Colorectal cancer stoma nursing data processing method and device

Abstract

The invention discloses a colorectal cancer stoma nursing data processing method and device, and relates to the technical field of nursing data processing, and the method comprises the steps: carrying out the time alignment and abnormality elimination of a stoma part sensor signal, image data and subjective feedback of a patient, and forming initial data; the reading is corrected by using the environmental parameters, cross-source verification is carried out, and calibration verification data is output; then, individual baseline features are constructed, the comprehensive risk amount is calculated, and a self-adaptive threshold evaluation record is generated in combination with a first-order iteration dynamic threshold strategy; and when the risk quantity exceeds a threshold value, an alarm is triggered, and the nurse locally verifies and returns a correction factor, so that the model is gradually adaptive. The method can improve the recognition and response efficiency of acute problems such as skin injury, is more accurate in use in various clinical scenes and patient cases, and remarkably improves the stoma nursing quality.

Description

Technical Field

[0001] The present invention relates to the technical field of nursing data processing, and specifically to a method and device for processing colorectal stoma nursing data. Background Art

[0002] In recent years, the incidence of colorectal cancer has been on the rise year by year, and the number of patients undergoing stoma surgery has increased accordingly. For such patients, the skin and mucosa around the stoma are in a highly sensitive state, and any carelessness may lead to skin damage, infection or other complications, thus affecting postoperative rehabilitation and quality of life. Traditional stoma care methods mostly rely on regular inspections by nursing staff and experience-based judgments, making it difficult to capture early signs such as leakage or minor bleeding in a timely and accurate manner. At the same time, with the development of sensor technology and information-based medicine, comprehensive processing means based on multi-source data (such as blood stain sensors, image monitoring devices, and patient self-perceived feedback) have gradually been applied in clinical practice, with the expectation of achieving continuous monitoring and personalized intervention for stoma care. However, how to efficiently fuse these heterogeneous data and dynamically adjust according to different patient states remains a challenging technical issue.

[0003] In a Chinese invention patent with the application publication number CN118448063A, an information-based intelligent nursing system for urostomy patients is disclosed, which relates to the technical field of medical nursing. The system can monitor the urine data and stoma data of urostomy patients in real time through a patient monitoring module and a machine learning module. These monitored data can be processed and analyzed by a machine learning model to quickly and accurately capture any abnormal changes. Through a monitoring and warning module, the system sets preset thresholds for comparing and evaluating the monitored data. Once any abnormal situation exceeding the preset threshold is detected, the system will immediately send out a warning message. This timely alarm mechanism enables nursing staff to take actions quickly, intervene and treat according to the specific conditions of the patients, effectively reducing the risks faced by the patients, improving the nursing efficiency and the overall safety of the patients. In addition, since the system can monitor and evaluate the physiological conditions of the patients in real time, medical staff can adjust the treatment plan more timely.

[0004] In the existing stoma care monitoring solutions, the most prominent problem is the lack of a processing mechanism that can calibrate multi-source data in real time and accurately map individual differences. Specifically, when a nursing system attempts to integrate sensor readings, image information, and patient self-evaluation feedback, it often encounters interference caused by sensor drift, environmental interference, and patient physical differences, resulting in difficulties in ensuring the accuracy and stability of monitoring results. At the same time, if the risk assessment threshold remains unchanged, it is very easy to have situations of over-alarm or missed alarm, and it cannot meet the dynamic needs of postoperative patients at different rehabilitation stages.

[0005] To this end, the present invention provides a method and device for processing colorectal stoma care data. Summary of the Invention

[0006] (1) Technical problems to be solved

[0007] In view of the deficiencies of the prior art, the present invention provides a method and device for processing colorectal stoma care data. By aligning the sensor signals, image data, and subjective feedback of the patient at the stoma site in time and removing abnormal data, initial data is formed; then, the readings are corrected using environmental parameters and cross-source verification is performed to output calibrated and verified data; subsequently, individual baseline characteristics are constructed and the comprehensive risk amount is calculated, and an adaptive threshold evaluation record is generated in combination with the first-order iterative dynamic threshold strategy; when the risk amount exceeds the threshold, an alarm is triggered, and the nursing staff verifies locally and returns the correction factor, enabling the model to gradually adapt. It can improve the recognition and response efficiency of acute problems such as skin damage, and become more accurate with increasing use in various clinical scenarios and patient cases, significantly improving the quality of stoma care, thereby solving the technical problems described in the background art.

[0008] (2) Technical solutions

[0009] To achieve the above objectives, the present invention is realized through the following technical solutions:

[0010] A method for processing colorectal stoma care data includes:

[0011] Align the sensor signals at the stoma site in time, combine with images and subjective feedback, remove obvious abnormal data, and summarize the corresponding timestamps and sensor identifiers together to generate stoma initial data, completing the acquisition and preliminary integration of multi-source sensor data;

[0012] Construct a calibration factor from environmental parameters to correct the readings of the sensor array, compare with images and subjective feedback of the patient, screen out suspected failure or conflict signals, output the calibrated and verified reliable data in the multi-source calibration verification record, and explicitly save the attention factor, calibration coefficient, and conflict sensor list;

[0013] Construct individual baseline characteristics based on the multi-source calibration verification record, use the comprehensive risk formula to quantify the possibility of skin damage, and dynamically adjust the alarm threshold in combination with the first-order iterative method to generate an adaptive threshold evaluation record; including: the change process of the dynamic threshold at each moment, the comparison result of the risk value and the threshold, and whether the skin damage risk marker has been triggered;

[0014] When the comprehensive risk amount R(t k ) exceeds the corresponding dynamic threshold Γ(t k) When it is triggered, an alarm is issued and recorded in the alarm care closed-loop record. The nursing staff then conducts intervention and local verification based on this. If it is confirmed as a real skin injury or a false alarm, a correction factor is transmitted back to dynamically optimize the model and the threshold strategy.

