Intelligent pressure sore risk management and nursing decision support system and method thereof

Abstract

The invention discloses an intelligent pressure sore risk management and nursing decision support system and method. A patient risk file is established by collecting patient inspection index data, skin image data and Braden scale evaluation data, a normal index area is identified to form an index reference domain, and multi-part associated monitoring configuration is established; performing feature extraction on the inspection index data to form an index change curve, and performing high-risk combination mode recognition in combination with Braden scale assessment data to generate a grading risk assessment benchmark; executing time axis dynamic tracking on the skin image data to generate a skin state change sequence, and generating a burnishing positioning result through burnishing area detection; and matching the red pressing positioning result with the skin characteristic parameters to identify a risk area, executing risk level dynamic adjustment according to the risk area in combination with a grading risk assessment criterion to generate an injury degree value, matching personalized nursing measures to generate a nursing decision result, and providing intelligent decision support for clinical nursing.

Description

Technical Field

[0001] This invention relates to the field of medical and nursing information processing technology, and in particular to an intelligent pressure ulcer risk management and nursing decision support system and method. Background Technology

[0002] Pressure ulcers are a common complication in long-term bedridden patients, with the elderly, critically ill, and those who are also long-term bedridden being high-risk groups. Once pressure ulcers develop, they not only increase the difficulty of care and hinder wound healing, but can also lead to infection, worsening of the condition, and even exacerbation of the illness, resulting in high subsequent treatment costs and placing a heavy burden on patients and their families. In clinical practice, nursing staff need to comprehensively consider multiple dimensions of information, including patients' laboratory test results, changes in skin appearance, and scores on clinical assessment scales. This requires a high level of continuous attention from healthcare professionals and places immense pressure on them. However, current pressure ulcer risk assessment methods mainly rely on the subjective experience and manual observation of nursing staff, resulting in inconsistent assessment standards, fragmented data recording, and delayed risk warnings, making it difficult to achieve early identification and continuous tracking of high-risk patients.

[0003] Furthermore, the current clinical practice commonly relies on nurses taking photos for handover between shifts. This method lacks systematicity and continuity, making it impossible to conduct time-series comparative analysis of skin condition changes, and even more difficult to predict whether a patient's condition will worsen. In particular, newly hired nurses often lack the ability to classify pressure ulcers and predict risks, while family members of patients who have been caring for them at home for a long time often lack relevant knowledge and do not pay enough attention. Clinically, it is common for patients who have not paid attention to their skin condition at home for a long time to have pressure ulcers that have penetrated to the bone upon admission. Summary of the Invention

[0004] This invention discloses an intelligent pressure ulcer risk management and nursing decision support system and method, which aims to integrate multi-source data such as patient test results, skin image data, and Braden scale assessment data to establish patient risk profiles and multi-site correlation monitoring configurations. By extracting indicator features and identifying high-risk combination patterns, a graded risk assessment benchmark is generated. Combined with time-axis dynamic tracking of skin image data and detection of pressure redness areas, dynamic monitoring of skin condition is achieved. Finally, based on risk area identification and responsiveness assessment of nursing measures, personalized nursing decisions are generated, completing closed-loop management from risk prediction to nursing intervention.

[0005] The first aspect of this invention proposes an intelligent pressure ulcer risk management and nursing decision support method, comprising the following steps: Collect test data, skin image data, and Braden scale assessment data of the patient's pressure points, and establish a patient risk profile based on the patient identifier associated with the test data and the skin image data; Based on the patient risk profile, normal indicator areas are identified to form an indicator benchmark domain, and a multi-site correlation monitoring configuration is established based on the indicator benchmark domain; The test indicator data are subjected to indicator feature extraction to form an indicator change curve. The indicator change curve is then compared with the Braden scale assessment data to generate a graded risk assessment benchmark by high-risk combination pattern recognition. Image pattern recognition is performed on the skin image data to extract skin feature parameters. Based on the skin image data, dynamic time-axis tracking is performed to generate a skin state change sequence. Based on the multi-site associated monitoring configuration, redness region detection is performed on the skin state change sequence to generate redness localization results. The redness localization result is matched with the skin feature parameters to identify risk areas. Based on the risk areas and the graded risk assessment benchmark, the risk level is dynamically adjusted to generate a damage degree value. Based on the damage degree value, personalized nursing measures are matched to generate a nursing decision result.

[0006] A second aspect of this invention proposes an intelligent pressure ulcer risk management and nursing decision support system, comprising: The data acquisition module is used to collect test index data, skin image data and Braden scale assessment data of the patient's pressure site, and establish a patient risk profile based on the patient identifier associated with the test index data and the skin image data; The baseline establishment module is used to identify normal indicator areas based on the patient risk file to form an indicator baseline domain, and to establish a multi-site correlation monitoring configuration based on the indicator baseline domain. The risk assessment module is used to extract indicator features from the test indicator data to form indicator change curves, and to perform high-risk combination pattern recognition by combining the indicator change curves with the Braden scale assessment data to generate a graded risk assessment benchmark. The image analysis module is used to perform image pattern recognition to extract skin feature parameters from the skin image data, perform time-axis dynamic tracking to generate a skin state change sequence based on the skin image data, and perform redness region detection on the skin state change sequence based on the multi-site correlation monitoring configuration to generate redness localization results. The decision output module is used to perform damage feature matching between the redness positioning result and the skin feature parameters to identify risk areas, dynamically adjust the risk level based on the risk areas and the graded risk assessment benchmark to generate damage severity values, and match personalized nursing measures based on the damage severity values ​​to generate nursing decision results.

[0007] The beneficial effects of this invention are reflected in the following aspects: First, by integrating three data sources—patient laboratory test data, skin image data, and Braden scale assessment data—a patient risk profile is established. Data cross-validation ensures data reliability, and a multi-site correlation monitoring configuration and hierarchical early warning mechanism for areas such as the sacrum, heel, and elbow are established based on the indicator baseline domain. This overcomes the limitations of traditional methods, which involve scattered data and isolated monitoring of each site, achieving systematic risk monitoring across multiple sites. Second, it innovatively integrates the indicator change curves of laboratory test data with Braden scale assessment data for analysis. Through deviation calculation and synergistic deterioration effect identification, high-risk combination patterns of simultaneous deterioration of multiple indicators are discovered. This captures complex risk states that are difficult to detect through single indicator analysis, enabling predictive assessment of pressure ulcer risk. Finally, dynamic time-axis tracking technology based on skin image data enables visual comparison of skin condition changes. Combined with nursing intervention responsiveness assessment, a two-way adjustment mechanism of positive and negative feedback is established. The intensity of the nursing plan is dynamically matched according to the degree of injury, and the implementation effect is continuously tracked, forming a closed-loop management of monitoring-assessment-intervention-feedback, realizing a shift from a passive treatment to a proactive prevention nursing model. Attached Figure Description

[0008] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0009] Figure 1 This is a flowchart illustrating the intelligent pressure ulcer risk management and nursing decision support method of the present invention.

[0010] Figure 2 This is a schematic diagram of the indicator change curve and trend analysis of the present invention.

[0011] Figure 3 This is a schematic diagram of the temporal registration and change detection of skin images according to the present invention.

[0012] Figure 4 This is a structural block diagram of the intelligent pressure ulcer risk management and nursing decision support system of the present invention.

[0013] Wherein: 1-Serve albumin raw data points; 2-Serve albumin fitted curve; 3-Hemoglobin raw data points; 4-Hemoglobin fitted curve; 5-Descending trend marker; 6-Stable trend marker; 7-Ascending trend marker; 8-Normal range reference line; 9-Warning threshold line; 10-Trend regression line; 11-Original image at time T1; 12-Original image at time T2; 13-Original image at time T3; 14-Registered T1 image; 15-Registered T2 image; 16-Registered T3 image; 17-Differential image between T1 and T2; 18-Differential image between T2 and T3; 19-No change area; 20-Deterioration area; 21-Improvement area; 22-Feature point matching line. Detailed Implementation

[0014] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0015] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0016] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0017] The technical solutions of the embodiments of this application will be described below.

[0018] like Figure 1 As shown, this embodiment of the invention provides an intelligent pressure ulcer risk management and nursing decision support method, including the following steps S110-S150: Step S110: Collect test index data, skin image data and Braden scale assessment data of the pressure site of the patient, and establish a patient risk profile based on the patient identifier associated with the test index data and skin image data.

