A wound care decision support system and method based on multi-source data

Abstract

The present application relates to the technical field of data processing analysis, in particular to a wound care decision support system and method based on multi-source data, the method provided by the present application obtains the basic data of the current patient from the data source, and the basic data is preprocessed to generate the first standard database of the current patient; logical mapping is set, different decision strategies are obtained based on the logical mapping, and a reward function is established for a plurality of basic parameter variables of the patient in a time period using different decision strategies; the decision strategies are sorted from high to low according to the reward function score, which is the priority recommendation order of the measures of the decision support system, providing auxiliary judgment information for medical staff and improving the efficiency of medical staff in work.

Description

Technical Field

[0001] This invention relates to the field of data processing and analysis technology, and more specifically, to a wound care decision support system and method based on multi-source data. Background Technology

[0002] Nursing Decision Support Systems (NDSS) can serve as a powerful support for nursing staff in this process, standardizing, streamlining, and automating the assessment, planning, treatment, and evaluation of wound patients. It can accurately assess, monitor, and record patients' wounds, provide precise decision recommendations based on individual patient characteristics, provide basic data for the management of different types of wounds, and assist clinical nurses in improving the quality of wound care.

[0003] A database search revealed that there are currently few patents related to wound care decision support systems. Some patents involve wound assessment or treatment selection, but none fully demonstrate a complete wound management process, automated procedures, or evidence-based and adaptive intervention improvements.

[0004] For example, patent CN119581060A discloses a wound care intelligent decision-making method and system based on high-precision image recognition. The key technical points of this patent are wound status recognition based on high-precision wound images, assisted positioning based on augmented reality technology, wound healing environment assessment based on multimodal sensor information, and nursing strategy information matching and pushing based on patient information retrieval. However, the patent has several shortcomings: (1) It assumes that the patient’s health information is extracted from the electronic health record, but for first-time patients or patients with new wounds, nurses may still need to manually enter the relevant information before it can be extracted by the decision system; (2) It does not provide sufficient information on the methods and sources of the “preset nursing plan database”, only stating that it may include nursing measures such as disinfection methods, dressing selection, and drug use. The generated strategy information can be a detailed nursing guide document, but it does not clarify how to formulate nursing plans, the level of detail of the provided plans, or whether there are ways to update or improve them. This part is the point of protection that the technology focuses on; (3) Based on the patient information retrieval, the hospital or pharmacy address is pushed according to the specific treatment strategy. These are different wound management methods. It does not further clarify which wounds require emergency hospital treatment and which wounds can be treated by themselves. It also does not clarify the method of pushing different nursing strategies to doctors, nurses and patients in different scenarios; (4) It does not clarify how to determine whether the current treatment and treatment are effective. This part of the judgment usually requires professional judgment. Therefore, there are many uncertainties and risks in clinical application.

[0005] Patent CN112309554A discloses a cross-team management information system for chronic wounds based on shared decision-making. The key technical point of this patent is the separation of wound strategies and decisions between doctors and nurses. Doctors provide different wound management strategies, patients provide their preferences for different strategies, and wound nurses make the final decisions. Furthermore, different doctors may offer different solutions, requiring nurses to synthesize multiple options before making a final decision. This patent emphasizes multi-party communication among medical staff, nurses, and patients, but it does not explain how to ensure the scientific validity of treatment plans. The overall process relies heavily on physician experience to develop plans, and nurses' experience to compare and select options. Moreover, the scope of actions that doctors and nurses can perform during wound treatment differs. For example, debridement, negative pressure suction, and antibiotic administration can mostly only be performed by doctors; nurses focus on wound assessment, non-sharp instrument debridement, wound cleaning and disinfection, and dressing changes. Therefore, treatment plans provided by doctors are not entirely applicable to nurses. The overall wound management approach of this patent leans towards the medical perspective, but it does not further explain how to guide nurses in their wound care work.

[0006] Patent CN11523506A discloses an AI-powered wound assessment and management system, primarily used for patient information management, wound area calculation and assessment, and ultimately generating patient wound reports. The wound assessment includes image-based wound area calculation and tissue classification, and wound odor assessment based on an electronic nose. Other patient information requires doctors to input, modify, and delete it. The system lacks sufficient automation in acquiring information throughout the wound patient's treatment process and also fails to provide treatment suggestions based on different wound characteristics.