[0015] Furthermore, real-time signals are obtained from the sensor array at the stoma position, and at the same time, local stoma images output by the imaging device and subjective feedback data of the patient are collected; a global timestamp is uniformly added to all signals and feedback; the collected original information is uniformly classified into the original data record, including the timestamp, sensor identification, sensor value, and basic patient identification elements.

[0016] At the same timestamp t k the values obtained from multiple sensors are compared. Among them,

[0017] assuming there are n effective sensors, then for the i-th sensor at timestamp t k the value S i (t k ) is calculated as follows; let S i (t k ) represent the reading of the i-th sensor at timestamp t k , and the control data S j (t j ) of n moments or sensor combinations are taken, where j ∈ [1, n]. According to the formula, the data abnormality degree M i (t k ) is defined; when the data abnormality degree M i (t k ) exceeds the corresponding threshold Θ, the marked abnormal data is subject to secondary verification;

[0018] Furthermore, the preliminary integrated data of the initial stoma data is read, which includes all sensor readings, image data, and subjective feedback data of the patient; environmental parameters are selected from the preliminary integrated data to construct a calibration factor: Λ k = ξ·f(H k , T k , P k ), where ξ is a constant coefficient and f is a multivariate function based on stoma care experience and observed data; the sensor reading S i (t k ) is calibrated for the environment to obtain: where Ψ(·) is a non-linear mapping function.

[0019] Furthermore, the sensor readings in the stoma environment calibration data are called and compared with the image data and patient feedback data. For the i-th and j-th sensors at timestamp t k , the cross-conflict amount Ω ij (tk )

[0020]

[0021] wherein, v ij is the attention factor, cos(·, ·) can calculate the cosine similarity after taking the inner product of the feature vectors of each sensor. When the cross-conflict volume Ω ij (t k ) continuously stays in the high range, if the image shows no obvious leakage trace while the sensor reading abnormally increases, it is determined that the sensor reading fails.

[0022] Furthermore, the sensor readings determined to be credible through cross-validation are stored in the post-sensor-verification data, and the failed or conflicting readings are marked; after integrating the stoma environment calibration data and the post-sensor-verification data, for each time stamp t k and sensor i, the final credible readings, associated image identifiers, and possible anomaly marks are generated; the integration results are uniformly output to the multi-source calibration verification record, including the calibrated value time stamp t k , sensor identifier i, and verification result;

[0023] Read the verified sensor readings and corresponding time stamps from the multi-source calibration verification record, combine with the patient's previous stoma care records, summarize and statistically analyze the current environmental information (such as room temperature, humidity, pressure, etc.) to form an environmental profile record; use a multi-dimensional mapping function to integrate it into the patient individual baseline feature vector Bp, B p and store it in the baseline feature information in the form of a data structure.

[0024] Furthermore, obtain the calibration value of sensor number i at time stamp t k in the multi-source calibration verification record Call B in the baseline feature information p , apply the non-linear mapping Φ(·) to , multiply it with the corresponding weight ω i and accumulate. The comprehensive risk volume R(t k ) is defined as follows:

[0025]

[0026] In the formula: represents the calibrated and verified valid value of the i-th sensor at time stamp t k ; δ ij is the cooperation factor between the i-th and j-th sensors, Θ p (B p ) is the baseline correction function, It is a class of binary non - linear mappings;

[0027] For each time stamp t k The risk value R(t k ) is temporarily stored in the stoma comprehensive risk score data in a data structure form, and necessary marks are retained.

[0028] Furthermore, after obtaining the comprehensive risk quantity R(t k ), the alarm threshold is dynamically optimized. The threshold target mapping function χ(·) is introduced as the target value and is defined as follows:

[0029] Γ(t k+1 ) = Γ(t k ) + v[χ(R(t k )) - Γ(t k )]

[0030] In the formula: Γ(t k ) represents the alarm threshold at the current time stamp t k , χ(R(t k )) is a class of threshold target mapping functions:

[0031] χ(r) = Γ base + λg(r)

[0032] Where Γ base is the basic threshold, λ is the sensitivity coefficient, g(·) can be any function that maps the risk quantity to the corresponding interval; v is the learning rate or update rate; t k+1 is the time stamp of the next moment;

[0033] For any t k , if R(t k ) > Γ(t k ), it is determined that there is a significant risk of skin damage, marked for alarm, and each comparison result and judgment identifier are stored in the adaptive threshold record.

[0034] Furthermore, read the comprehensive risk quantity R(t k ) and the corresponding dynamic threshold Γ(t k ) at the current time stamp t k , and at the same time obtain the over - limit mark, indicating whether R(t k ) > Γ(t k ) has been satisfied;

[0035] For all records where R(t k ) > Γ(t k) In the case of , enter the state of waiting for an alarm. According to the actual usage environment, a predefined multi-channel notification mechanism can be called to send an alarm signal; write this alarm and its trigger timestamp t into the alarm record data object, including the corresponding sensor set number, risk value, threshold, and whether the alarm is automatically or manually confirmed.

[0036] Furthermore, when the caregiver receives the alarm instruction triggered in step 401, they need to check and intervene in the patient's stoma area, collect new clinical observation results, the patient's subjective description, and nursing operation items.

[0037] If it is found that the alarm is a false alarm, mark the false alarm in the local verification record and explain the reason; if it is a real skin injury, carry out targeted nursing; integrate the nursing operation and the key point information of local verification into the local inspection data.

[0038] A data processing device for colorectal stoma care, including

[0039] A multi-source data integration module, which aligns the sensor signals of the stoma area in time, combines images and subjective feedback, eliminates obvious abnormal data, and summarizes it together with its corresponding timestamp and sensor identifier to generate initial stoma data, completing the acquisition and preliminary integration of multi-source sensor data.

