interface is simple in design and easy to operate, providing medical staff with efficient and accurate decision support during the mechanical ventilation weaning process.

[0266] The embodiments and functional operations of the subject matter described in this specification can be implemented in the following ways: digital electronic circuits, tangibly implemented computer software or firmware, computer hardware, including the structures disclosed in this specification and their equivalents, or combinations thereof. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, that is, one or more modules of computer program instructions encoded on one or more tangible non-transitory program carriers, for execution by a data processing device or to control the operation of the data processing device.

[0267] Alternatively or additionally, program instructions may be encoded on artificially generated propagation signals, such as machine-generated electrical, optical, or electromagnetic signals, which are then generated as coded information to be transmitted to an appropriate receiver device executed by data processing equipment. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or one or more combinations of the above.

[0268] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather as descriptions of features that can embody specific embodiments of a particular invention. Specific features described in this specification within the context of independent embodiments may also be implemented in combination with individual embodiments. Conversely, various features described within the context of individual embodiments may also be implemented independently.

[0269] The features may be implemented in multiple embodiments or in any suitable sub-combination. Furthermore, although the features may be described as having a combined effect and even initially required in this way, one or more features from the required combination may be removed from the combination in some cases, and the required combination may be redirected to a sub-combination or a variation of the sub-combination.

Claims

1. A decision-making method for the entire process of mechanical ventilation and weaning, characterized in that, Includes the following steps: A patient indicator database was constructed by continuously collecting various indicator data from patients at the highest frequency. The offline window period is identified by using key offline window indicator data from the patient indicator database and AI-captured dynamic behavioral signals. During the weaning window, a spontaneous breathing test is initiated and performed. Key weaning index data are collected at a second frequency to generate a spontaneous breathing test tolerance score. Based on the spontaneous breathing test tolerance score, a recommendation on whether to wean is made to the medical staff. After the medical staff remove the ventilator, the key outcome indicators are monitored at a third frequency to assess the patient's recovery and obtain the assessment results. Based on the assessment results, it is determined whether to intervene.

2. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 1, characterized in that, The offline window period is identified by using key offline window metrics data and AI to capture dynamic behavioral signals, including: Real-time monitoring of key indicators in the disconnection window, including the first respiratory function indicators, which include inhaled oxygen concentration, positive end-expiratory pressure and shallow rapid breathing index. A basic warning is triggered when the inhaled oxygen concentration is ≤40%, the positive end-expiratory pressure is ≤5cmH2O, and the shallow rapid breathing index is <105 breaths / (min・L). The stability of spontaneous breathing is obtained by using AI visual analysis of the patient’s chest rise and fall frequency, limb range of motion, and audio analysis of breath sound characteristics. When the basic early warning conditions are met and the spontaneous breathing stability reaches the set threshold, it is determined that the patient has entered the off-ventilator window period.

3. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 2, characterized in that, During the stated weaning window, a spontaneous breathing test was initiated and performed. Key weaning indicator data were collected at a second frequency to generate a spontaneous breathing test tolerance score, including: During the weaning window, key weaning indicator data are collected at a set second frequency. The key weaning indicator data includes a second respiratory function indicator, a first blood gas indicator, and an assessment of the impact of medication. A spontaneous breathing tolerance score is generated based on the above key weaning indicator data.

4. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 3, characterized in that, The weaning suggestion based on the spontaneous breathing test tolerance score includes: when the spontaneous breathing tolerance score reaches the set threshold for 30-120 minutes, and the score remains stable above the threshold during this period, it is determined that the patient can tolerate the spontaneous breathing test well, and a weaning suggestion is pushed.

5. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 4, characterized in that, After weaning from ventilator, key outcome indicators are monitored at a third frequency to assess the patient's recovery. Based on the assessment results, the decision to intervene is made, including: real-time monitoring of key outcome indicators after weaning, including the third respiratory function indicator, which includes peak cough flow (PCF) and the Kubota water swallowing test results. When the PCF is <60 L / min or the Kubota water swallowing test is ≥3, an intervention warning is triggered and a corresponding intervention plan is recommended.

6. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 5, characterized in that, The second frequency is greater than the first frequency and the third frequency.

7. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 6, characterized in that, The first respiratory function index, the second respiratory function index, and the third respiratory function index are obtained by capturing the ventilator display interface in real time through a binocular camera deployed at the foot of the bed and recognizing the ventilator mode and ventilator parameters in the ventilator interface through AI vision. The blood gas parameters are captured in real time by a binocular camera deployed at the foot of the bed by a blood gas analyzer. The interface displays various blood and qi data.

8. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 3, characterized in that, The first blood gas parameters include: PaO2 / FiO2 ratio, pH value, and lactate (Lac).

9. The decision-making method for the entire process of mechanical ventilation and de-ventilation as described in claim 7, characterized in that, The second frequency is used to collect the key indicators of the off-ventilator test and generate the spontaneous breathing test tolerance score, which is calculated according to the formula SBT_score = (W1 * F_resp + W2 * F_gas + W3 * F_hemo + W4 * F_ai) * F_trend * F_med; Where: SBT_score is the spontaneous breathing test tolerance score, ranging from 0 to 100 points, with a higher score indicating better tolerance; F_resp is a sub-item of the respiratory function score, which is calculated by standardizing secondary respiratory function indicators such as the shallow and rapid breathing index RSBI, respiratory rate RR, and tidal volume Vt according to preset rules. F_gas is a sub-item of gas exchange score, which is calculated from the first blood gas indicators such as PaO2 / FiO2 ratio, pH value, and arterial blood carbon dioxide partial pressure PaCO2 after standardization according to preset rules; F_hemo is a hemodynamic score sub-item, which is calculated from vital signs such as heart rate (HR) and systolic blood pressure (SBP) after standardization according to preset rules. F_ai is a sub-item of AI behavior assessment, which is derived by fusing the breathing pattern stability score obtained from AI visual analysis and the breathing sound quality score obtained from audio analysis. F_trend is a trend adjustment factor. Based on the time series data collected at the second frequency, if the key indicators (such as RSBI, HR) show a stable or improving trend during the evaluation period, then F_trend > 1; if they show a deteriorating trend, then F_trend < 1. F_med is the drug effect adjustment factor. Based on the drug effect assessment, if the current dose of sedative and analgesic drugs is lower than the effect threshold, then F_med ≈ 1; if it is higher than the threshold, then F_med < 1. W1, W2, W3, and W4 are the weight coefficients of each scoring sub-item, and W1+W2+W3+W4=1, among which the respiratory function scoring sub-item W1 has the highest weight.

10. A decision-making system for the entire process of mechanical ventilation and de-ventilation, characterized in that, The system includes at least one processor and a memory storing instructions that, when executed by the at least one processor, perform the steps of the method according to any one of claims 1-9.

Images