[0007] Patent CN118037698A discloses an intelligent wound monitoring and management system for field operations. Its focus is on identifying wound type, area, and depth based on convolutional neural networks, while simultaneously monitoring the healing process. It improves recognition accuracy in field environments through transfer learning; collects multi-parameter wound data based on intelligent dressings and combines machine learning methods to assess infection risk and healing status; and dynamically adjusts the drug release and dressing change frequency of the intelligent dressing based on wound monitoring data and treatment feedback using a dynamic treatment adjustment algorithm. It's important to note that this system is only suitable for emergency treatment in resource-constrained field environments. Its purpose is to reduce reliance on medical personnel, enabling wound monitoring and limited treatment options (drug release and dressing change frequency) based on intelligent dressings in the absence of medical personnel. However, it does not cover many details involved in wound care, such as wound cleaning solution selection, disinfectants, debridement methods, etiological treatment, dressing selection, and health guidance. Therefore, it represents a different wound management approach from decision support systems that provide treatment recommendations to medical personnel.

[0008] Patent CN115426939A discloses a machine learning method for wound assessment, healing prediction, and treatment. This method uses wound images combined with patient information to predict the likelihood of wound healing within 30 days and selects between standard wound treatment or advanced therapy based on this likelihood. However, the specific method of standard or advanced treatment is determined by the medical staff themselves and is not recommended by the system.

[0009] In summary, the aforementioned patents still fail to fully address the practical problems faced in the field of wound care decision support. Therefore, a wound care decision support system based on multi-source data is proposed to address some of the shortcomings of the aforementioned patents. Summary of the Invention

[0010] The purpose of this invention is to provide a wound care decision support system and method based on multi-source data to solve the above-mentioned problems in the prior art.

[0011] This invention is achieved through the following technical solution:

[0012] In a first aspect, the present invention provides a wound care decision support method based on multi-source data, comprising:

[0013] The basic data of the current patient is obtained from the data source, and the basic data is preprocessed to generate the first standard database of the current patient;

[0014] Set up a logical mapping, obtain different decision-making strategies based on the logical mapping, and establish a reward function based on several basic parameter variables of patients and the first standard database during the time period of using different decision-making strategies.

[0015] Decision-making strategies are ranked from highest to lowest based on their reward function scores, and this ranking serves as the priority recommendation order for measures in the decision support system.

[0016] Preferably, the step of establishing the reward function includes:

[0017] Several wound care measures are obtained from the decision-making strategy, and basic data on the wound conditions corresponding to the measures are obtained and preprocessed to generate several second standard databases.

[0018] A reward function is established, and based on the first standard database and several second standard databases, the matching is performed using the reward function to obtain several first matching degree values;

[0019] Obtain the historical usage frequency of several measures, generate the calculation weight of each measure based on the historical usage frequency, correct the first matching degree value, and obtain several second matching degree values;

[0020] Sort the second matching degree values ​​from largest to smallest, and output the priority recommendation order of the measures corresponding to each second matching degree value.

[0021] Preferably, the preprocessing of the basic data includes:

[0022] Image recognition is performed on the image data to obtain feature images with wound characteristics;

[0023] Speech recognition is performed on the audio data of the consultation to obtain the text data of digital text. Keyword samples are set to extract keywords from the text data.

[0024] The feature images, text data, oxygen, oxygen saturation data, local tissue pressure sensing detection data, and vital sign data are stored in a first standard database or a second standard database.

[0025] Preferably, the image recognition of the image data includes:

[0026] When acquiring image data, a white graduated measuring ruler is set and placed at a preset distance from the wound as a reference for color correction and size measurement, thereby acquiring the image data;

[0027] Based on the unified correction of image brightness, color temperature, and hue using a measuring ruler, Gaussian filtering is applied, and edge detection is performed on the Gaussian-filtered image to delineate the edge range of the wound;

[0028] Calculate the wound area, longest diameter, shortest diameter, and diameter based on the number of pixels within the wound edge area;

[0029] The image data is classified into tissue morphology based on color, and the RGB values ​​are identified and classified into red, yellow and black tissues. Edge detection is performed on different image data and the corresponding area is calculated to obtain the proportion of different tissues in the wound bed.

[0030] Preferably, the method further includes calculating the similarity between first image data from a first standard database and second image data from a second standard database;

[0031] Wavelet transform is performed on the first image data and the second image data. SURF features of the low-frequency image are extracted from the wavelet-transformed first image data and the second image data to obtain the first feature point set of the first image data and the second feature point set of the second image data.

[0032] Cross-matching is performed on the first feature point set and the second feature point set, and the RANSAC algorithm is used to remove mismatched feature points to obtain the target feature points that are successfully matched in the first feature point set.

[0033] The similarity of the current second image data is obtained by using the number of target feature points and the number of feature points in the first feature point set.

[0034] Preferably, the cross-matching of the first feature point set and the second feature point set includes:

[0035] Calculate the Euclidean distance between all feature points in the first and second feature point sets in both directions;

[0036] A Euclidean distance threshold is set. If the Euclidean distance between a feature point in the first feature point set and a feature point in the second feature point set is less than the Euclidean distance threshold, then the feature point in the first feature point set is marked as the target feature point.