[0040] A multi-source cross-verification module, which constructs a calibration factor from environmental parameters to correct the readings of the sensor array, compares them with images and the patient's subjective feedback, screens out suspected failure or conflict signals, outputs the calibrated and verified reliable data in the multi-source calibration verification record, and explicitly saves the attention factor, calibration coefficient, and conflict sensor list.

[0041] A risk assessment module, which constructs individual baseline characteristics based on the multi-source calibration verification record, quantifies the possibility of skin injury using a comprehensive risk formula, combines a first-order iterative method to dynamically adjust the alarm threshold, and generates an adaptive threshold assessment record; including: the change process of the dynamic threshold at each moment, the comparison result of the risk value and the threshold, and whether the skin injury risk marker has been triggered.

[0042] An alarm trigger feedback module, when the comprehensive risk amount exceeds the corresponding dynamic threshold, triggers an alarm and records it in the alarm nursing closed-loop record. The caregiver conducts intervention and local verification based on this. If it is confirmed as a real skin injury or a false alarm, a correction factor is returned to dynamically optimize the model and threshold strategy.

[0043] (III) Beneficial effects

[0044] The present invention provides a data processing method and device for colorectal stoma care, having the following beneficial effects:

[0045] 1. Uniformly encapsulate and label valid sensor readings, image information, and patient feedback, format and store multi-source data, which can lay the foundation for subsequent refined calibration and risk analysis. Relying on a unified timestamp and pre-denoising of heterogeneous data, the original data can be initially cleaned and the basic structure can be ensured to be consistent.

[0046] 2. Provide a high-confidence, traceable set of sensor and image information for risk assessment. By making full use of environmental information and multi-source verification means, more reliable sensor data is obtained, avoiding the interference of single sensor failure or noise on the overall analysis.

[0047] 3. Enable subsequent risk calculations to be personalized according to the sensitivity and complication history of different patients, avoiding one-size-fits-all threshold settings. Non-linearly weight and couple multiple sensor readings for calculation, and can combine methods such as synergy factors and non-linear mapping to improve the sensitivity and accuracy of anomaly detection, achieving multi-dimensional risk assessment.

[0048] 4. Introduce an adaptive threshold algorithm to iteratively update the alarm threshold Γ(t k ) and generate an adaptive threshold record; the alarm threshold can be flexibly adjusted in different nursing scenarios and patient states, reducing false alarms and missed alarms, enhancing intelligence and adaptability; based on individual differences and real-time environmental parameters, a dynamic risk model is implemented, which can adaptively give a warning about the possibility of stoma skin damage, significantly improving the accuracy and timeliness of clinical monitoring.

[0049] 5. According to the out-of-limit judgment in the adaptive threshold evaluation record, trigger an alarm in real time and generate an alarm record, which can ensure that when the skin damage risk exceeds the dynamic threshold, the nursing staff or the patient can be reminded immediately for immediate intervention. The authenticity of the alarm can be evaluated through experimental verification means, which can not only handle real skin damage in a timely manner but also detect and mark false alarms.

[0050] 6. It can be more accurate with more use in various clinical scenarios and patient cases; closely combine alarm triggering, clinical intervention, and model feedback to construct a closed-loop process of diagnosis-disposal-relearning, which not only improves the processing efficiency of acute events such as skin damage but also enables the subsequent model to continuously iterate and improve the warning strategy, significantly optimizing the overall performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic flowchart of the data processing method for colorectal stoma care of the present invention;

[0052] Figure 2 It is a schematic structural diagram of the data processing system for colorectal stoma care of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0053] 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 ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] See also Figure 1 The present invention provides a method for processing data for colorectal cancer stoma nursing, comprising:

[0055] Step 1: By time-aligning the sensor signals at the stoma site and combining them with images and subjective feedback, we eliminate obviously abnormal data and aggregate the corresponding timestamps and sensor identifiers to generate initial stoma data, completing the acquisition and preliminary integration of multi-source sensor data.

[0056] The step 1 includes the following:

[0057] Step 101: Acquire real-time signals from the sensor array at the stoma location. The sensor array includes a blood sensor, a pressure sensor, and a humidity sensor. Simultaneously, it collects local stoma images output by an imaging device and patient subjective feedback data. Global timestamps are added to all signals and feedback. k , ensuring that data in the same time slice can be aligned and analyzed in subsequent links; the collected original information is unified into the original data record, including basic elements such as timestamp, sensor identification, sensor value, patient identification, etc.; by giving a unified timestamp to different sensors and auxiliary inputs, accurate alignment of stoma site sensors, patient subjective feedback and image data is achieved; in subsequent analysis, the temporal and spatial consistency between data can be guaranteed to avoid misjudgment caused by information mismatch.

[0058] Step 102: At the same time stamp t k The values obtained by multiple sensors are compared, and the high-order Minkowski distance (p>2) is used to measure the abnormality of a sensor data.

[0059] Suppose there are n valid sensors, then for the i-th sensor at timestamp t k The value S i (t k ) is calculated as follows; let S i (t k ) represents the i-th sensor at timestamp t k The reading at the time, take the comparison data S of n moments or sensor combinations j (t j ), where j∈[1, n], the formula defines the data abnormality M i(t k ):

[0060]

[0061] Where: S i (t k ) is the value of the i-th sensor at timestamp t k The original collected value at the time; Φ(·) is a function that performs nonlinear transformation on the sensor readings, which is used to preprocess data with a large value range or uneven distribution. For example, you can set:

[0062] Φ(x)=ln(1+α·x)

[0063] Where α is a scaling factor, which can reduce its gradient when the value is large and enhance the resolution when the value is small;

[0064] β j is the sensor trust factor or historical stability weight. A larger β value can be assigned to sensors that have performed reliably in the past. j For sensors with high failure rates or those that have experienced offset, β can be appropriately reduced. j ;