[0019] Specifically, laboratory test data, skin images, and Braden scale assessment data are collected from the patient's pressure points. The system automatically retrieves patient test data by connecting to the hospital's Laboratory Information System (LIS). When connection is not possible, nurses can manually input data or photograph test reports for OCR data extraction. The test data includes serum albumin, hemoglobin, serum sodium, and serum chloride levels. Serum albumin and hemoglobin levels are used for nutritional status assessment, while serum sodium and chloride levels are used for dehydration risk warning. The normal range for serum albumin is 35-55 g / L. A serum albumin level below 30 g / L indicates malnutrition and decreased tissue repair capacity. The normal range for hemoglobin is 120-160 g / L for men and 110-150 g / L for women. Low hemoglobin levels indicate anemia and insufficient tissue oxygen supply. The normal range for serum sodium ions is 135-145 mmol / L, and the normal range for serum chloride ions is 96-106 mmol / L. Significantly elevated levels trigger a dehydration warning, indicating possible dehydration, poor skin elasticity, and fragile tissues. A portable skin imaging device equipped with a high-resolution image sensor was used to perform standardized imaging of pressure points such as the sacrum, heel, and elbow to acquire skin image data. Standardized imaging required the same angle and lighting conditions for the same area. The image sensor resolution was no less than 12 megapixels, and the shooting distance was controlled within 15-25 cm to ensure image clarity. Skin image data includes high-resolution images of pressure sites and capture timestamps. At least three images from different angles are captured for each pressure site to ensure complete coverage. The system generates a unique image file index for each image for subsequent data association. Nursing staff use a tablet to assess patients based on the Braden Pressure Ulcer Risk Assessment Scale, automatically calculating the total score and risk level to obtain Braden scale assessment data. The Braden scale assessment data includes scores for six dimensions: sensation, skin moisture, mobility, nutrient uptake, friction, and shear stress. Sensation, skin moisture, mobility, nutrient uptake, and nutrient uptake are scored on a scale of 1-4, while friction and shear stress are scored on a scale of 1-3. The total score ranges from 6 to 23 points. A total score below 12 indicates high risk, 12-14 indicates medium risk, and above 14 indicates low risk.

[0020] In some embodiments, establishing a patient risk profile based on the association of the test index data and the skin image data with a patient identifier includes: performing data quality verification on the test index data to generate verification-passed data; obtaining serum albumin and hemoglobin values ​​from the verification-passed data to form a nutritional status label; extracting pressure site location information from the skin image data to generate a site identifier; and associating the verification-passed data, the nutritional status label, and the site identifier with a patient identifier to establish a patient risk profile.

[0021] Data quality verification is performed on laboratory test data to generate verified data. Laboratory test data collected from the same patient at different time points are arranged in ascending order of collection time to create a time-series dataset. The variation range between two consecutive test results is compared. An abnormality warning is triggered when the single variation range of serum albumin exceeds 20%, when the single variation range of hemoglobin exceeds 15%, and when the single variation range of serum sodium or chloride ion exceeds 10%. These thresholds can be configured and adjusted according to actual clinical needs. Abnormal warning data requires manual review to confirm the accuracy of data collection and entry (e.g., if the patient's previous serum albumin value was 40 g / L and the current test result is 30 g / L, the variation reaches 25%, the system automatically marks it as abnormal and prompts the nurse to verify whether there is specimen confusion or acute change in condition). The test indicator data, after passing the time-series variation check, were further validated for numerical range. The effective range for serum albumin was set at 15-65 g / L, for hemoglobin at 40-200 g / L, for serum sodium at 110-170 mmol / L, and for serum chloride at 70-130 mmol / L. Values ​​outside the effective range were marked as abnormal and removed. The test indicator data, after passing the numerical range check, were then validated for logical consistency. This logical consistency check included verifying the match between the test items and the patient's age and gender. The lower limit of normal for hemoglobin was 120 g / L for male patients and 110 g / L for female patients. Data that passes the time-series change verification, numerical range verification, and logical consistency verification is marked as verified data. Verified data is also given a data quality score, which is based on a 100-point scale. Data with a score above 85 is marked as high confidence, data with a score between 60 and 85 is marked as medium confidence, and data with a score below 60 is marked as low confidence.

[0022] Nutritional status labels are generated from the validated data by extracting serum albumin and hemoglobin values. Serum albumin values ​​reflect the patient's protein nutritional status and liver synthesis function. A serum albumin value below 25 g / L is labeled as severe malnutrition, between 25 and 30 g / L as moderate malnutrition, between 30 and 35 g / L as mild nutritional risk, and above or equal to 35 g / L as normal. Hemoglobin values ​​reflect the patient's tissue oxygenation capacity and red blood cell oxygen-carrying capacity. A hemoglobin value below 60 g / L is labeled as severe anemia, between 60 and 90 g / L as moderate anemia, between 90 g / L and the lower limit of normal as mild anemia, and above the lower limit of normal as normal. Based on the combined serum albumin and hemoglobin values ​​in the validated data, a nutritional status label is generated. This label uses a three-level classification system, categorized into three levels: well-nourished, at nutritional risk, and malnourished. Well-nourished is indicated when both serum albumin and hemoglobin values ​​are within the normal range; at nutritional risk, either value is below the lower limit of normal; and malnourished is indicated when both values ​​are below the lower limit of normal. The nutritional status label also records the specific values ​​of both values ​​and the corresponding data quality score.

[0023] Based on skin image data, the location information of the pressure points is extracted to generate site identifiers. The skin image data undergoes image preprocessing, which includes three steps: brightness correction, contrast enhancement, and color normalization. Brightness correction uses a histogram equalization algorithm to adjust the image brightness to a standard range. Contrast enhancement uses an adaptive contrast-limited histogram equalization algorithm to improve the recognizability of skin texture. Color normalization uses white balance correction to eliminate color differences caused by different lighting conditions. Deep learning image recognition algorithms are used to analyze the preprocessed skin image data and identify anatomical landmarks in the image, including bony prominences such as the sacral protuberance, ischial tuberosity, greater trochanter of the femur, calcaneal tuberosity, and olecranon of the ulna. The specific location of the pressure point is determined based on the positional relationship of the anatomical landmarks, generating site identifiers. Site identifiers adopt a standardized hierarchical coding system, with the first level being body region coding: S for the sacrococcygeal region, H for the heel region, E for the elbow region, P for the scapular region, and O for the occipital region. The second level is lateral coding, with the left side coded as L, the right side as R, and the midline as M. The third level is sequence coding, used to distinguish multiple pressure points within the same region. The sacral and coccygeal pressure area identified in the skin image data generates a location identifier SM-01, the left heel pressure area generates a location identifier HL-01, and the right elbow pressure area generates a location identifier ER-01. These location identifiers are bound to their corresponding image file indexes, forming a location-image association record.

[0024] Patient risk profiles are established by associating validated data, nutritional status labels, and site identifiers with patient identification. The primary key constraint for data association is established using the patient's hospital number as the unique index key. The hospital number is generated according to the hospital's unified coding rules, with a format of four digits for the year followed by an eight-digit serial number, totaling twelve digits. Serum albumin and hemoglobin values ​​from the validated data are bound to the nutritional status label to form the patient's nutritional risk data. This nutritional risk data includes fields for original test values, nutritional status labels, data collection time, and data quality scores. Site identifiers are associated with the patient's hospital number to form the patient's pressure site information, which includes a site identifier code field and an image file index field. By integrating nutritional risk data and pressure site information, a complete patient risk profile is established. The patient risk profile is stored in a relational database. The main table stores basic patient information and hospital number index. The nutritional risk sub-table stores nutritional status tags and verification data through a foreign key associated with the hospital number. The pressure site sub-table stores site identifiers and image indexes through a foreign key associated with the hospital number. The skin monitoring record sub-table stores monitoring time and skin condition description through a foreign key associated with the hospital number. The nursing execution data sub-table stores nursing operation time, operation type, and operation content through a foreign key associated with the hospital number.

[0025] Step S120: Based on the patient's risk record, identify the normal indicator area to form an indicator benchmark domain, and establish a multi-site correlation monitoring configuration based on the indicator benchmark domain.

[0026] Based on patient risk records, a baseline domain for normal indicators was established by identifying normal indicator zones. Nutritional risk data, including historical sequences of serum albumin and hemoglobin values ​​and their data quality scores, was retrieved from patient risk records. Only high-confidence data with a data quality score higher than 85 were included in the statistical analysis. Distribution statistics were performed on the historical serum albumin values ​​selected from the patient risk records to identify intervals with concentrated numerical distributions as reference ranges for normal indicator zones. Combining clinical criteria, serum albumin values ​​greater than or equal to 35 g / L were defined as the normal indicator zone, values ​​between 30 and 35 g / L as the mild risk zone, values ​​between 25 and 30 g / L as the borderline indicator zone, and values ​​below 25 g / L as the abnormal indicator zone. Based on historical hemoglobin value data selected from patient risk records, distribution statistics were performed to identify intervals with concentrated value distributions as reference ranges for the normal indicator zone. Combining clinical standards and patient gender, the interval with hemoglobin values ​​greater than or equal to the lower limit of normal (120 g / L for males, 110 g / L for females) was defined as the normal indicator zone; the interval with hemoglobin values ​​between 90 g / L and the lower limit of normal was defined as the borderline indicator zone; and the interval with hemoglobin values ​​below 90 g / L was defined as the abnormal indicator zone. The boundary values ​​of the normal, borderline, and abnormal indicator zones for each test indicator were integrated to form the indicator benchmark domain. The indicator benchmark domain uses a multidimensional interval representation method, with each test indicator corresponding to one dimension, containing four boundary parameters: upper limit of the abnormal zone, upper limit of the borderline zone, lower limit of the normal zone, and upper limit of the normal zone. The baseline range of indicators can be dynamically adjusted according to individual patient differences. The lower limit of the normal range for hemoglobin in elderly patients can be appropriately lowered to 100 g / L, and the lower limit of the normal range for serum albumin in long-term bedridden patients can be appropriately lowered to 28 g / L. The above adjustment parameters can be configured according to clinical guidelines.