[0037] Preferably, the reward function;

[0038]

[0039] In the formula, Let i be the first matching degree value of the i-th second database. The number of target feature points to match the first image data. The number of feature points in the first feature point set. This represents the number of identical keywords in the text data between the first standard database and the i-th second standard database. This represents the total number of keywords in the text data of the first standard database. The oxygen saturation values ​​are from the first standard database. Let i be the oxygen saturation value of the i-th second standard database. The values ​​are local tissue pressure sensor readings from the first standard database. For the i-th local tissue pressure sensor detection value in the second standard database, The blood pressure values ​​are from the first standard database. Let be the blood pressure value of the i-th second standard database.

[0040] Preferably, the correction of the first matching degree value includes:

[0041]

[0042] In the formula, This is the second matching degree value. The number of times all measures are used. The frequency at which the i-th incorrect number is used.

[0043] Preferably, it further includes setting a logical mapping, the logical mapping including:

[0044] Based on the common wound types, wounds are classified, key interventions in wound care are identified, and clinical problems are set.

[0045] Based on each clinical problem, corresponding free terms and subject terms are selected to generate recommendations, which include the priority order of the corresponding measures.

[0046] The recommendations are decomposed using GEM to map free text and heterogeneous knowledge into a computer-understandable structured language;

[0047] After selecting specific recommendation text, GEM cut is used to identify and label the knowledge components of the text, setting them as knowledge maps in IF-THEN format, which serve as different decision-making strategies.

[0048] Secondly, the present invention also provides a wound care decision support system based on multi-source data, comprising:

[0049] The data source module is configured to obtain the basic data of the current patient from the data source, and preprocess the basic data to generate the first standard database of the current patient;

[0050] The data acquisition module is configured to obtain several wound care measures from a literature database, acquire basic data on the wound condition corresponding to the measures, perform preprocessing, and generate several second standard databases.

[0051] The measure evaluation module is configured to establish a reward function, match a first standard database and several second standard databases using the reward function to obtain several first matching degree values; obtain the historical usage frequency of several measures, generate the calculation weight of each measure based on the historical usage frequency, and correct the first matching degree values ​​to obtain several second matching degree values;

[0052] The functional application module is configured to sort the second matching degree values ​​from largest to smallest and output the priority recommendation order of the measures corresponding to each second matching degree value.

[0053] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0054] The method provided by this invention obtains basic patient data from a data source, preprocesses the basic data to generate a first standard database of the current patient, sets up logical mappings, obtains different decision-making strategies based on the logical mappings, and establishes a reward function for several basic parameter variables of the patient within the time period of using different decision-making strategies. The decision-making strategies are ranked from high to low according to the reward function scores, which serves as the priority recommendation order for the measures of the decision support system, providing medical staff with information to assist in judgment and improving the efficiency of medical staff in their work. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0057] Figure 2 This is the control flowchart of the present invention. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0059] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. The naming or numbering of steps in this application does not imply that the steps in the method flow must be executed in the chronological / logical order indicated by the naming or numbering. The execution order of named or numbered process steps can be changed according to the desired technical objective, as long as the same or similar technical effect is achieved.

[0060] The independently described modules or sub-modules may or may not be physically separated; they may be implemented in software or hardware, and some modules or sub-modules may be implemented in software, with the processor calling the software to implement the function of these modules or sub-modules, while other modules or sub-modules may be implemented in hardware, such as through hardware circuits. Furthermore, some or all of the modules can be selected to achieve the purpose of this application's solution according to actual needs.

[0061] Please refer to Figures 1-2 This invention provides a wound care decision support method based on multi-source data, comprising:

[0062] S101: Obtain the basic data of the current patient from the data source, and preprocess the basic data to generate the first standard database of the current patient;

[0063] The decision support system may involve various data sources during the assessment, planning, execution and evaluation phases, including but not limited to comprehensive platforms that integrate historical patient data such as HIS, EMR, LIS, RIS, PACS, NIS, wound management systems and clinical patient data centers, as the basis for subsequent data collection and acquisition of patient data.

[0064] Various data may be involved during a patient's visit, including but not limited to the patient's current visit details, historical medical records, hospitalization records, surgical records, nursing records, examination and test results, medical orders, wound images obtained during the current visit, and audio recordings of the consultation. Wound images are acquired using an image acquisition terminal that interoperates with the HIS system, while audio recordings are acquired using an audio acquisition terminal that interoperates with the outpatient system. Furthermore, other terminals for collecting wound-related information can be expanded, such as those for odor recognition, temperature sensing and humidity / oxygen saturation measurement, electrophysiological testing, and local tissue pressure sensing.