[0065] Δ(t k , t j ) represents the current timestamp t k Compare with timestamp t j The time difference between the two can be taken as an absolute value or other measurement method to measure the time distance between historical data and current data, ω(Δ(t k , t j )) is a time decay weighting function, which is used to place lower weight on data that is farther away from the current moment, thereby increasing sensitivity to recent trends. It often uses an exponential decay form:

[0066] ω(Δ(t k , t j ))=exp(-γ·Δ(t k , t j ))

[0067] Where γ>0 controls the decay rate; p is a high-order exponent, p>2, which can improve the amplification of abnormal values (such as spike noise or fault readings) and enhance the sensitivity to extreme situations. i (t k ) is the value of the i-th sensor at timestamp t k The abnormality metric under, the larger the value, the greater the deviation from the acceptable normal range;

[0068] When compared with the given threshold Θ, if the data abnormality M i (t k) > Θ indicates that the sensor may have significant anomalies at this moment and needs to be excluded or marked. In ostomy care data, not only the reasonable range of the current sensor readings should be concerned, but also the fluctuation trend of recent or historical readings should be considered. This formula adjusts the dynamic contribution of history through ω(Δ(t k , t j ))). The high-order Minkowski metric with p > 2 can more sensitively capture extreme values or significant offsets in multi-dimensional data distributions. The Φ(·) function can be individually adjusted according to the ranges and characteristics of different types of sensors, especially suitable for multi-source data with large differences in original magnitudes.

[0069] When M i (t k ) exceeds a certain threshold Θ, it is determined that the sensor data is abnormal at the current timestamp and needs to be excluded or marked. For the data marked as abnormal, it can be double-checked in subsequent processes (such as manual intervention or in-depth inspection), but it will not enter the next set of trusted data; the normal data screened by the above method, together with its corresponding timestamp and sensor identifier, are summarized to form a filtered data set;

[0070] In use, high-order Minkowski or advanced distance metric methods are adopted to identify and exclude outliers for multi-sensor values that occur at the same moment; reduce the interference caused by sensor failures or instantaneous noise, and provide a relatively clean preliminary data set for subsequent steps.

[0071] Step 103: According to the filtered data set screened in Step 102, at the same timestamp t k , organize and merge the sensor readings, patient subjective feedback, and image information; generate initial ostomy data, and this output retains necessary identification information, such as timestamp t k , sensor identifier, sensor availability mark, and the associated reference of the image or feedback;

[0072] In use, combine the content in Steps 101 to 103:

[0073] Unify the encapsulation and marking of valid sensor readings, image information, and patient feedback, format and store multi-source data, which can lay the foundation for subsequent refined calibration and risk analysis. Relying on the unified timestamp and pre-denoising of heterogeneous data, the original data can be initially cleaned and the basic structure can be ensured to be consistent, providing high-quality data input for accurate modeling and analysis in subsequent steps.

[0074] Step 2: Construct calibration factors from environmental parameters, correct the sensor array readings, compare them with the images and the patient's subjective feedback, screen out suspected failure or conflict signals, output the calibrated and verified reliable data in the multi-source calibration verification record, and explicitly save the attention factor, calibration coefficient, and conflict sensor list;

[0075] The content of Step 2 is as follows:

[0076] Step 201: Read the preliminary integrated data of the initial stoma data, which includes all sensor readings, image data, and patient subjective feedback data; select environmental parameters from the preliminary integrated data, such as humidity H k , temperature T k or pressure P k ;

[0077] Use these environmental parameters to construct calibration factors, such as:

[0078] Λ k = ξ·f(H k , T k , P k )

[0079] where ξ is a constant coefficient and f is a multivariate function based on stoma care experience and observational data;

[0080] Perform environmental calibration on the sensor readings S i (t k ) to obtain:

[0081]

[0082] where Ψ(·) is a non-linear mapping function, such as exponential transformation or logarithmic transformation, used to suppress large-scale interference and amplify subtle differences, and store the calibrated data in the stoma environment calibration data uniformly to provide input for the next cross-source verification;

[0083] When in use, considering environmental factors such as temperature, humidity, and pressure, and using a correction function to perform dynamic calibration on the sensor readings can further reduce the measurement deviation caused by environmental interference and make the sensor readings more in line with the true physiological conditions.

[0084] Step 202: Call the sensor readings in the stoma environment calibration data and compare them with the image data and the patient feedback data. For the i-th and j-th sensors at time stamp t k , define the cross-conflict amount Ω ij (t k ):

[0085]

[0086] where, v ij is the attention factor, and cos(·,·) can calculate the cosine similarity after taking the inner product of the feature vectors of each sensor to determine whether the readings of two sensors are highly consistent at the same moment:

[0087] When the cross-conflict volume Ω ij (t k ) continuously stays in the high range, it indicates that there is a significant conflict between the two, and further comparison and confirmation with the image or patient feedback data are required.

[0088] If the image shows no obvious leakage marks, but the reading of a certain sensor increases abnormally, it can be determined that the reading of this sensor fails;

[0089] If local pain or a sense of irritation is reported in the patient feedback data, but there is no corresponding change in the sensor, it also needs to be recorded as suspicious and subject to subsequent verification. The sensor readings determined to be credible through cross-validation are stored in the sensor-verified data, and the failed or conflicting readings are marked for subsequent analysis or manual review;

[0090] During use, by introducing image data and patient feedback and comparing them with the calibrated sensor values, it is possible to effectively filter out poor probe contact or occasional incorrect readings, and greatly improve the credibility of the data.

