Central vein vascular access complication risk management method and related equipment

Abstract

The invention discloses a central vein vascular access complication risk management method and related equipment. The method comprises the steps that based on basic physiological information and edema related information of a patient, ideal pressure values of different edema parts of the patient are determined through a treatment pressure decision model, and the treatment pressure decision model is obtained through training according to historical clinical treatment data; displaying the ideal pressure values of the different edema parts of the patient; in the binding process, the actual pressure value between the bandage and the skin of the patient is obtained, and the actual pressure value is displayed. The problems that in the prior art, complication supervision of a patient after a central venous catheter is implanted through a peripheral vein and infusion port implantation stays in static inquiry, the response speed is low, emergency situations cannot be recognized in time, the data collection mode is single, and the evaluation accuracy is insufficient can be solved.

Description

Technical Field

[0001] This application relates to the field of smart healthcare, and more specifically, to a method and related equipment for managing the risk of complications in central venous access. Background Technology

[0002] Following the placement of a peripherally inserted central venous catheter (PICC) or port-a-catheter, patients typically record their daily symptoms on paper forms, which are then manually assessed by healthcare professionals during follow-up visits. This method suffers from incomplete records, inaccurate symptom descriptions, and lacks real-time monitoring capabilities. Hospitals also periodically inquire about patients' conditions by phone, using standardized questionnaires to collect information. This approach is inefficient, has limited coverage, and struggles to achieve personalized consultations. Summary of the Invention

[0003] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. The summary section of this invention is not intended to limit the key features and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0004] To address the shortcomings of current methods for monitoring complications in patients following peripherally inserted central venous catheter (PICC) and port-a-catheter implantation, which rely on static consultations, resulting in slow response times, inability to promptly identify critical situations, limited data collection methods, and insufficient assessment accuracy, this invention proposes a method for managing the risk of complications related to central venous access. This method includes:

[0005] Extract key symptom information associated with complications of the central venous access from patient user feedback messages regarding pre-existing issues;

[0006] The consultation path is adjusted in real time based on the key symptom information and the knowledge graph database of central venous access complications until the obtained key symptom information meets the risk prediction conditions of the knowledge graph database.

[0007] The obtained key symptom information is input into the XGBoost gradient boosting tree model to classify and output the risk prediction level, so as to trigger the associated treatment and management process based on the risk prediction level.

[0008] Optionally, the extraction of key symptom information associated with complications of the central venous access from the patient's feedback messages on the preceding questions includes:

[0009] The system calls the BERT-BiLSTM model to convert the words in the patient user's feedback message to the preceding question into context-related semantic vectors and models word order and dependency relations to generate a word vector sequence.

[0010] Key symptom information associated with complications of the central venous access can be obtained by label classification based on word vector sequences or by similarity matching with a symptom word vector library.

[0011] Optional, also includes:

[0012] A multi-head attention mechanism is used to calculate the relevance of all key symptom information extracted from patient user feedback messages on pre-existing questions.

[0013] The weights of each key symptom information are obtained through inner product weighting or attention matrix.

[0014] Optionally, the step of adjusting the consultation path in real time based on the key symptom information and the central venous access complication knowledge graph database until multiple key symptom information obtained meet the risk prediction conditions of the complication knowledge graph database includes:

[0015] Select the key symptom information with the highest weight and query the complication knowledge graph database for complications associated with the key symptom information with the highest weight;

[0016] The risk assessment of the complication requires matching at least one symptom condition group. Each symptom condition group includes at least one key symptom information. Meeting any of the symptom condition groups satisfies the risk prediction conditions for the complication.

[0017] Based on the symptom condition group, candidate questions are generated that are associated with the key symptom information with the highest weight;

[0018] Priority scores will be assigned based on the importance of the corresponding complication paths in the knowledge graph for each candidate question;

[0019] Based on prioritizing inquiries, the process continues until multiple key symptom information obtained meets the risk prediction conditions of the complication knowledge graph database.

[0020] Optionally, the extraction of key symptom information associated with complications of the central venous access from the patient's feedback messages on the preceding questions includes:

[0021] Based on the ResNet50 convolutional neural network, image features are extracted from the image information in the patient's feedback message to the previous question, and image feature vectors including at least one type of information such as color distribution, redness and swelling area outline and effusion trace are obtained.

[0022] Based on the symptom evolution information in the patient's feedback messages on the preceding issues, a time-stamped symptom sequence is constructed, the symptom development trajectory pattern is identified, and a time-series state feature vector is output.

[0023] The key symptom information is obtained by fusing the image feature vector, the time series state feature vector, and the text information feature vector in the feedback message.