[0065] S102: Obtain several wound care measures from the literature database, obtain basic data on the wound conditions corresponding to the measures, perform preprocessing, and generate several second standard databases;

[0066] The literature database includes an evidence-based measures repository. This repository primarily consists of high-quality research evidence in the field of wound care, including guidelines, consensus statements, best practices, systematic reviews, and randomized controlled trials (RCTs). Expert-developed rules provide reliable, empirical opinions from expert teams for various situations lacking evidence-based medicine but common in clinical practice. Nurse feedback, derived from actual use, is used to rank recommended measures based on the frequency of selection by wound care nurses and their skill level (level and seniority). The evidence-based measures and expert rules are written with specific decision conditions using IF-THEN statements, while nurse feedback automatically updates the ranking of recommended measures through an adaptive reinforcement learning algorithm. Furthermore, the corresponding rules and priority order in the logical mapping module can be regularly checked and updated by designated personnel to ensure the scientific rigor and reliability of the content.

[0067] S103: Establish a reward function, and match the first standard database and several second standard databases using the reward function to obtain several first matching degree values;

[0068] S104: Obtain the historical usage frequency of several measures, generate the calculation weight of each measure based on the historical usage frequency, correct the first matching degree value, and obtain several second matching degree values.

[0069] The method provided by this invention obtains basic patient data from a data source, preprocesses the basic data to generate a first standard database of the current patient, sets up logical mappings, obtains different decision-making strategies based on the logical mappings, and establishes a reward function for several basic parameter variables of the patient within the time period of using different decision-making strategies. The decision-making strategies are ranked from high to low according to the reward function scores, which serves as the priority recommendation order for the measures of the decision support system, providing medical staff with information to assist in judgment and improving the efficiency of medical staff in their work.

[0070] In practice, the main functions include patient assessment, assessment record generation, wound assessment-based nursing intervention recommendations, development and implementation of nursing order plans, regular reminders of nursing orders, wound outcome evaluation, quality control monitoring, and nurse feedback.

[0071] The interactive interface used by nursing staff follows the nursing procedure framework of assessment-planning-execution-evaluation to complete the corresponding nursing tasks.

[0072] In one exemplary embodiment of the present invention, the basic data includes wound image data, consultation audio data, oxygen saturation data, local tissue pressure sensor detection data, and vital sign data.

[0073] Specifically, the preprocessing of the basic data includes:

[0074] Image recognition is performed on the image data to obtain feature images with wound characteristics;

[0075] Speech recognition is performed on the audio data of the consultation to obtain the text data of digital text. Keyword samples are set to extract keywords from the text data. The keyword samples are non-repeating keywords. In subsequent statistics, the number of repeated occurrences is not counted. If the keywords of two texts appear at the same time, the keyword is marked as the target keyword.

[0076] The feature images, text data, oxygen saturation data, local tissue pressure sensing detection data, and vital sign data are stored in a first standard database or a second standard database.

[0077] Image recognition of image data includes:

[0078] All wound images were captured using a standardized white graduated ruler, placed 3-5 cm from the wound edge as a reference for color correction and size measurement. After capturing the images, brightness, color temperature, and hue were uniformly corrected based on the ruler, followed by Gaussian filtering. Edge detection was then performed on the Gaussian-filtered images to delineate the wound area. Subsequently, the wound area, longest diameter, shortest diameter, and diameter were calculated based on the number of pixels within the wound edge area. Depth was measured indirectly using sterile cotton swabs to detect wound depth and photographs for identification; the results are expressed in cm. Simultaneously, the wound bed tissue was classified based on color, identifying RGB values ​​and categorizing them as red, yellow, and black tissues. Edge detection was performed on each tissue type, and the corresponding area was calculated to obtain the proportion of each tissue in the wound bed.

[0079] Secondly, audio recognition and conversion: During patient visits or hospitalizations, real-time audio streams are recorded using a microphone array. Beamforming technology combined with a deep learning voiceprint recognition model is used to distinguish multiple voice tracks of medical staff and patients. Semantic segmentation automatically cuts continuous audio into meaningful speech fragments. Combined with medical terminology, high-precision speech recognition is performed to extract medical named entities, attributes, relationships, and timelines. Based on the patient's wound medical record template, the identified entities, attributes, relationships, and timelines are accurately filled into the corresponding fields of the template. Simultaneously, a sequence-to-sequence (Seq2Seq) natural language summarization method is used, combined with the extracted structured information and dialogue context, to generate medical record narrative text such as medical record summaries, present medical history, and nursing records.

[0080] After patient data acquisition, further processing is performed based on whether the data has been standardized. For patient data intended for subsequent modeling, cleaning, encoding, and / or normalization are performed according to model requirements. Image or audio data obtained from external sensors are identified and transformed according to pre-defined models, including image segmentation / amplification / normalization, and audio-to-text conversion. Subsequently, natural language processing is performed on the transcribed audio text, identifying and splitting text words, mapping each word to medical terminology, and finally summarizing it with other data to form a standardized wound patient database.