[0091] Step 203, after integrating the stoma environment calibration data and the sensor-verified data, generate the final credible readings, associated image identifiers, and possible anomaly marks for each timestamp t k and sensor i;

[0092] The integration results are uniformly output to the multi-source calibration verification record, including the calibrated values timestamp t k , sensor identifier i, verification results (marking whether there are anomalies and the corresponding references to the image and feedback information), and explicitly save the attention factor, calibration coefficient, and list of conflicting sensors in the multi-source calibration verification record;

[0093] During use, the results of the completed calibration and verification are uniformly output, and the failed or suspicious data are marked; it can provide a high-confidence, traceable set of sensor and image information for risk assessment. By making full use of environmental information and multi-source verification means, more credible sensor data are obtained, laying a solid foundation for personalized risk modeling and avoiding the interference of single-sensor failure or noise on the overall analysis.

[0094] Step 3: Construct individual baseline characteristics based on the multi-source calibration and verification records, quantify the likelihood of skin damage using the comprehensive risk formula, and dynamically adjust the alarm threshold in combination with the first-order iteration method to generate an adaptive threshold evaluation record, including: the change history of the dynamic threshold at each moment, the comparison result between the risk value and the threshold, and whether the skin damage risk marker has been triggered;

[0095] The said Step 2 includes the following contents:

[0096] Step 301: Read the verified sensor readings and corresponding timestamps from the multi-source calibration and verification records, and combine the patient's previous stoma care records (such as the frequency of postoperative skin damage, skin sensitivity, and types of complications that have occurred) to construct a personalized set of basic health parameters for distinguishing different constitutions and postoperative recovery stages; summarize and statistically analyze the current environmental information (such as room temperature, humidity, pressure, etc.) to form an environmental profile record;

[0097] After obtaining the basic health parameters and environmental profile, use a multi-dimensional mapping function to integrate them into the patient's individual baseline feature vector B p , which can be used to correct the sensitivity of different sensor readings; B p Store it in the baseline feature information in the form of a data structure;

[0098] When used in combination with information such as the patient's pre-operative and post-operative records and current physiological status, construct baseline feature information to reflect individual differences; this allows subsequent risk calculations to be personalized according to the sensitivity and complication history of different patients, avoiding a one-size-fits-all threshold setting.

[0099] Step 302: In the multi-source calibration and verification records, obtain the calibration value of sensor number i at timestamp t k under Call B in the baseline feature information p to reflect the patient's individual differences and environmental adaptability;

[0100] To more flexibly capture potential skin damage risks, first apply a non-linear mapping Φ(·) to , then multiply and accumulate it with the corresponding weight ω i , and the comprehensive risk quantity R(t k ) is defined as follows:

[0101]

[0102] In the formula: represents the calibrated and verified valid value of the i-th sensor at timestamp t k ; δ ij is the cooperation factor between the i-th and j-th sensors. When δ ijWhen it is relatively large, it means that there is a high degree of complementarity or enhancement between sensor i and sensor j in the judgment of skin damage risk; if δ ij is relatively small or negative, it indicates that there may be conflicts or overlaps between the two, and the combined effect needs to be weakened:

[0103] Θ p (B p ) is a baseline correction function used to make personalized adjustments according to the patient's individual feature vector B p , for example, giving additional amplification to patients with extremely sensitive skin, or slightly tightening for patients with a low-risk history; is a type of binary non-linear mapping that can have multiple specific implementations in different scenarios:

[0104] Ψ(x, y) = Φ(x) + Φ(y) (simple sum), or Ψ(x, y) = [Φ(x) · Φ(y)] α (product form);

[0105] where Φ(·) is a function for further transforming the sensor readings (such as logarithmic, exponential or custom functions), α is used to control the amplification or suppression degree of the cumulative term, m is the number of sensors actually enabled in the current care plan, and in the double summation of this formula, j starts from i to avoid symmetric repetition (if all i and j need to be traversed, the summation upper and lower limits can also be changed to j = 1 to m and corresponding corrections are made).

[0106] Through the synergy factor δ ij and the binary mapping , this formula can explicitly represent the interaction between sensors, no longer limited to simply accumulating the values of each sensor channel, and can adjust δ ij and Ψ according to clinical experience or data-driven methods to adapt to complex care scenarios (such as when there is non-linear coupling among multiple sensors), and with the help of Θ p (B p ) to achieve differential processing of the patient's historical records, postoperative recovery status, etc., and enhance the adaptability of the model.

[0107] Temporarily store the risk value R(t k ) at each time stamp t k in the form of a data structure into the stoma comprehensive risk score data, and retain the necessary marks (such as the corresponding sensor group, baseline correction coefficient, etc.) for dynamically optimizing the alarm threshold in step 303; when in use, perform non-linear weighting and coupling calculations on multiple sensor readings to obtain the real-time risk quantity R(t k ), and can improve the sensitivity and accuracy of anomaly detection by combining methods such as the synergy factor and non-linear mapping to achieve multi-dimensional risk assessment.

[0108] Step 303, after obtaining the comprehensive risk quantity R(tk ), the alarm threshold is dynamically tuned to cope with environmental changes and patient differences. Compared with the previous simple exponential amplification or reduction, a threshold target mapping function χ(·) is introduced as the target value, and the idea of first-order iterative update is adopted to make the threshold gradually approach the appropriate level. It is defined as follows:

[0109] Γ(t k+1 )=Γ(t k )+v[χ(R(t k ))-Γ(t k )]

[0110] Where: Γ(t k ) represents the current timestamp t k The alarm threshold under k )) is a type of threshold target mapping function, according to the current comprehensive risk amount R(t k ) gives the threshold target under ideal conditions, for example, it can be defined as:

[0111] χ(r)=Γ base +λg(r),

[0112] where Γ base is the basic threshold, λ is the sensitivity coefficient, and g(·) can be any function that can map the risk amount to the corresponding interval, such as piecewise linear, Sigmoid, or custom nonlinear;

[0113] v is the learning rate or update rate, which controls the speed at which the threshold moves toward the target value in this iteration. A larger v causes the threshold to change rapidly, while a smaller v causes it to change more steadily.