[0027] In some embodiments, establishing a multi-site associated monitoring configuration based on the indicator benchmark domain includes: determining the monitoring priority of each pressure site according to the indicator benchmark domain to generate a site priority table; defining association rules for the sacrum, coccyx, heel, and elbow based on the site priority table to generate an association rule table; configuring cross-site linkage early warning triggering conditions according to the association rule table to form a triggering condition set; and establishing a multi-site associated monitoring configuration based on the triggering condition set.

[0028] A site priority table is generated to determine the monitoring priority of each pressure site based on the baseline index range. The status range of serum albumin and hemoglobin values ​​within the baseline index range determines the patient's overall nutritional risk level. When both indicators are within the normal range, the basic monitoring priority configuration is used for each pressure site. When the baseline index range shows that any indicator is in the borderline range, the monitoring priority of the sacrococcygeal region is increased by one level, and the monitoring priority of the heel is increased by one level. When the baseline index range shows that any indicator is in the abnormal range, the monitoring priority of all pressure sites is increased to the highest level. Based on the correspondence between the status range of the baseline index range and the site risk, a site priority table is generated, using a three-level priority labeling system. Level 1 priority sites include the sacrococcygeal region (site label SM) and the ischial tuberosity area. These sites bear the greatest weight pressure and have the highest risk of pressure ulcers; the monitoring interval is set to once every 2 hours. Level 2 priority sites include the heel (site label HL / HR), the lateral malleolus, and the medial malleolus areas. These sites have prominent bony prominences and thin subcutaneous tissue; the monitoring interval is set to once every 4 hours. Level 3 priority areas include the elbow (elbow / er), scapula (pl / pr), occipital region (om), and auricle. These areas have a relatively low incidence of pressure ulcers, and the monitoring interval is set to once per shift. The area priority table records the area identification code, priority level, and monitoring interval configuration for each area, and supports dynamic updates based on changes in patient position. The priority of the greater trochanter of the femur is automatically increased when the patient is in the lateral decubitus position, and the priority of the anterior superior iliac spine area is automatically increased when the patient is in the prone position.

[0029] A correlation rule table was generated by defining association rules for the sacrococcygeal region, heel, and elbow based on the site priority table. As three key monitoring sites in the site priority table, the anatomical correlation and pressure sequence correlation of the sacrococcygeal region, heel, and elbow need to be analyzed to establish linkage rules. Based on the priority levels in the site priority table, the sacrococcygeal region is a first-priority site, the heel a second-priority site, and the elbow a third-priority site, establishing cascading trigger rules from high-priority sites to low-priority sites. When skin abnormalities occur in the sacrococcygeal region, simultaneous examination of the ischial tuberosity region is automatically triggered based on anatomical proximity. Based on the monitoring interval configuration in the site priority table, the monitoring frequency of the heel region is automatically increased from every 4 hours to every 2 hours (the sacrococcygeal region and heel are both major pressure sites in the supine position; early injury in the sacrococcygeal region indicates decreased overall skin tolerance, and the risk of pressure sores on the heel increases simultaneously). When skin abnormalities occur in the heel, simultaneous examination of the lateral and medial malleolus regions is automatically triggered based on lower limb circulation correlation, and investigation of other bony prominences is automatically triggered based on nutritional and metabolic correlation. A bidirectional linkage rule was established between the sacrococcygeal region and the heel, automatically increasing the monitoring frequency of the other region when a problem is detected in one area. A nutritional linkage rule was also established between the heel and the elbow, triggering a nutrition consultation request when problems occur in both areas simultaneously. These linkage rules were structured and encoded to generate a linkage rule table, stored in IF-THEN rule format. The condition section describes the triggering scenario, and the action section describes the execution operation. The linkage rule table contains five fields: triggering location, triggering condition, associated location, execution action, and rule priority. The triggering condition for the sacrococcygeal-heel linkage rule is the appearance of a non-reactive white-red spot on the sacrococcygeal region, and the execution action is increasing the heel monitoring frequency to once every 2 hours. The linkage rule table supports rule conflict detection; when multiple rules are triggered simultaneously, they are executed according to their priority.

[0030] Based on the association rule table, cross-site linkage early warning trigger conditions are configured to form a trigger condition set. The condition part of all IF-THEN rules constitutes the basis for early warning triggering. The condition expressions are logically decomposed and standardized. The trigger conditions in the association rule table are classified into two categories according to type: single-site trigger conditions and multi-site joint trigger conditions. Single-site trigger conditions include three types: skin color change trigger, skin temperature change trigger, and skin integrity damage trigger. The threshold for skin color change trigger is set as the appearance of a white-red spot or a purplish-red color change that does not change with pressure. The threshold for skin temperature change trigger is set as the local temperature being more than 2°C higher or lower than the surrounding skin (an increase in local temperature indicates an inflammatory response or deep tissue damage, while a decrease in local temperature indicates impaired blood circulation; both are early warning signals for pressure ulcer formation). The threshold for skin integrity damage trigger is set as the appearance of blisters, epidermal peeling, or ulcers. Multi-site joint trigger conditions include three types: simultaneous abnormal triggering of adjacent sites, simultaneous abnormal triggering of symmetrical sites, and abnormal triggering of multiple sites throughout the body. Warning levels are configured for various trigger conditions: a yellow warning for mild abnormalities in a single area, an orange warning for moderate abnormalities in a single area, and a red warning for severe abnormalities in a single area or combined abnormalities in multiple areas. The categorized trigger conditions and warning level configurations are integrated to form a trigger condition set. The trigger condition set includes four fields: trigger condition code, condition expression, warning level, and response time limit. The response time limit for a red warning is set to a nurse assessing the patient at the bedside within 15 minutes; for an orange warning, it is set to a nurse assessing the patient at the bedside within 30 minutes; and for a yellow warning, the assessment must be completed before the next shift handover.

[0031] A multi-site associated monitoring configuration is established based on a set of trigger conditions. The correspondence between the warning level and the monitoring task parameters determines the monitoring intensity under different states. Under a red warning state, continuous monitoring mode is activated, with the monitoring frequency increased to once per hour, and nursing record templates are automatically generated. Under an orange warning state, enhanced monitoring mode is activated, with the monitoring frequency increased to once every 2 hours, and pressure relief measure reminders are automatically pushed. Under a yellow warning state, regular monitoring mode is maintained, and monitoring tasks are executed according to the site priority table, with skin protection measure reminders automatically pushed. The warning rules of the trigger condition set are mapped to each pressure site, establishing a three-element association relationship of site-condition-action. The execution method and recording requirements of the monitoring tasks are configured. The monitoring execution methods include three modes: direct bedside examination, mobile terminal photo recording, and voice description to text. The recording requirements include four items: skin color description, skin temperature and touch, skin integrity status, and measurement data. The trigger condition binding, monitoring parameter configuration, execution method configuration, and recording requirement configuration are integrated to establish a complete multi-site associated monitoring configuration. Nurses receive monitoring task reminders through mobile terminals, enter the results on the terminal after completing the monitoring, and determine whether the warning conditions are triggered based on the trigger condition set. The monitoring data generated by the multi-site associated monitoring configuration is automatically written back to the skin monitoring record sub-table in the patient's risk profile, forming a complete skin condition tracking record.

[0032] Step S130: Perform indicator feature extraction on the test indicator data to form indicator change curves. Then, perform high-risk combination pattern recognition on the indicator change curves and Braden scale assessment data to generate a graded risk assessment benchmark.

[0033] Specifically, feature extraction is performed on the test indicator data to generate indicator change curves. For example... Figure 2As shown, historical measurement results of serum albumin and hemoglobin values ​​from patient risk records are retrieved and sorted by collection time to establish a two-dimensional coordinate system with time as the horizontal axis and deviation as the vertical axis. Deviation reflects the degree to which the indicator value deviates from the normal range; different indicators can be compared within the same coordinate system after deviation conversion. Curve fitting is performed on the discrete measurement points in the test indicator data. The original serum albumin data point 1 is used to generate a serum albumin fitting curve 2 using a cubic spline interpolation algorithm, and the original hemoglobin data point 3 is used to generate a hemoglobin fitting curve 4 using the same algorithm. The interpolation interval is set to 6 hours to reflect the intraday trend of the indicators. Feature extraction is performed on the fitted curves, including trend direction and magnitude of change. The trend direction is determined using linear regression analysis and is divided into three categories: downward trend 5, stable trend 6, and upward trend 7. The trend regression line 10 is used to assist in determining the trend direction. The normal range reference line 8 and the warning threshold line 9 are used to identify the normal range and warning boundary of the indicators. The feature extraction results of various indicators are integrated to generate indicator change curves. These curves consist of three components: the original data point sequence, the fitted curve parameters, and trend labels. The decreasing trend of serum albumin in the indicator change curves indicates deteriorating nutritional status, suggesting a continued decline in tissue repair capacity. The decreasing trend of hemoglobin in the indicator change curves indicates worsening anemia, suggesting persistent tissue oxygen deficiency.