[0081] An exemplary embodiment of the present invention further includes calculating the similarity between first image data from a first standard database and second image data from a second standard database;

[0082] Wavelet transform is performed on the first image data and the second image data. SURF features of the low-frequency image are extracted from the wavelet-transformed first image data and the second image data to obtain the first feature point set of the first image data and the second feature point set of the second image data.

[0083] Cross-matching is performed on the first feature point set and the second feature point set, and the RANSAC algorithm is used to remove mismatched feature points to obtain the target feature points that are successfully matched in the first feature point set.

[0084] The similarity of the current second image data is obtained by using the number of target feature points and the number of feature points in the first feature point set.

[0085] Specifically, the cross-matching of the first feature point set and the second feature point set includes:

[0086] Calculate the Euclidean distance between all feature points in the first and second feature point sets in both directions;

[0087] A Euclidean distance threshold is set. If the Euclidean distance between a feature point in the first feature point set and a feature point in the second feature point set is less than the Euclidean distance threshold, then the feature point in the first feature point set is marked as the target feature point.

[0088] Among them, the RANSAC algorithm is often used to remove mismatched feature points in feature image matching.

[0089] Assuming the dataset consists of n pairs of matching points, set 3 pairs of matching point data.

[0090] Step 1: Randomly select 3 pairs of non-collinear initial matching points from the dataset, and use the least squares method to obtain their initial transformation matrix;

[0091] Step 2: For the remaining n-3 pairs of matching points, calculate the distance between each point and the model. If the distance is less than the error threshold, the feature point is considered an interior point of the matching pair; otherwise, it is considered an exterior point. Count the number of interior points.

[0092] Step 3: Repeat Step 1 and Step 2. When the number of interior points no longer changes and is greater than the preset threshold, the corresponding set of interior points is the maximum interior point domain, and the transformation matrix at this time is the optimal model matrix.

[0093] In one exemplary embodiment of the present invention, the reward function;

[0094]

[0095] In the formula, Let i be the first matching degree value of the i-th second database. The number of target feature points to match the first image data. The number of feature points in the first feature point set. This represents the number of identical keywords in the text data between the first standard database and the i-th second standard database. This represents the total number of keywords in the text data of the first standard database. The oxygen saturation values ​​are from the first standard database. Let i be the oxygen saturation value of the i-th second standard database. The values ​​are local tissue pressure sensor readings from the first standard database. For the i-th local tissue pressure sensor detection value in the second standard database, The blood pressure values ​​are from the first standard database. Let be the blood pressure value of the i-th second standard database.

[0096] Specifically, the correction of the first matching degree value includes:

[0097]

[0098] In the formula, This is the second matching degree value. The number of times all measures are used. The frequency at which the i-th incorrect number is used.

[0099] An exemplary embodiment of the present invention further includes setting a logical mapping, which includes: classifying wounds based on common wound types and obtaining key interventions in wound care, setting clinical questions; based on each clinical question, filtering corresponding free terms and subject terms to generate recommendations, the recommendations including a priority order of corresponding measures. The recommendations are decomposed using GEM to map free text and heterogeneous knowledge to a computer-understandable structured language; after selecting specific recommendation text, GEM cut is used to identify and label the knowledge components of the text, setting them as IF-THEN format knowledge mappings as different decision-making strategies; the patient's pain improvement, quality of life score, and healing speed during different decision-making strategy time periods are used as reward functions; the decision-making strategies are ranked from high to low according to the reward function scores, serving as the priority order of measures in the decision support system.

[0100] Specifically, the logical mapping includes an evidence-based practice library, expert-developed rules, and priority of measures based on nurse feedback. The evidence-based practice library mainly consists of high-quality research evidence in the field of wound care from the literature, including guidelines, consensus statements, best practices, systematic reviews, and RCTs. The acquisition process follows the steps of formulating a clinical question—PICO decomposition—systematic literature search—quality assessment—evidence synthesis—formulation of recommendations. First, based on common wound types, wounds are categorized into 13 major categories and 77 subcategories, including surgical and laparoscopic wounds, pressure injuries, incontinence dermatitis, burns, mastitis, traumatic wounds, vascular ulcers, diabetic foot ulcers, radiation injuries, immune ulcers, infected wounds, dermatological wounds, and other wounds. Combining this with key interventions in wound care—including wound monitoring, wound treatment, consultation and communication, and health education—clinical questions are posed, such as: "What indicators should be assessed and monitored during the care of a certain type of wound?", "What treatment should be given to a certain type of wound based on its characteristics?", "In what situations is it necessary to consult with other specialists?", and "What kind of health education should be provided to patients with a certain type of wound?". Subsequently, for each clinical question, based on the format of population (P) - intervention / exposure (I) - control (C) - outcome (O), corresponding free terms and subject terms are selected, and searches are conducted in PubMed, Embase, Web of Science, CINAHL, and The Cochrane. We obtained relevant literature from the Library database and the website of the Wound Medicine Association, used the JBI Pre-Grading Standard for Evidence to grade the quality of evidence, and used a meta-quantitative merging method to obtain the final conclusion for those with conflicting results from different literatures, and formed recommendations in the form of PICO.