[0114] t k+1 is the timestamp of the next moment (or the next sampling point), and whenever the new risk amount R(t k ) is calculated, Γ(t k+1 ), ensuring that the threshold dynamically adapts to the current risk level.

[0115] Through the first-order difference form (Γ(t k+1 )-Γ(t k )) can achieve smooth and gradual progress, and can gradually tighten the threshold for sudden risks, and can also avoid severe threshold jitter when the risk tends to be stable, χ(R(t k )) can be flexibly set according to nursing experience or model training results, providing higher controllability in different patients or seasonal environments. This iterative formula can not only quickly integrate new risk information, but also combine the nursing feedback in step four to further correct the evolution direction of the interval value, forming a continuous self-learning process.

[0116] For any t k , if R(tk ) > Γ(t k ) If it is determined that there is a significant risk of skin damage, mark it for pending alarm,

[0117] Store the result of each comparison and the judgment flag in the adaptive threshold record, which clearly records information such as the timestamp, the risk value threshold, and whether it exceeds the limit; after completing the above threshold update and over-limit determination, form the final product of this step, the adaptive threshold evaluation record, which contains the following core elements: the dynamic threshold Γ(t k ) The change history at each moment, the risk value R(t k ) The comparison result with the threshold, and whether the skin damage risk mark has been triggered;

[0118] When in use, combine the content in steps 301 to 304: introduce the adaptive threshold algorithm, iteratively update the alarm threshold Γ(t k ) according to the recent risk level and generate the adaptive threshold record; the alarm threshold can be flexibly adjusted under different nursing scenarios and patient states to reduce false alarms and missed alarms, and enhance intelligence and self-adaptability; based on individual differences and real-time environmental parameters, dynamic risk modeling is realized, which can adaptively give warnings for the possibility of stoma skin damage, greatly improving the accuracy and timeliness of clinical monitoring.

[0119] Step Four: When the comprehensive risk quantity R(t k ) exceeds the corresponding dynamic threshold Γ(t k ), trigger an alarm and record it in the alarm nursing closed-loop record. The nursing staff will intervene and conduct local verification based on this. If it is confirmed as a real skin damage or a false alarm, then transmit the correction factor back to dynamically optimize the model and the threshold strategy;

[0120] The said Step Four includes the following content;

[0121] Step 401: Read the current timestamp t k from the adaptive threshold evaluation record, the comprehensive risk quantity R(t k ) and the corresponding dynamic threshold Γ(t k ), and at the same time obtain the over-limit flag, indicating whether R(t k ) > Γ(t k ) has been satisfied;

[0122] For all cases in the records that satisfy R(t k ) > Γ(t k ), enter the pending alarm state. According to the actual usage environment, a predefined multi-channel notification mechanism (such as wearable device warning, mobile terminal push, or nurse workstation pop-up window) can be called to send an alarm signal; write the current alarm and its trigger timestamp t into the alarm record data object, including the corresponding sensor set number, risk value, threshold, and whether the alarm is automatically or manually confirmed;

[0123] During use, based on the over-limit judgment in the record evaluated by the adaptive threshold, alarms are triggered in real time and alarm records are generated, which can ensure that when the skin damage risk exceeds the dynamic threshold, the nursing staff or the patient can be reminded immediately to achieve immediate intervention.

[0124] Step 402: When the nursing staff receives the alarm instruction triggered in Step 401, they need to check and intervene in the patient's stoma site, such as changing the dressing, cleaning the periphery of the stoma, or adjusting the activity posture, etc.; collect new clinical observation results (such as image data), the patient's subjective description, and nursing operation items (such as changing the chassis or using hemostatic gel, etc.);

[0125] After the nursing operation is completed, local verification is performed: such as verifying whether there is actual abnormal skin damage through a rapid test strip or a simple sensor, or collecting a new sensor reading and image to determine whether the system is still in a high-risk state.

[0126] If it is found that the alarm is a false alarm, mark the false alarm in the local verification record and explain the reason (for example: temporary noise interference, instantaneous signal fluctuation due to patient body position change, etc.); if it is a real skin damage, carry out targeted nursing treatment and make a detailed record.

[0127] Integrate the key information of the nursing operation and local verification into local inspection data, which includes: the corresponding alarm timestamp t k , the content and result of the nursing operation; the new rapid test or sensor test data; whether it is a false alarm and the reason explanation;

[0128] During use, after receiving the alarm, the nursing staff can evaluate the authenticity of the alarm through existing verification means by performing actual intervention and collecting new inspection data (such as rapid test strips, additional images), which can not only timely handle real skin damage but also discover and mark false alarms.

[0129] Step 403: Define the nursing feedback correction factor γ k , where k is the time index or the identifier of the alarm batch, used to correct the previous comprehensive risk amount or sensor cooperation factor, and the formula is as follows:

[0130] γ k =κ×[λ true 1(true alarm) - λ false 1(false alarm)]

[0131] Where: 1(·) is an indicator function, which takes 1 when the local verification confirms that it is a real skin damage alarm and takes 0 if it is a false alarm;

[0132] λ true , λ falseare the penalty / reward coefficients for true alarms and false alarms respectively, and κ is the overall adjustment rate, which is used to control the amplitude of subsequent risk calculation or cofactor adjustment; this correction factor can be used to dynamically lower (or raise) the weight of the sensor that causes false alarms (or missed alarms) or correct the evolution speed of the adaptive threshold during the next risk assessment;

[0133] Compare the local inspection data with the above γ k Store it in the feedback record structure;

[0134] In the next iteration of step three to calculate the comprehensive risk amount R(t k+1 ) and the threshold Γ(t k+1 ), the records in the feedback record can be referred to for self-learning incremental updates of relevant parameters (such as the cofactor δ ij , the individual correction function Θ p (B p ) and the threshold target mapping χ(·));

[0135] Integrate all alarm trigger information, nursing intervention items, local verification data, and feedback correction factors to form the final product of this step. The alarm nursing closed-loop record should at least include:

[0136] The timestamp of the alarm that has occurred, the sensor combination, the risk value and the threshold, the nursing operation and the local verification result, the mark of whether it is a true alarm or a false alarm; the correction factor (γ k ) for the model and local suggestions.