[0034] In some embodiments, the step of generating a graded risk assessment benchmark by identifying high-risk combination patterns using the indicator change curve and the Braden scale assessment data includes: identifying malnutrition markers and tissue oxygen deficiency markers based on the indicator change curve to obtain a risk marker set; calculating the deviation of the risk marker set from a preset normal indicator range to generate a deviation value; identifying a synergistic deterioration effect based on the deviation value to generate a synergistic deterioration coefficient; and integrating the synergistic deterioration coefficient with the Braden scale assessment data to generate a graded risk assessment benchmark.

[0035] A risk marker set was obtained by identifying markers of malnutrition and insufficient tissue oxygenation based on the indicator change curves. The characteristics of the serum albumin curve reflect changes in the patient's nutritional status. A current serum albumin value below 35 g / L is marked as a sign of malnutrition, indicating insufficient protein reserves, decreased tissue repair capacity, and slowed wound healing. A continuous downward trend in serum albumin with a weekly decrease exceeding 5 g / L is marked as a sign of rapid nutritional deterioration, suggesting possible severe protein loss or hypercatabolism, requiring immediate nutritional intervention. The characteristics of the hemoglobin curve reflect changes in the patient's tissue oxygenation status. A current hemoglobin value below 90 g / L is marked as a sign of insufficient tissue oxygenation, indicating decreased red blood cell oxygen-carrying capacity, insufficient tissue oxygenation, and reduced skin pressure tolerance. A continuous downward trend in hemoglobin with a weekly decrease exceeding 10 g / L is marked as a sign of rapid deterioration in tissue oxygenation, suggesting possible occult bleeding or hematopoietic dysfunction, making pressure ulcers difficult to heal once formed. The various markers identified in the indicator change curves are summarized to generate a risk marker set. The risk marker set is represented by a Boolean vector, with each dimension corresponding to the presence or absence of a risk marker. The risk marker set includes four dimensions: malnutrition marker, rapid nutritional deterioration marker, insufficient tissue oxygen supply marker, and rapid deterioration of tissue oxygen supply marker. A value of 1 for each dimension indicates the presence of the marker, and a value of 0 indicates its absence.

[0036] The deviation degree is calculated by comparing the risk indicator set with the preset normal indicator range. The Boolean values ​​of each dimension in the risk indicator set identify the indicators that require deviation calculation. When a dimension in the risk indicator set has a value of 1, it indicates that the indicator corresponding to that dimension has a risk indicator and must be included in the deviation calculation range. The normal range boundary values ​​of each test indicator in the indicator benchmark domain serve as the reference benchmark for deviation calculation. The normal range boundary values ​​are read from the indicator benchmark domain. Indicators to be calculated are selected based on the indicator status of the risk indicator set. The current value of each selected indicator is compared with the normal range to calculate the deviation degree. The deviation degree calculation formula is D=(LV) / (UL), where D is the deviation degree, V is the current value of the indicator, L is the lower limit of the normal range, and U is the upper limit of the normal range. This formula calculates the deviation degree when the indicator is below the lower limit of the normal range. When the indicator value is within the normal range or above the upper limit, the deviation degree is 0. When the indicator value is below the lower limit of the normal range, the deviation degree is positive; the larger the deviation degree, the further it deviates from the normal range. The deviations of each indicator are weighted and summed using the formula Dt = Σ(Wi × Di), where Dt is the total deviation value, Wi is the weight coefficient of the i-th indicator, and Di is the deviation of the i-th indicator. The weight coefficients are determined based on the indicator's impact on pressure ulcer occurrence; the weight coefficient for serum albumin and hemoglobin is set to 0.5. These weight coefficients can be adjusted based on clinical experience. The weighted summation results generate a deviation value, ranging from 0 to 1, where 0 represents complete normality and 1 represents severe deviation. When a cluster of risk markers shows signs of rapid deterioration, the deviation of the corresponding indicator is weighted by 0.1 for deterioration after calculation.

[0037] For example, the step of identifying the synergistic deterioration effect and generating the synergistic deterioration coefficient based on the deviation value includes: splitting the deviation value into individual deviations of each indicator to form an individual deviation set; performing multi-indicator synchronous deterioration detection and identifying a synchronous deterioration indicator group based on the individual deviation set; performing cross-factor superposition effect analysis based on the synchronous deterioration indicator group to generate a superposition amplification value; and generating the synergistic deterioration coefficient through the superposition amplification value.

[0038] The deviation value is broken down into individual deviations for each indicator, forming an individual deviation set. The deviation value includes two components: serum albumin deviation and hemoglobin deviation. These two components are calculated independently before weighted aggregation, reflecting risk contributions from different dimensions. The two components of the deviation value are further split according to indicator category, with each sub-deviation stored independently and labeled with its indicator name. Trend analysis is performed on each sub-deviation, extracting the deviation sequence corresponding to the three most recent test results to determine whether the deviation of each indicator shows an upward, downward, or stable trend. Trend analysis can reveal the dynamic direction of indicator deterioration or improvement. The current deviation value and trend label of each indicator are integrated to form an individual deviation set, stored in key-value pair format. The key is the indicator name, and the value is a structure containing the deviation value and trend label. The serum albumin record in the individual deviation set includes the deviation value and trend label. A deviation of 0.25 with an upward trend indicates that serum albumin deviates from the normal range and the deviation is worsening. The records for hemoglobin in the single deviation set include a deviation value and a trend label. A deviation of 0.40 with an increasing trend indicates that the hemoglobin is deviating from the normal range and the deviation is worsening. The single deviation set supports sorting the output by deviation magnitude.

[0039] To identify synchronous deterioration indicator groups, a multi-indicator synchronous deterioration detection method was developed for single deviation sets. All indicator records in the single deviation set were traversed, and indicators with deviations exceeding a preset threshold (set at 0.2, meaning indicators deviating more than 20% from the normal range) were included in the detection range. Trend analysis was performed on the selected indicators, identifying those with an upward trend label. An upward trend indicates that the deviation is worsening and the patient's condition is continuously deteriorating. When two indicators in the single deviation set simultaneously meet the condition of exceeding the threshold and showing an upward trend, it is determined to be a state of synchronous deterioration of multiple indicators, indicating that the patient faces the combined threat of multiple risk factors. Indicators that simultaneously meet the conditions are grouped to generate synchronous deterioration indicator groups, and the name, deviation, and trend label of each indicator within the group are recorded. When the single deviation set shows an upward trend in serum albumin deviation and an upward trend in hemoglobin deviation, the two indicators constitute a synchronous deterioration indicator group, labeled "Nutrition-Anemia Synergistic Deterioration Group." This group indicates that the patient has both malnutrition and anemia as risk factors, and both are worsening. Analyzing the clinical significance of the synchronous deterioration indicator groups, the nutrition-anemia synergistic deterioration group indicates that patients' protein metabolism and hematopoietic function are simultaneously impaired, with the two influencing each other to form a vicious cycle (malnutrition leads to impaired absorption of hematopoietic raw materials and reduced globin synthesis, while anemia leads to hypoxia of the gastrointestinal mucosa and decreased digestive and absorptive functions, further aggravating malnutrition), significantly increasing the risk of pressure ulcers. The synchronous deterioration indicator groups are arranged in descending order of total deviation within the group; a higher total deviation within the group indicates a more severe degree of synergistic deterioration.

[0040] The superposition amplification value was generated by performing cross-factor superposition effect analysis based on the synchronous deterioration indicator group. The deviation values, trend labels, and medical correlation strength of each indicator within the group jointly determine the magnitude of the superposition effect. The medical correlation strength reflects the degree of pathophysiological association between the two indicators. The medical correlation strength between serum albumin and hemoglobin was set to 0.8. Both are closely related to protein metabolism. Malnutrition leads to insufficient amino acid supply, resulting in a lack of globin raw materials required for hemoglobin synthesis, thus aggravating anemia. The cross-factor superposition effect of the synchronous deterioration indicator group was calculated. The superposition effect calculation is based on a comprehensive consideration of the product of deviation values ​​of each indicator within the group and the medical correlation strength. The superposition amplification value calculation formula is A=D1×D2×R×K, where A is the superposition amplification value, D1 is the deviation value of the first indicator in the synchronous deterioration indicator group, D2 is the deviation value of the second indicator in the synchronous deterioration indicator group, R is the medical correlation strength, and K is the amplification base. The amplification base K was set to 2.5 to amplify the deviation product to a reasonable numerical range. When the trend labels of both indicators in the synchronous deterioration indicator group are increasing, it indicates that both indicators are continuously deteriorating. The superposition amplification value is calculated by adding 0.2 to the formula result to reflect the risk of accelerated deterioration. When the synchronous deterioration indicator group is the "nutrition-anemia synergistic deterioration group", the superposition amplification value is calculated based on serum albumin deviation, hemoglobin deviation, trend label, and medical association strength. The superposition amplification value ranges from 0 to 2.2, where 0 indicates no superposition effect, values ​​between 0 and 0.5 indicate a mild superposition effect, values ​​between 0.5 and 1 indicate a moderate superposition effect, and values ​​exceeding 1 indicate a significant cross-deterioration effect.