[0101] The obtained recommendations are decomposed using a Guideline Elements Model (GEM) to map free text and heterogeneous knowledge into a computer-understandable structured language. After selecting specific recommendation text, the GEM cut function identifies and labels the knowledge components of the text, setting them as IF-THEN format knowledge maps. For example:

[0102] Original guideline: Pressure injury risk assessment should be performed on hospitalized patients aged ≥65 years.

[0103] After conversion, the content is broken down into:

[0104] [Action Description 1] Pressure Injury Risk Assessment

[0105] [Trigger 1] Age ≥ 65 years old

[0106] [Trigger 2] Inpatients

[0107] IF [Trigger 1] AND [Trigger 2] = TRUE, [Action Description 1]

[0108] Each action and trigger is assigned a unique code for backend management.

[0109] Once accessed, the decision support system presents the information to nurses in the following format:

[0110] For both [Trigger 1] and [Trigger 2], [Action Description 1] should be executed.

[0111] The triggers here include patient information, wound characteristics, and other data.

[0112] Many of these triggers may not be easily summarized by precise numerical values. For example, the amount of wound exudate (small, medium, or large); the wound area (small, medium, or large); the wound color (red, yellow, or black); the wound's status as uninfected, suspected, or confirmed infected; and the degree of wound odor. For such cases, it is necessary to define fuzzy sets and membership functions, clarify their approximate ranges, and construct fuzzy rules based on guidelines and expert opinions.

[0113] For example:

[0114] [Action Description 1] Bacterial Culture (Priority: High)

[0115] [Trigger 1] Wound color: Yellow

[0116] [Trigger 2] Exudate: Large amount

[0117] [Trigger 3] Pain: Severe

[0118] IF([Trigger Item 1] AND [Trigger Item 2]) OR([Trigger Item 1] AND [Trigger Item 3]), THEN[Action Description 1].

[0119] Based on the actual situation of the wound, signs of infection, exudate, and pain can all be converted into membership vectors. For example, "exudate = 15ml / d" can be converted into the corresponding membership (small amount: 0.3, moderate: 0.7, large amount: 0), and wound color (red 15%, yellow 75%, black 10%) can be converted into the corresponding membership (red 0.15, yellow 0.75, black 0.1). Then, according to the rules, the action triggering priority is determined by the value with the highest membership.

[0120] (5) Ranking of recommended measures

[0121] In addition to expert-based rules, nurse feedback is used to add and prioritize recommended strategies. During the application of the wound decision support system, nurses' choices of recommended measures and patient outcomes are continuously collected. The patient's pain improvement, quality of life score, and healing speed over different decision-making strategy periods are used as reward functions, designed as: Reward = Wound area reduction / time norm + Pain reduction norm + Medical cost norm + Quality of life score norm (all variables are linearly summed after normalization). Strategies used by nurses are ranked from highest to lowest based on reward function scores, serving as the priority order for subsequent recommendations when nurses use the decision support system; meanwhile, recommended measures that nurses do not use are defaulted to the bottom of the recommendation order. Furthermore, nurses can add "other measures" when selecting wound care measures based on their personal experience. These added measures are effective for the patient at the time of application and are added to the alternative measure library in the decision support system's knowledge base, where they are periodically checked by designated personnel to confirm whether corresponding decision rules have been added.

[0122] In the above modules, specifically regarding the work of wound care nurses, the functions and processes involved are as follows:

[0123] Patient Comprehensive Assessment: After a wound patient is admitted, the nurse can directly extract the patient's existing data from the existing data source and synchronize it to the decision support system's assessment module. Other key wound-related assessments (wound history, injury mechanism, patient comorbidities, treatment history, wound condition, etc.) are conducted through on-site consultation and physical examination. During the consultation, voice recordings are used to simultaneously upload the consultation data and convert it into text, enabling automatic completion of the patient's wound assessment data. Local wound examination information can be collected through image acquisition, skin temperature monitoring, etc., simultaneously identifying wound size, depth, wound stage, tissue type, color, exudate, wound temperature, etc., and filling this information into the assessment system. Subsequently, the nurse can view the current pre-filled wound assessment form and modify, submit, or delete it accordingly. The aforementioned wound image information is processed using a pre-trained machine learning model. Specific model construction methods include, but are not limited to, edge detection and convolutional neural networks, used for quantitative calculation and qualitative classification of wound parameters.