[0137] When in use, the nursing intervention and the local verification result are transmitted back to the risk assessment model (for example, by adjusting the sensor weight or the threshold evolution strategy through the correction factor), and finally generating the alarm nursing closed-loop record can achieve closed-loop self-learning optimization, making the system more and more accurate in various clinical scenarios and patient cases; closely combining alarm triggering, clinical intervention, and model feedback, a closed-loop process of diagnosis-disposal-relearning is constructed, which not only improves the processing efficiency of acute events such as skin damage, but also enables the subsequent model to continuously iterate and improve the warning strategy, significantly optimizing the overall performance of the system.

[0138] Based on multi-source sensor data acquisition and preliminary integration, by introducing an environmental calibration and cross-source cross-validation mechanism, outliers caused by temperature and humidity interference or poor device contact are effectively eliminated, significantly improving the accuracy and reliability of stoma monitoring data. Subsequently, combined with the patient's previous care records and individual baseline characteristics, through comprehensive risk quantification and adaptive threshold modeling, flexible response to different postoperative stages and environmental changes is achieved, which can not only reduce the interference of false alarms but also improve the sensitivity to capture sudden situations such as potential bleeding. In addition, this application adopts a closed-loop feedback strategy. After an alarm is triggered, nursing interventions and local verification results are collected, and a correction factor is generated and dynamically transmitted back to the risk assessment model, enabling the system to continuously self-learn and optimize in real applications.

[0139] Please refer to Figure 2 , the present invention provides a data processing device for colorectal stoma care, including,

[0140] A multi-source data integration module, which aligns sensor signals at the stoma site in time, combines images and subjective feedback, eliminates obvious abnormal data, and summarizes it together with its corresponding timestamp and sensor identifier to generate initial stoma data, completing the acquisition and preliminary integration of multi-source sensor data;

[0141] A multi-source cross-validation module, which constructs a calibration factor from environmental parameters to correct the readings of the sensor array, compares them with images and the patient's subjective feedback, screens out suspected failed or conflicting signals, outputs calibrated and verified reliable data in the multi-source calibration verification record, and explicitly saves the attention factor, calibration coefficient, and conflict sensor list;

[0142] A risk assessment module, which constructs individual baseline characteristics based on the multi-source calibration verification record, quantifies the possibility of skin damage using a comprehensive risk formula, and dynamically adjusts the alarm threshold in combination with a first-order iterative method to generate an adaptive threshold assessment record; including: the change history of the dynamic threshold at each moment, the comparison result of the risk value and the threshold, and whether the skin damage risk marker has been triggered;

[0143] An alarm trigger feedback module, when the comprehensive risk amount exceeds the corresponding dynamic threshold, triggers an alarm and records it in the alarm nursing closed-loop record. The nursing staff conducts interventions and local verification based on this. If it is confirmed as a real skin damage or a false alarm, a correction factor is transmitted back to dynamically optimize the model and the threshold strategy.

[0144] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0145] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0146] In 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 illustrative. For example, the division of the units is only for some logical function divisions, and there can be other division methods in actual implementation. For example, 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 displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0147] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0148] As described above, only the specific implementation manners of this application are provided, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application and should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A data processing method for colorectal stoma care, characterized in that: including, By time-aligning the sensor signals at the stoma site, combining images with subjective feedback, eliminating significantly abnormal data, and summarizing the corresponding timestamps and sensor identifiers, initial stoma data is generated to complete the acquisition and preliminary integration of multi-source sensor data; After constructing calibration factors from environmental parameters, the readings of the sensor array are corrected, compared with images and the patient's subjective feedback, suspected failed or conflicting signals are screened out, and the calibrated and verified reliable data is output in the multi-source calibration verification record, and the attention factor, calibration coefficient, and conflicting sensor list are explicitly saved; Construct individual baseline characteristics based on the multi-source calibration verification record, quantify the possibility of skin damage around the stoma using the comprehensive risk formula, and dynamically adjust the alarm threshold in combination with the first-order iteration method to generate an adaptive threshold evaluation record; including: the change history of the dynamic threshold at each moment, the comparison result of the risk value and the threshold, and whether the skin damage risk marker has been triggered; When the comprehensive risk quantity R(t k ) exceeds the corresponding dynamic threshold Γ(t k ), an alarm is triggered and recorded in the alarm care closed-loop record, and intervention and local verification are carried out accordingly. If it is confirmed as a real peristomal skin injury or a false alarm, a correction factor is transmitted back to dynamically optimize the model and the threshold strategy.

2. The method for processing colorectal stoma care data according to claim 1, characterized in that: Obtain real-time signals from the sensor array at the stoma position, and simultaneously collect the local stoma image output by the imaging device and the patient's subjective feedback data; uniformly add a global timestamp to all signals and feedback; classify the collected original information into the original data record, including the timestamp, sensor identifier, sensor value, and patient identifier; At the same timestamp t k compare the values obtained by multiple sensors, where Given n effective sensors, the value S of the i-th sensor at time stamp t k is calculated as follows for i (t k ): Let S i (t k ) denote the reading of the i-th sensor at timestamp t k . Take the control data S j (t j ) of n moments or sensor combinations, where j ∈ [1, n], and define the data abnormality degree M i (t k ) according to the formula; when the data abnormality degree M i (t k ) exceeds the corresponding threshold Θ, mark the data as abnormal and conduct a secondary verification.