[0041] The synergistic deterioration coefficient is generated by superimposing amplification values. The superimposed amplification value reflects the additional risk arising from the synergistic deterioration of multiple indicators and needs to be integrated with the risk of deviation from individual indicators to comprehensively assess the patient's overall risk status. The superimposed amplification value is integrated with the total deviation value, which is taken from the weighted sum of deviation values ​​and represents the baseline risk level of individual indicator deviation. The formula for calculating the synergistic deterioration coefficient is C=(Dt+A) / N, where C is the synergistic deterioration coefficient, Dt is the total deviation value, A is the superimposed amplification value, and N is the normalization factor. The normalization factor N is set to 3.2 to normalize the calculation results to the range of 0-1, allowing the synergistic deterioration coefficient to be uniformly compared and integrated with other normalized indicators. When the total deviation value is low and the superimposed amplification value is small, the synergistic deterioration coefficient is at a low level, indicating that the patient's risk is controllable. When the total deviation value is high and the superimposed amplification value is large, the synergistic deterioration coefficient increases significantly, indicating that the patient faces the dual threat of single indicator abnormalities and synergistic deterioration of multiple indicators. The synergistic deterioration coefficient is truncated at an upper limit; when the calculated result exceeds 1, it is truncated to 1 to ensure that the coefficient value is within the valid range. The physical meaning of the synergistic deterioration coefficient is the overall risk level after comprehensively considering the deviation of single indicators and the synergistic deterioration of multiple indicators; the larger the coefficient value, the higher the risk of pressure ulcers. The synergistic deterioration coefficient is stored in a structured format, recording three fields: total deviation value, superposition amplification value, and final coefficient value. The synergistic deterioration coefficient supports time-series storage; each calculation result is saved in timestamp order.

[0042] A tiered risk assessment benchmark is generated by integrating the synergistic deterioration coefficient with the Braden scale assessment data. A comprehensive risk scoring model is established by fusing the synergistic deterioration coefficient with the total score of the Braden scale assessment data. The Braden scale assessment data provides scores and a total score across six dimensions, ranging from 6 to 23. The total score is converted into a normalized risk value ranging from 0 to 1, with lower total scores indicating higher risk. The conversion formula is Rb = (23 - S) / (23 - 6), where Rb is the Braden normalized risk value and S is the total score of the Braden scale. The synergistic deterioration coefficient and the normalized risk value of the Braden scale assessment data are then weighted and fused. The weight of the synergistic deterioration coefficient is set to 0.4, and the weight of the normalized risk value of the Braden scale assessment data is set to 0.6. The resulting comprehensive risk score is then used to classify risk levels based on the comprehensive risk score, generating a tiered risk assessment benchmark. When the comprehensive risk score is greater than or equal to 0.75, the tiered risk assessment benchmark is marked as extremely high risk, requiring the activation of the highest intensity preventive measures. When the comprehensive risk score is between 0.5 and 0.75, the risk assessment benchmark is marked as high-risk, requiring the initiation of enhanced preventative measures. When the comprehensive risk score is between 0.25 and 0.5, the risk assessment benchmark is marked as medium-risk, requiring the initiation of routine preventative measures. When the comprehensive risk score is less than 0.25, the risk assessment benchmark is marked as low-risk, requiring the maintenance of basic care measures. The risk assessment benchmark also outputs a major risk factor analysis, including the main indicators leading to high risk and a predicted trend of deterioration. The risk assessment benchmark supports dynamic updates, automatically recalculating the risk level when test result data or Braden scale assessment data change.

[0043] Step S140: Perform image pattern recognition to extract skin feature parameters from the skin image data, perform time-axis dynamic tracking based on the skin image data to generate a skin state change sequence, and perform redness detection on the skin state change sequence based on multi-site correlation monitoring configuration to generate redness localization results.

[0044] Specifically, image pattern recognition is performed on skin image data to extract skin feature parameters. Skin image data is retrieved from the pressure site sub-table of the patient's risk file. This data includes standardized images of pressure sites such as the sacrum, heel, and elbow. Each image is captured under the same angle and lighting conditions to ensure comparability. A dedicated preprocessing operation for feature extraction is performed on the skin image data. This preprocessing process includes three steps: image denoising, color correction, and geometric correction. Image denoising uses a Gaussian filtering algorithm to remove random noise from the shooting process. Color correction uses a white balance algorithm to eliminate color casts caused by different lighting conditions. Geometric correction uses a perspective transformation algorithm to eliminate distortion caused by the shooting angle. Image pattern recognition is then performed on the preprocessed skin image data. The recognition algorithm uses a convolutional neural network model. The model input is the standardized image, and the model output is skin state classification labels and feature data. Skin feature parameters are extracted from the recognition results. These parameters include three categories: color features, texture features, and morphological features. Color features include the RGB mean, HSV hue value, and erythema index of the skin region. The erythema index reflects the degree of skin redness; a higher erythema index indicates more pronounced redness. Texture features include skin texture roughness and contrast; increased roughness suggests dry skin or abnormal keratinization. Morphological features include the area and perimeter of abnormal regions, which are used to quantify the size and extent of inflamed areas. Skin feature parameters are stored in feature vector format, along with the timestamp of feature extraction and the corresponding image file index.

[0045] In some embodiments, the step of performing time-axis dynamic tracking to generate a skin state change sequence based on the skin image data includes: extracting the acquisition timestamps of the skin image data to form time stamps; arranging the skin image data in a time sequence according to the time stamps to generate a time-series image set; registering images of the same body part in the time-series image set to generate a registered image sequence; and performing change rate quantization and region labeling on the registered image sequence to form a skin state change sequence.

[0046] The acquisition timestamps of skin image data are extracted to form time stamps. The acquisition timestamps are recorded in the metadata of the skin image data and stored in the EXIF ​​information of the image files, in a standard time format of year-month-day-hour-minute-second. The acquisition timestamps of each image in the skin image data are parsed to extract the six time components (year, month, day, hour, minute, second), which are then converted to a unified Unix timestamp format for easy comparison and sorting. The acquisition timestamps of all skin image data from the same patient and the same area are summarized to generate a time stamp sequence, arranged chronologically. The time stamps are stored in a structured format, with each record containing three fields: a Unix timestamp, a formatted time string, and an image file index. The time stamp sequence supports filtering by time range, allowing extraction of all records within a specified date or time period. The completeness of the time stamps directly affects the accuracy of dynamic timeline tracking; when abnormal jumps in time intervals are detected, data missing information is automatically indicated.

[0047] A time-series image set is generated by arranging skin image data according to time stamps. The order of the time stamp sequence determines the organization of the skin image data. The skin image data is arranged in ascending order according to the Unix timestamps of the time stamps to ensure that the image sequence accurately reflects the changes in skin condition over time. The first image in the arrangement is the earliest taken image, and the last image is the most recent taken image. The time interval between images reflects the monitoring density and regularity. An index relationship is established for the arranged skin image data. The index adopts a doubly linked list structure. Each node stores the file path of the current image and records the node pointers of the preceding and following images. The time-series arranged skin image data and its index relationship are encapsulated to generate a time-series image set. The time-series image set is stored using an ordered set data structure to maintain the temporal order of the images. The time-series image set supports forward and backward traversal. Forward traversal accesses images from early to late time to track disease development, while backward traversal accesses images from late to early time to trace the origin of the disease. The time-series image set supports random access, allowing quick location of a specific image based on the index position or timestamp.

[0048] For a time-series image set, perform image registration of the same region to generate a registered image sequence. For example... Figure 3As shown, original images 11 at time T1, 12 at time T2, and 13 at time T3 in the time-series image set are used as inputs for registration processing. Images from earlier times are used as reference images, and images from later times are used as images to be registered. Feature points are extracted from both the reference and images to be registered in the time-series image set. The SIFT algorithm or ORB algorithm is used for feature point extraction. SIFT extracts feature points with scale and rotation invariance, while ORB is faster and suitable for real-time processing. The extracted feature points are then matched. Matching lines 22 represent the correspondence between feature points in adjacent images. The nearest neighbor algorithm is used for feature point matching, calculating the descriptor distance between feature points in the image to be registered and those in the reference image. The feature point pair with the smallest distance is marked as a matching pair. The image transformation matrix is ​​calculated based on the matching pair. The transformation matrix uses an affine transformation model, including translation, rotation, and scaling components. The transformation matrix is ​​robustly estimated using the RANSAC algorithm to eliminate the influence of mismatched points. The transformation matrix is ​​applied to the image to be registered, transforming it into the coordinate system of the reference image. The transformed image is spatially aligned with the reference image. Registration is performed sequentially on all images in the time-series image set, generating a registered image sequence. The coordinate systems of registered images T1 (14), T2 (15), and T3 (16) are identical, meaning the same anatomical location has the same pixel coordinates in images at different time points.

[0049] A skin condition change sequence is generated by performing rate of change quantization and region annotation on the registered image sequence. Differential analysis is then performed on adjacent images in the registered image sequence, such as... Figure 3As shown, the difference between the registered T1 image 14 and the registered T2 image 15 generates the T1-T2 difference image 17, and the difference between the registered T2 image 15 and the registered T3 image 16 generates the T2-T3 difference image 18. Each region in the difference image is marked into three states based on pixel value changes: regions where pixel value changes exceed a degradation threshold are marked as degraded regions 20 and displayed in red; regions where pixel value changes exceed an improvement threshold are marked as improved regions 21 and displayed in green; and regions where pixel value changes do not exceed the threshold are marked as unchanged regions 19 and displayed in gray. Thresholding segmentation is performed on the difference images, marking pixels with differences exceeding the threshold as changed pixels. The threshold is set to 5% of the image's dynamic range, and this threshold can be configured and adjusted according to the characteristics of the image acquisition device. After thresholding, a binary mask of the changed regions is obtained. Connectivity analysis is performed on the changed regions, aggregating spatially adjacent changed pixels into changed patches, filtering out noise patches with an area less than 0.01% of the total image area. For each changed patch, a quantitative analysis was performed to extract its area, centroid coordinates, and average change magnitude. The change analysis results at each time point in the registered image sequence were summarized to generate a skin condition change sequence. This skin condition change sequence is stored using a time-series data structure, with each record containing three fields: timestamp, image index, and a list of changed areas.