[0124] Nursing plan development: Based on the patient's characteristics data obtained from preliminary assessments and other sources, and considering the patient's overall condition and wound condition, a comprehensive decision is made. Using the pre-defined decision logic of the logic mapping module, recommended wound care measures are generated corresponding to the specific assessment content. The content of the nursing measures follows the TIME (Time of Action) principle of wound management, including wound cleaning, disinfection, debridement, dressing application, other auxiliary treatments, and health education. The generated nursing measures can be selected, added, or deleted by wound care personnel. After a wound care measure is selected, the frequency of its execution must also be chosen to generate a nursing plan related to the patient's wound. The nurse selects the nursing measures and frequency and submits the plan, generating a nursing plan containing specific nursing measures and frequencies. Subsequently, based on the nursing frequency in the nursing plan, the system generates regular reminders on the user interface. For example, after generating a "wound dressing change qd" nursing order, the system will remind the nurse daily on the user interface according to the scheduled time for that patient's wound dressing change.

[0125] Nursing intervention implementation: After the nursing intervention is implemented, it is automatically executed in the system and a nursing record is generated.

[0126] In another embodiment, the following can also be configured:

[0127] Outcome evaluation: For patients who return for follow-up visits, the evaluation can be based on the actual effect of the wound follow-up visit to determine whether the wound recovery is good. For wounds that have completely healed, the "cured" option can be checked; during the orderly repair process of the wound, the "improved" or "no significant change" option can be checked; if there is malignant transformation of the wound or delayed healing, the "worsened" option can be checked.

[0128] Nurse feedback is mainly reflected in the modification of nursing assessments, the selection of nursing interventions, and the evaluation of wound outcomes. These results will be fed back to the logical mapping module of the decision support system, which will continuously adjust the accuracy of nursing assessments and the priority of modifying nursing interventions based on the frequency of specific nursing intervention selections and the revision of nursing assessment content.

[0129] Quality control monitoring: This mainly includes monitoring and extracting wound quality control indicators to provide a reference for understanding the basic situation, treatment information, and evaluating treatment effectiveness of wound patients. For individual patients, wound nurses can use the quality control monitoring function to view the patient's treatment measures and wound assessment information at different time periods, and plot the wound healing progress curve to determine whether the patient's healing progress is normal. From a batch management perspective, wound nurses can use the quality control monitoring function to understand the total number of patients, their basic characteristics, wound conditions, and treatment effects, and further analyze the effectiveness of different treatment measures to provide suggestions for revising the current nursing intervention knowledge base.

[0130] Secondly, the present invention also provides a wound care decision support system based on multi-source data, comprising:

[0131] The data source module is configured to obtain the basic data of the current patient from the data source, and preprocess the basic data to generate the first standard database of the current patient;

[0132] The data acquisition module is configured to obtain several wound care measures from a literature database, acquire basic data on the wound condition corresponding to the measures, perform preprocessing, and generate several second standard databases.

[0133] The measure evaluation module is configured to establish a reward function, match a first standard database and several second standard databases using the reward function to obtain several first matching degree values; obtain the historical usage frequency of several measures, generate the calculation weight of each measure based on the historical usage frequency, and correct the first matching degree values ​​to obtain several second matching degree values;

[0134] The functional application module is configured to sort the second matching degree values ​​from largest to smallest and output the priority recommendation order of the measures corresponding to each second matching degree value.

[0135] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0136] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. This computer software product, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-source data based wound care decision support method, characterized in that, The method comprises the following steps: obtain the basic data of the current patient from the data source, and preprocess the basic data to generate a first standard database of the current patient; set a logical mapping, obtain different decision strategies based on the logical mapping, and establish a reward function using the first standard database and a plurality of basic parameter variables of the patient in a time period using different decision strategies; sort the decision strategies from high to low according to the reward function scores, and use the sorted order as the priority recommendation order of the measures of the decision support system; the sorting of the decision strategies from high to low according to the reward function scores comprises: obtain a plurality of measures for wound care from the decision strategies, obtain the basic data of the wound condition corresponding to the measures and preprocess the basic data to generate a plurality of second standard databases; establish a reward function, match the first standard database and the plurality of second standard databases through the reward function, and obtain a plurality of first matching degree values; obtain the historical usage frequency of the plurality of measures, generate a calculation weight for each measure through the historical usage frequency, correct the first matching degree values, and obtain a plurality of second matching degree values; sort the second matching degree values from large to small, and output the priority recommendation order of the measures corresponding to each second matching degree value; the reward function comprises: In the formula, is the first matching degree value of the ith second database, is the number of target feature points matched with the first image data, is the number of feature points in the first feature point set, is the number of keywords in the text data of the first standard database that are the same as the text data of the ith second standard database, is the total number of keywords in the text data of the first standard database, is the oxygen saturation value of the first standard database, is the oxygen saturation value of the ith second standard database, is the local tissue pressure sensing detection value of the first standard database, is the local tissue pressure sensing detection value of the ith second standard database, is the blood pressure value of the first standard database, is the blood pressure value of the ith second standard database; the correction of the first matching degree values comprises: wherein is a second matching degree value, is the number of times all measures are used, is the frequency of the ith mistake number used; the selection of the recommended measures and the patient outcomes are continuously collected, the pain improvement, the quality of life score and the healing speed of the patient in the time period using different decision strategies are used as the reward function, and the reward function is designed as Reward= wound area reduction amount / time norm+ pain reduction norm+ medical cost norm+ quality of life score norm; according to the reward function score, the strategy used by the nurse is sorted from high to low, which is used as the priority order of the recommended measures when the nurse uses the decision support system subsequently; at the same time, the recommended measures not used by the nurse are defaulted to the last position of the recommended order; when collecting image data, a white scale ruler is set at a preset distance from the wound, which is used as a reference for color correction and size measurement, so as to obtain the image data; based on the scale ruler, the image brightness, color temperature and hue are uniformly corrected, Gaussian filtering is performed, and edge detection is performed on the Gaussian filtered image to circumscribe the edge range of the wound; a fuzzy set and a membership function are defined, fuzzy rules are constructed, and the value with the highest membership degree is used to determine the priority of action triggering.