3. The method for processing colorectal stoma care data according to claim 2, characterized in that: Read the preliminary integrated data of the initial stoma data, which includes all sensor readings, image data, and patient subjective feedback data; select environmental parameters from the preliminary integrated data to construct a calibration factor: Λ k = ξ·f(H k , T k , P k ), where ξ is a constant coefficient and f is a multivariate function based on stoma care experience and observational data; perform environmental calibration on the sensor reading S i (t k ) to obtain: where Ψ(·) is a non-linear mapping function.

4. The method for processing colorectal stoma care data according to claim 3, characterized in that: Call the sensor readings in the stoma environment calibration data And compare the image data with the patient feedback data. For the time stamp t k For the i-th and j-th sensors at time t, define the cross-conflict quantity Ω ij (t k ): where v ij is the attention factor, cos(·, ·) can calculate the cosine similarity after taking the inner product of the feature vectors of each sensor. When the cross-conflict quantity Ω ij (t k ) continuously stays in the high range, if the image shows no obvious leakage trace while the sensor reading abnormally increases, it is determined that the sensor reading fails.

5. The method for processing colorectal stoma care data according to claim 4, characterized in that: Store the sensor readings determined to be trustworthy through cross-validation into the post-sensor-verification data, and mark the invalid or conflicting readings; after integrating the stoma environment calibration data with the post-sensor-verification data, generate the final trustworthy readings, associated image identifiers, and possible anomaly marks for each timestamp t k and sensor i; uniformly output the integration result to the multi-source calibration verification record, including the calibrated value timestamp t k 、sensor identifier i, verification result; Read the verified sensor readings and corresponding timestamps from the multi-source calibration verification record, combine with the patient's previous stoma care records, construct a personalized set of basic health parameters, summarize and statistically analyze the current environmental information to form an environmental profile record; Integrate it into the patient individual baseline feature vector B using a multi-dimensional mapping function p , B p Store it in the baseline feature information in the form of a data structure.

6. The method for processing colorectal stoma care data according to claim 5, characterized in that: Obtain the calibration value of sensor number i at timestamp t in the multi-source calibration verification record k under Call B in the baseline feature information p , and apply the non-linear mapping Φ(·) to , multiply and accumulate with the corresponding weight ω i , and the comprehensive risk quantity R(t k ) is defined as follows: wherein: represents the calibrated and validated effective value of the i-th sensor at timestamp t k ; δ ij is the cooperation factor between the i-th and j-th sensors, and Θ p (B p ) is the baseline correction function, is a type of binary non-linear mapping; For each timestamp t k the risk value R(t k ) is temporarily stored in the stoma comprehensive risk score data in the form of a data structure, and necessary markers are retained.

7. The method for processing colorectal stoma care data according to claim 6, characterized in that: After obtaining the comprehensive risk quantity R(t k ), the alarm threshold is dynamically optimized, and the threshold target mapping function χ(·) is introduced as the target value, which is defined as follows: Γ(t k+1 ) = Γ(t k ) + v[χ(R(t k )) - Γ(t k )] Where: Γ(t k ) represents the alarm threshold at the current timestamp t k , and χ(R(t k )) is a type of threshold target mapping function: χ(r) = Γ base + λg(r) where Γ base is the base threshold, λ is the sensitivity coefficient, and g(·) can be any function that maps the risk amount to the corresponding interval; v is the learning rate or update rate; t k+1 is the timestamp of the next moment; For any t k , if R(t k ) > Γ(t k ), it is determined that there is a significant risk of skin damage, marked for pending alarm, and each comparison result and judgment identifier are stored in the adaptive threshold record.

8. The method for processing colorectal stoma care data according to claim 7, characterized in that: Read the current timestamp t k The comprehensive risk quantity R(t k ) and the corresponding dynamic threshold Γ(t k ), and at the same time obtain the over-limit flag to indicate whether R(t k ) > Γ(t k ) is satisfied; For all cases in the records where R(t k ) > Γ(t k ), enter the state to be alarmed. According to the actual usage environment, a predefined multi-channel notification mechanism can be called to send an alarm signal; Write the current alarm and its triggering timestamp t into the alarm record data object, including the corresponding sensor set number, risk value, threshold, and whether the alarm is automatically or manually confirmed.

9. The method for processing colorectal stoma care data according to claim 8, characterized in that: Examine and intervene in the patient's stoma site, collect new clinical observation results, the patient's subjective description, and nursing operation items; If it is found that the alarm is a false alarm, mark the false alarm in the local verification record and explain the reason; If it is a real skin injury, carry out targeted care; integrate the nursing operation and the key points of local verification into the local inspection data.

10. A data processing device for colorectal stoma care, characterized in that: including, The multi-source data integration module aligns the sensor signals at the stoma site in terms of time, combines images and subjective feedback, eliminates significantly abnormal data, and summarizes it together with its corresponding timestamps and sensor identifiers to generate initial stoma data, completing the acquisition and preliminary integration of multi-source sensor data; The multi-source cross-validation module constructs calibration factors from environmental parameters to correct the readings of the sensor array, compares them with images and the patient's subjective feedback, screens out suspected failure or conflict signals, outputs the calibrated and verified reliable data in the multi-source calibration verification record, and explicitly saves the attention factor, calibration coefficient, and conflict sensor list; The risk assessment module constructs individual baseline characteristics based on the multi-source calibration verification record, quantifies the likelihood of skin damage using a comprehensive risk formula, dynamically adjusts the alarm threshold in combination with the first-order iterative method, and generates an adaptive threshold assessment record; including: the change history of the dynamic threshold at each moment, the comparison result between the risk value and the threshold, and whether the skin damage risk marker has been triggered; The alarm trigger feedback module triggers an alarm and records it in the alarm care closed-loop record when the comprehensive risk quantity exceeds the corresponding dynamic threshold. The nursing staff conducts intervention and local verification based on this. If it is confirmed as a real skin damage or a false alarm, a correction factor is sent back to dynamically optimize the model and the threshold strategy.

Images