[0050] In some embodiments, the step of performing pressure ulcer region detection and generating pressure ulcer localization results based on the multi-site associated monitoring configuration on the skin condition change sequence includes: determining a list of sites to be detected based on the multi-site associated monitoring configuration to form a detection site sequence; performing early weak pressure ulcer enhancement processing on the skin condition change sequence according to the detection site sequence to generate an enhanced image sequence; performing automatic pressure ulcer staging calibration on the enhanced image sequence to generate a staging identifier; and integrating the staging identifier with the detection site sequence to form a pressure ulcer localization result.

[0051] A list of pressure ulcer sites is determined based on the multi-site associated monitoring configuration, forming a detection site sequence. This configuration includes a list of codes and priority settings for all pressure ulcer sites, using a standardized hierarchical coding system as defined earlier. Based on the priority settings, the site codes are prioritized, with first-priority sites at the beginning of the sequence and third-priority sites at the end. The sorted list is then filtered by status, prioritizing sites in a warning state according to the monitoring configuration. Sites currently in a warning state are moved to the front of the sequence. The filtered and sorted list is then encapsulated to form the detection site sequence, which determines the execution order of pressure ulcer detection. This sequence is stored using a priority queue data structure, with the highest priority sites at the head and the lowest priority sites at the tail. The sacrum-coccygeal region (SM) code is typically placed first in the detection sequence because it is the most common site for pressure ulcers. Sites in a warning state are temporarily prioritized to ensure they are detected first.

[0052] Early, subtle pressure-induced redness enhancement was performed on the skin condition change sequence according to the detection site sequence to generate enhanced image sequences. Based on the order of the detection site sequence, image data corresponding to each site was extracted sequentially from the skin condition change sequence, ensuring that images of high-priority sites were processed first. Early, subtle pressure-induced redness enhancement was performed on the extracted image data. This enhancement aimed to improve the detection rate of early stage I pressure ulcers. Stage I pressure ulcers are characterized by erythema that does not turn white upon pressure. Early erythema may be very faint and difficult to detect with the naked eye (in low light or when the patient's skin is dark, the erythema of early stage I pressure ulcers may only show slight tonal changes; enhancement using the Lab color space can significantly improve the detection rate). The enhancement process combined color space transformation and contrast stretching. First, the image was converted from the RGB color space to the Lab color space. The a channel in the Lab color space reflects red-green contrast, which is suitable for detecting erythematous areas. Histogram equalization was then performed on the a channel of the Lab color space to enhance the contrast between red areas and normal skin areas. Thresholding segmentation was performed on the equalized a-channel to extract suspected erythema regions. The threshold was dynamically determined based on the statistical distribution of a-channel values ​​in normal skin areas; pixels exceeding the mean plus twice the standard deviation were marked as suspected erythema. Enhancement processing was sequentially applied to images of all areas in the skin condition change sequence. The enhanced images were then reorganized according to their original temporal order to generate an enhanced image sequence. The contrast of erythema regions in each image of the enhanced image sequence was significantly improved. The enhancement parameters of the enhanced image sequence were differentiated according to the characteristics of each area in the detection sequence; the heel area required increased enhancement intensity due to its thicker stratum corneum.

[0053] Automatic staging of pressure ulcers was implemented using enhanced image sequences. Erythematous regions detected in the enhanced image sequences were used for pressure ulcer staging, adopting the NPUAP / EPUAP / PPPIA pressure ulcer staging guidelines, which are internationally recognized standards. Morphological and color features of the erythematous regions in the enhanced image sequences were used as the basis for staging. Morphological features included region area and boundary clarity, while color features included erythema intensity and color uniformity. The extracted features were comprehensively analyzed to determine the pressure ulcer stage of the erythematous regions. Stage I pressure ulcers were defined as intact skin with erythematous patches that did not turn white upon pressure; areas with erythema intensity exceeding a threshold and good skin integrity in the enhanced image sequence were designated as Stage I. Stage II pressure ulcers were defined as partial loss of skin, manifesting as superficial ulcers or intact or ruptured serous vesicles. Stage III pressure ulcers were defined as full-thickness skin loss, with visible subcutaneous fat but no exposed bone, tendons, or muscles. Stage IV pressure ulcers were defined as full-thickness tissue loss, with exposed bone, tendons, or muscles. The criteria for identifying suspicious deep tissue injury are intact skin with purple or brownish-red discoloration; areas showing purple-red discoloration in enhanced image sequences are designated as suspicious deep tissue injury. Based on the pressure ulcer staging results, staging identifiers are generated using standardized codes: Stage I is coded as STG-1, Stage II as STG-2, Stage III as STG-3, Stage IV as STG-4, and Suspicious deep tissue injury as SDTI. The staging identifiers also record the confidence score of the staging determination, using a 0-100 scale. A score above 80 indicates high confidence, a score between 60 and 80 indicates the need for manual review, and a score below 60 indicates an unreliable determination.

[0054] The staging identifier and the detection site sequence are integrated to form the pressure ulcer localization result. A correlation is established between the staging identifier and the detection site sequence, binding the staging identifier of each pressure ulcer area to its corresponding anatomical site code. The detection site sequence provides a standardized code for the location of the pressure ulcer area, while the staging identifier provides the pressure ulcer stage and confidence score. The staging code of the staging identifier is appended to the site code to form a complete pressure ulcer localization code. The complete code for stage I pressure ulceration in the sacrococcygeal region is SM-01-STG-1, and the complete code for suspected deep tissue injury in the left heel is HL-01-SDTI. The spatial coordinate information of the pressure ulcer area in the enhanced image sequence, including the center point coordinates and the area, is integrated with the staging information. The site code, staging identifier, center coordinates, and area are integrated to form a complete pressure ulcer localization result. The pressure ulcer localization result is stored in a structured format, with each record containing five fields: localization code, staging identifier, confidence score, center coordinates, and area. The pressure ulcer localization result supports filtering by site and by stage. The results of redness localization are automatically updated to the skin monitoring record sub-table in the patient's risk profile, forming a complete skin monitoring and tracking record. Changes in the redness localization results automatically trigger a risk level reassessment. If a new stage I redness is detected, the system automatically feeds this information back to the tiered risk assessment benchmark, triggering a recalculation of the risk level, and the risk level is automatically upgraded by one level.

[0055] Step S150: The pressure ulcer location results are matched with skin feature parameters to identify risk areas. Based on the risk areas and the graded risk assessment benchmark, the risk level is dynamically adjusted to generate a damage severity value. Based on the damage severity value, personalized nursing measures are matched to generate nursing decision results, thus completing intelligent pressure ulcer risk management and nursing decision support.

[0056] Specifically, the pressure erythema localization results are matched with skin feature parameters to identify risk areas. The pressure erythema localization results provide the localization code, stage identifier, confidence score, center coordinates, and area of ​​each pressure erythema region. Skin feature parameters provide the corresponding color, texture, and morphological features. The center coordinates from the pressure erythema localization results are spatially aligned with the feature vectors from the skin feature parameters to extract local feature parameters within the pressure erythema region. Subsequent damage feature analysis is performed only on regions with a confidence score higher than 60. Damage feature analysis is performed on the areas indicated by the pressure erythema localization results. Damage features include erythema depth, erythema boundary features, and erythema progression features. Erythema depth is assessed using the spatial distribution gradient of the erythema index; a significantly higher erythema index in the central region than in the peripheral region suggests a risk of deep tissue damage. Erythema boundary features are assessed using boundary clarity; clear and regular boundaries indicate acute pressure injury, while blurred and irregular boundaries suggest chronic tissue damage. Erythema progression features are assessed by comparing with historical data; an increase in area or erythema index indicates damage progression. Based on the damage characteristic analysis results, the areas in the redness localization results are divided into different risk area levels. The risk areas are classified into three levels: high-risk areas are areas with an erythema index exceeding 0.5 and showing a progressive trend; medium-risk areas are areas with an erythema index between 0.3 and 0.5 or with blurred boundaries; and low-risk areas are areas with an erythema index between 0.15 and 0.3 and are stable.

[0057] In some embodiments, the step of dynamically adjusting the risk level based on the risk area and the graded risk assessment benchmark to generate a damage severity value includes: analyzing the damage area and damage depth based on the risk area to form a damage quantification index; matching the damage quantification index with the graded risk assessment benchmark to generate an initial risk level; conducting a nursing intervention responsiveness assessment for the initial risk level to generate an intervention effectiveness index; and dynamically adjusting the initial risk level based on the intervention effectiveness index to generate a damage severity value.