2. A multi-source data based wound care decision support method according to claim 1, characterized in that, the preprocessing of the basic data comprises: image recognition is performed on the image data to obtain a feature image with wound characteristics; voice recognition is performed on the audio data of the inquiry to obtain text data of digital text, key word samples are set, and the key words in the text data are extracted; the feature image, the text data, the oxygen saturation data, the local tissue pressure sensing detection data and the vital sign data are stored as the first standard database or the second standard database.

3. A multi-source data based wound care decision support method according to claim 2, characterized in that, the image recognition of the image data comprises: based on the number of pixels in the wound edge range, the wound area, the longest diameter, the shortest diameter and the diameter are calculated; The image data is classified based on color, RGB values are identified and classified as red, yellow, black tissue, and edge detection is performed on different image data to calculate the corresponding area to obtain the proportion of different tissues in the wound bed.

4. A multi-source data based wound care decision support method according to claim 2, characterized in that, Also includes calculating the similarity of the first image data of the first standard database and the second image data of the second standard database; Wavelet transform is performed on the first image data and the second image data, and SURF features of low-frequency images of the wavelet-transformed first image data and second image data are extracted to obtain a first feature point set of the first image data and a second feature point set of the second image data; Cross matching is performed on the first feature point set and the second feature point set, and the RANSAC algorithm is used to remove mis-matched feature points to obtain target feature points in the first feature point set that are successfully matched; The similarity of the current second image data is obtained by the number of target feature points and the number of feature points in the first feature point set.

5. A multi-source data based wound care decision support method according to claim 4, characterized in that, The cross matching of the first feature point set and the second feature point set includes: Bidirectional calculation of the Euclidean distance of all feature points in the first feature point set and the second feature point set; And set the Euclidean distance threshold, if the Euclidean distance between a feature point in the first feature point set and a feature point in the second feature point set is less than the Euclidean distance threshold, mark the feature point in the first feature point set as a target feature point.

6. A multi-source data based wound care decision support method according to claim 4, characterized in that, Also includes setting a logical mapping, the logical mapping includes: Based on the current common wound type, classify the wound, obtain the main intervention measures in wound care, and set the clinical problems; Based on each of the clinical problems, filter the corresponding free words and subject headings, generate the recommended opinions, and the recommended opinions include the priority recommendation order of the corresponding measures; The recommended opinions are disassembled using GEM to realize the mapping of free text and heterogeneous knowledge to computer-understandable structured language; After selecting a specific recommended opinion text, identify and label the knowledge components of the text using GEM cut, set the knowledge mapping in IF-THEN format as different decision strategies.

7. A multi-source data based wound care decision support system for performing a multi-source data based wound care decision support method according to any one of claims 1-6, characterized by It includes: A data source module configured to obtain basic data of a current patient from a data source and preprocess the basic data to generate a first standard database of the current patient; A data acquisition module configured to obtain several measures for wound care from a literature database, and obtain the basic data of the wound condition corresponding to the measures and preprocess the basic data to generate several second standard databases; A measure evaluation module configured to establish a reward function, match the first standard database and the several second standard databases based on the reward function to obtain several first matching degree values; obtain the historical usage frequency of the several measures, generate a calculation weight for each measure based on the historical usage frequency, and correct the first matching degree values to obtain several second matching degree values; A function application module configured to sort the second matching degree values from large to small, and output the priority recommendation order of the measures corresponding to each second matching degree value.

Images