[0058] Damage quantification indicators are formed based on the analysis of damage area and depth in risk zones. The three-level classification results of risk zones determine the calculation weights for damage quantification: high-risk zones have a damage quantification weight of 1.0, medium-risk zones have a weight of 0.6, and low-risk zones have a weight of 0.3. The area of ​​each risk level zone is statistically analyzed, with the area derived from the area field in the pressure erythema localization results, in square centimeters. A depth analysis of the erythema index (EI) of each risk level zone is performed, based on the color features in the skin characteristic parameters. The maximum EI value within each zone is extracted, reflecting the depth of the most severe injury. The damage area and EI assessment results for high-risk, medium-risk, and low-risk zones are multiplied by their corresponding weights and then integrated to form the damage quantification indicators. The damage quantification indicators include three components: absolute damage area, damage area percentage, and peak EI value. The damage area percentage is calculated by dividing the weighted damage area by the total area of ​​the pressure site. The damage quantification indicators support multi-region aggregation. When multiple risk zones exist for the same patient, they are accumulated separately according to the three-level classification before being weighted and aggregated to reflect the patient's overall damage burden.

[0059] Initial risk levels are generated by matching quantified damage indicators with a risk assessment grading benchmark. Two core indicators in the quantified damage indicators—damage area percentage and peak erythema index—determine the severity of local damage, while the risk level and co-morbidity coefficient in the risk assessment benchmark reflect the patient's overall risk status. A mapping relationship between quantified damage indicators and risk levels is established: a high-risk level is defined as a damage area percentage exceeding 20% ​​or a peak erythema index exceeding 0.5; a medium-risk level is defined as a damage area percentage between 10-20% or a peak erythema index between 0.3-0.5; and a low-risk level is defined as a damage area percentage less than 10% and a peak erythema index less than 0.3. These thresholds can be adjusted based on clinical experience. The risk level mapped from the quantified damage indicators is compared with the current risk level in the risk assessment grading benchmark, and the higher level is taken as the preliminary result. The preliminary result is then corrected using the co-morbidity coefficient; if the co-morbidity coefficient in the risk assessment benchmark exceeds 0.6, the preliminary result level is upgraded by one level. The revised result is the initial risk level, which adopts a four-level classification system, including very high risk, high risk, medium risk and low risk, and will serve as the baseline reference for the nursing intervention response assessment.

[0060] For example, the step of conducting a nursing intervention responsiveness assessment to generate intervention effectiveness indicators based on the initial risk level includes: retrospectively executing nursing interventions based on the initial risk level to form intervention execution records; extracting skin condition changes before and after intervention execution based on the intervention execution records to generate condition comparison data; performing intervention effectiveness quantification scoring based on the condition comparison data to generate intervention score values; and establishing intervention effectiveness indicators based on the intervention score values.

[0061] Nursing interventions are performed based on the initial risk level, and retrospective intervention records are generated. The time window for assessment retrospection is determined according to the initial risk level: the retrospection window is set to the past 24 hours for very high and high risk levels, and the retrospection window is set to the past 72 hours for intermediate and low risk levels. All nursing operation records related to the patient within this time window are retrieved from the nursing execution data table associated with the patient's risk file. These records include turning records, skin examination records, decompression measures records, and nutritional support records. The retrieved nursing operation records are structured and categorized by operation type and operation time. The organized nursing operation information is then encapsulated to form intervention execution records, which are stored in a time-series table format. Each record in the intervention execution record contains three fields: operation time, operation type, and operation content. An example of a turning record in the intervention execution record is "2024-01-15 08:00, turning, from supine position to left lateral decubitus position 30 degrees". An example of a skin examination record in the measure execution log is "2024-01-15 10:00, skin examination, sacrococcygeal skin intact without erythema". Measure execution records support filtering by operation type and by time range.

[0062] Based on the implementation records, skin condition changes before and after the implementation of measures were extracted to generate condition comparison data. The implementation time points of key nursing measures in the implementation records served as the dividing lines for condition comparison. Key nursing measures included measures with clear starting points, such as the first intensive turning, the first use of a pressure-reducing mattress, and the first nutritional intervention. Baseline skin condition data before the implementation of key measures was extracted from the skin condition change sequence. The baseline data included the erythema index and lesion area from the most recent skin examination before the implementation of the measures. Follow-up data of skin condition after the implementation of key measures was extracted from the skin condition change sequence. The follow-up data included skin examination results at three time points: 24 hours, 48 ​​hours, and 72 hours after the implementation of the measures. The measure information in the implementation records was correlated with the corresponding baseline and follow-up data to generate condition comparison data. The condition comparison data adopted a before-and-after comparison design. Each record included four fields: measure name, baseline skin condition, follow-up skin condition, and change amount. In the condition comparison data, the change amount was calculated by subtracting the baseline skin condition index from the follow-up skin condition index. A negative change in the erythema index indicated a reduction in erythema, and a negative change in the lesion area indicated a reduction in the area. An example of state comparison data is: "Strengthening the turning-over measure, baseline erythema index 0.35, erythema index 0.22 after 72 hours, change -0.13". State comparison data supports joint analysis of multiple measures, and can analyze the synergistic effect of the combination of measures when multiple measures are implemented simultaneously.

[0063] The effectiveness of the measures is quantified and scored based on the condition comparison data to generate a measure score. The changes in skin condition for each measure in the condition comparison data reflect the degree of improvement or deterioration of the skin condition after implementation; the greater the change, the more significant the effect of the measure. The change in erythema index is scored as follows: a reduction of more than 30% is scored as 100 points, a reduction of 20-30% as 80 points, a reduction of 10-20% as 60 points, a change within ±10% as 50 points, an increase of 10-20% as 30 points, and an increase of more than 20% as 10 points. The change in lesion area is also scored as follows: a reduction of more than 30% is scored as 100 points, a reduction of 20-30% as 80 points, a reduction of 10-20% as 60 points, a change within ±10% as 50 points, an increase of 10-20% as 30 points, and an increase of more than 20% as 10 points. The erythema index score and lesion area score in the condition comparison data are weighted and averaged with weights of 0.6 and 0.4, respectively. These weights can be adjusted according to clinical needs. The weighted average result is the intervention score. The intervention score ranges from 10 to 100 points, with higher scores indicating better intervention effectiveness. Intervention scores are calculated separately for each intervention type: scores for turning interventions, decompression interventions, and nutritional interventions are stored separately.

[0064] An effectiveness index for nursing interventions was established based on intervention score scores. The scores of all intervention types were summarized and analyzed. Interventions with scores exceeding 60 were marked as effective, those below 40 as ineffective, and those between 40 and 60 as having uncertain effects. The ratio of actual nursing procedures performed to planned procedures within the retrospective time window was defined as the intervention execution rate. The skin improvement rate was calculated by dividing the average change in erythema index in the status comparison data by the average baseline erythema index. A positive skin improvement rate indicates an improving trend, while a negative rate indicates a worsening trend. The above analysis results were integrated to establish an effectiveness index, which includes three components: a comprehensive effectiveness score, an intervention execution rate, and a skin improvement rate. The comprehensive effectiveness score is the weighted average of all intervention scores, with weights determined by the frequency of intervention execution; higher execution frequencies result in higher weights. A comprehensive effectiveness score of 82 indicates good overall nursing intervention effectiveness, and an intervention execution rate of 90% indicates good implementation of the nursing plan.

[0065] The initial risk level is dynamically adjusted based on the effectiveness indicators of the measures to generate damage severity values. The overall score of the effectiveness indicators and the skin improvement rate jointly determine the adjustment direction of the initial risk level. When the overall score is higher than 80 and the skin improvement rate is positive, the initial risk level is lowered by one level. When the overall score is lower than 50 or the skin improvement rate is negative, the initial risk level is raised by one level. When the overall score is between 50 and 80, the initial risk level remains unchanged. When the measure implementation rate is lower than 70%, priority is given to sending implementation reminders rather than upgrading the care plan, because the effectiveness of the plan itself cannot be judged when the measures are not fully implemented. The adjusted risk level is converted into a numerical damage severity value: extremely high risk corresponds to a damage severity value of 5, high risk corresponds to a damage severity value of 4, medium risk corresponds to a damage severity value of 3, low risk corresponds to a damage severity value of 2, and no risk corresponds to a damage severity value of 1. The damage severity value is fine-tuned within a range of ±0.5 based on the skin improvement rate. When the skin improvement rate is greater than 20%, the damage severity value decreases by 0.5; when the skin improvement rate is less than -20%, the damage severity value increases by 0.5. The damage severity value ranges from 1.0 to 5.5. Changes in the damage severity value automatically trigger an update of the monitoring frequency configured for multi-site correlation monitoring.

[0066] Personalized nursing interventions are matched to injury severity values ​​to generate nursing decision outcomes. Based on the numerical range of the injury severity value, a nursing plan of appropriate intensity is matched from a pre-built standardized nursing intervention knowledge base. When the injury severity value is in the range of 4-5.5, the nursing plan includes intensive decompression measures, frequent skin examinations, and nutritional support interventions. Intensive decompression measures shorten the turning interval to once every 1 hour and use a high-specification alternating air mattress; frequent skin examinations are conducted every 2 hours. When the injury severity value is in the range of 2-3, a standard intensive nursing plan is matched, with the turning interval set to once every 2 hours, and foam pads used for local decompression. When the injury severity value is less than 2, a basic preventive nursing plan is matched, maintaining the standard turning schedule and skin protection measures. The matched nursing interventions are structured and arranged to generate nursing decision outcomes, which include a list of nursing interventions and an implementation schedule. Nursing interventions in the nursing decision outcomes are targeted based on low scores on the Braden scale; for example, increased examination frequency is recommended for impaired sensory abilities, and an incontinence care plan is recommended for low moisture scores. Once the nursing decision is generated, it is pushed to the nurse's workstation, where nurses receive task reminders and complete the execution record on their mobile devices. When skin condition improves and the lesion severity value decreases, the monitoring frequency and intensity of nursing interventions are automatically reduced; when skin condition worsens and the lesion severity value increases, the nursing plan is automatically upgraded. Through a continuous monitoring-assessment-intervention-feedback cycle, the system shifts from "post-pressure ulcer management" to "predicting and preventing pressure ulcer occurrence," achieving intelligent pressure ulcer risk management and nursing decision support.

[0067] To implement the intelligent pressure ulcer risk management and nursing decision support method corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, see [link to documentation]. Figure 4 , Figure 4 This diagram illustrates the structural block diagram of the intelligent pressure ulcer risk management and nursing decision support system 400 provided in an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The intelligent pressure ulcer risk management and nursing decision support system 400 provided in this embodiment includes: The data acquisition module 401 is used to collect test index data, skin image data and Braden scale assessment data of the patient's pressure site, and establish a patient risk profile based on the patient identifier associated with the test index data and the skin image data. The baseline establishment module 402 is used to identify normal indicator areas based on the patient risk file to form an indicator baseline domain, and to establish a multi-site correlation monitoring configuration based on the indicator baseline domain. Risk assessment module 403 is used to perform indicator feature extraction on the test indicator data to form indicator change curves, and to perform high-risk combination pattern recognition on the indicator change curves and the Braden scale assessment data to generate a graded risk assessment benchmark. Image analysis module 404 is used to perform image pattern recognition to extract skin feature parameters from the skin image data, perform time-axis dynamic tracking to generate a skin state change sequence based on the skin image data, and perform redness region detection on the skin state change sequence based on the multi-site association monitoring configuration to generate redness localization results. The decision output module 405 is used to perform damage feature matching to identify risk areas by matching the redness positioning results with the skin feature parameters, dynamically adjust the risk level based on the risk areas and the graded risk assessment benchmark to generate damage degree values, and match personalized nursing measures based on the damage degree values ​​to generate nursing decision results.

[0068] The aforementioned intelligent pressure ulcer risk management and nursing decision support system 400 can implement the intelligent pressure ulcer risk management and nursing decision support method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0069] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. An intelligent pressure ulcer risk management and nursing decision support method, characterized in that, include: Collect test data, skin image data, and Braden scale assessment data of the patient's pressure points, and establish a patient risk profile based on the patient identifier associated with the test data and the skin image data; Based on the patient risk profile, normal indicator areas are identified to form an indicator benchmark domain, and a multi-site correlation monitoring configuration is established based on the indicator benchmark domain; The test indicator data are subjected to indicator feature extraction to form an indicator change curve. The indicator change curve is then compared with the Braden scale assessment data to generate a graded risk assessment benchmark by high-risk combination pattern recognition. Image pattern recognition is performed on the skin image data to extract skin feature parameters. Based on the skin image data, dynamic time-axis tracking is performed to generate a skin state change sequence. Based on the multi-site associated monitoring configuration, redness region detection is performed on the skin state change sequence to generate redness localization results. The redness localization result is matched with the skin feature parameters to identify risk areas. Based on the risk areas and the graded risk assessment benchmark, the risk level is dynamically adjusted to generate a damage degree value. Based on the damage degree value, personalized nursing measures are matched to generate a nursing decision result.

2. The method according to claim 1, characterized in that, The process of establishing a patient risk profile based on the association of the test index data and the skin image data with patient identifiers includes: Data quality verification is performed on the aforementioned test indicator data to generate verification-passed data; The serum albumin and hemoglobin values ​​are obtained from the verified data to form a nutritional status label; Based on the skin image data, the location information of the pressure site is extracted to generate a site identifier; The patient risk profile is established by associating the verified data, the nutritional status label, and the site identifier with the patient identifier.

3. The method according to claim 1, characterized in that, The establishment of a multi-site correlation monitoring configuration based on the indicator benchmark domain includes: Based on the aforementioned index benchmark domain, the monitoring priority of each pressure-bearing part is determined, and a part priority table is generated; Based on the aforementioned location priority table, association rules are defined for the sacrum, coccyx, heel, and elbow to generate an association rule table; Based on the aforementioned association rule table, cross-location linkage early warning triggering conditions are configured to form a triggering condition set; A multi-site correlation monitoring configuration is established based on the aforementioned trigger condition set.

4. The method according to claim 1, characterized in that, The step of generating a graded risk assessment benchmark by performing high-risk combination pattern identification using the indicator change curve and the Braden scale assessment data includes: Based on the curves showing changes in these indicators, a risk marker set was obtained by identifying markers of malnutrition and insufficient tissue oxygen supply. The deviation value is generated by calculating the degree of deviation between the risk indicator set and the preset normal indicator range. Based on the deviation value, a synergistic deterioration effect is identified, and a synergistic deterioration coefficient is generated; A graded risk assessment benchmark is generated by integrating the co-deterioration coefficient with the Braden scale assessment data.

5. The method according to claim 1, characterized in that, The step of performing pressure-induced redness region detection and generating pressure-induced redness localization results on the skin state change sequence based on the multi-site correlation monitoring configuration includes: Based on the multi-site associated monitoring configuration, a list of sites to be detected is determined to form a detection site sequence; For the skin condition change sequence, early weak redness enhancement processing is performed according to the detection site sequence to generate an enhanced image sequence; Automatic pressure ulcer staging and generation of staging identifiers are performed on the enhanced image sequence; The stage identifier is integrated with the detection site sequence to form the redness localization result.

6. The method according to claim 1, characterized in that, The step of generating a skin state change sequence by performing time-axis dynamic tracking based on the skin image data includes: Extract the acquisition timestamp of the skin image data to form a time stamp; Based on the time markers, the skin image data is arranged chronologically to generate a time-series image set; For the time-series image set, perform image registration of the same part to generate a registered image sequence; The registered image sequence is subjected to rate of change quantization and region labeling to form a skin state change sequence.

7. The method according to claim 1, characterized in that, The process of dynamically adjusting the risk level and generating a damage severity value based on the risk area and the graded risk assessment benchmark includes: Based on the analysis of the aforementioned risk areas, damage area and damage depth are used to form quantitative damage indicators; The damage quantification index is matched with the graded risk assessment benchmark to generate an initial risk level; A nursing intervention responsiveness assessment was conducted based on the initial risk level to generate indicators of intervention effectiveness. The initial risk level is dynamically adjusted based on the effectiveness index of the measures to generate a damage degree value.

8. The method according to claim 4, characterized in that, The step of identifying the synergistic deterioration effect and generating the synergistic deterioration coefficient based on the deviation value includes: The deviation value is broken down into individual deviations of each indicator to form an individual deviation set; For the single deviation set, perform multi-indicator synchronous deterioration detection and identify synchronous deterioration indicator groups; Based on the aforementioned synchronous deterioration index group, a superposition amplification value is generated through cross-factor superposition effect analysis. The synergistic deterioration coefficient is generated by the superimposed amplification value.

9. The method according to claim 7, characterized in that, The method of generating effectiveness indicators for nursing intervention responsiveness assessment based on the initial risk level includes: Nursing interventions were performed based on the initial risk level, and intervention execution records were generated retrospectively. Based on the implementation records of the aforementioned measures, skin condition changes before and after the implementation of the measures are extracted to generate condition comparison data; Based on the aforementioned status comparison data, the effectiveness of the measures is quantitatively scored to generate a measure score value; An effectiveness index for the measures will be established based on the scoring values ​​of the measures.

10. An intelligent pressure ulcer risk management and nursing decision support system, characterized in that, include: The data acquisition module is used to collect test index data, skin image data and Braden scale assessment data of the patient's pressure site, and establish a patient risk profile based on the patient identifier associated with the test index data and the skin image data; The baseline establishment module is used to identify normal indicator areas based on the patient risk file to form an indicator baseline domain, and to establish a multi-site correlation monitoring configuration based on the indicator baseline domain. The risk assessment module is used to extract indicator features from the test indicator data to form indicator change curves, and to perform high-risk combination pattern recognition by combining the indicator change curves with the Braden scale assessment data to generate a graded risk assessment benchmark. The image analysis module is used to perform image pattern recognition to extract skin feature parameters from the skin image data, perform time-axis dynamic tracking to generate a skin state change sequence based on the skin image data, and perform redness region detection on the skin state change sequence based on the multi-site correlation monitoring configuration to generate redness localization results. The decision output module is used to perform damage feature matching between the redness positioning result and the skin feature parameters to identify risk areas, dynamically adjust the risk level based on the risk areas and the graded risk assessment benchmark to generate damage severity values, and match personalized nursing measures based on the damage severity values ​​to generate nursing decision results.

Images