[0024] Optional, also includes:

[0025] The silent symptom prediction sub-model is used to infer potential symptoms that are not actively mentioned by the patient in the current consultation context, so as to infer abnormal silent symptoms when no key symptoms are directly expressed.

[0026] Optional, also includes:

[0027] Construct a semantic feature representation of the current patient based on the acquired explicit symptom features;

[0028] By comparing with the pre-defined symptom co-occurrence probability distribution, identify missing symptoms that are not present in the current conversation but are highly related to existing symptoms;

[0029] The missing symptom is used as a mask position and input into the mask language model for probability prediction to obtain the significance score of the missing symptom in the current context.

[0030] When the significance score exceeds the set threshold, the missing symptom is marked as an abnormal silent symptom, and the abnormal silent symptom is added to subsequent consultation questions.

[0031] Secondly, the present invention also proposes a central venous access complication risk management device, comprising:

[0032] Extraction unit, used to extract key symptom information associated with complications of the central venous access from the patient user's feedback messages on the preceding questions;

[0033] The analysis unit is used to adjust the consultation path in real time based on the key symptom information and the knowledge graph database of central venous access complications until the obtained key symptom information meets the risk prediction conditions of the knowledge graph database.

[0034] The treatment unit is used to input the obtained multiple key symptom information into the XGBoost gradient boosting tree model to classify and output the risk prediction level, so as to trigger the associated treatment management process based on the risk prediction level.

[0035] Thirdly, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the central venous access complication risk management method as described in any of the first aspects above.

[0036] Fourthly, the present invention also proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the central venous access complication risk management method of any of the above claims in the first aspect.

[0037] In summary, the central venous access complication risk management method proposed in this application extracts key symptom information associated with central venous access complications from patient feedback on pre-emptive questions; adjusts the consultation path in real time based on the key symptom information and a central venous access complication knowledge graph database until multiple key symptom information meets the risk prediction conditions of the knowledge graph database; inputs the obtained multiple key symptom information into an XGBoost gradient boosting tree model to classify and output risk prediction levels, triggering associated treatment and management processes based on the risk prediction levels. This significantly improves real-time performance, eliminating reliance on manual intervention for consultation and assessment, allowing for immediate identification and response to symptom changes; enhances personalization, as the consultation path moves beyond a fixed template and is dynamically controlled based on the knowledge graph, ensuring complete acquisition of key information; and strengthens objectivity in assessment, as machine learning-based model judgment replaces subjective experience, resulting in greater repeatability and accuracy. Furthermore, this solution supports multimodal input in practical deployment, such as combining uploaded puncture site images for image recognition to assist in judging infection signs, further enhancing the system's ability to identify heterogeneous symptoms and truly achieving data-driven complication risk prediction and early intervention.

[0038] The method for managing the risk of complications from central venous access according to the present invention, and other advantages, objectives and features of the present invention will be apparent in part from the following description, and in part from what those skilled in the art will understand through study and practice of the invention. Attached Figure Description

[0039] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0040] Figure 1 A schematic flowchart of a central venous access complication risk management method provided in this application embodiment;

[0041] Figure 2 A schematic diagram of a central venous access complication risk management device provided in this application embodiment;

[0042] Figure 3 This is a schematic diagram of an electronic device for managing the risk of complications in a central venous access, provided as an embodiment of this application. Detailed Implementation

[0043] The terms "first," "second," "third," "fourth," etc. (if present) 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. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0044] To address the shortcomings of current complication monitoring for patients following peripherally inserted central venous catheter (PICC) and port-a-catheter implantation, which relies on static consultations, slow response times, inability to promptly identify critical situations, limited data collection methods, and insufficient assessment accuracy, please refer to [the relevant documentation / reference]. Figure 1 This is a schematic flowchart of a central venous access complication risk management method provided in an embodiment of this application, which may specifically include steps S110 to S130.

[0045] S110, extract key symptom information associated with complications of the central venous access from the patient user's feedback messages on the preceding issues.

[0046] S120, adjust the consultation path in real time based on the key symptom information and the knowledge graph database of central venous access complications until the obtained key symptom information meets the risk prediction conditions of the knowledge graph database.

[0047] S130, the obtained multiple key symptom information is input into the XGBoost gradient boosting tree model to classify and output the risk prediction level, so as to trigger the associated treatment and management process based on the risk prediction level.

[0048] For example, to improve the efficiency and accuracy of postoperative complication risk management after PICC catheter or port implantation, the system uses subjective feedback (including text, voice, or images) submitted daily by patients via mobile devices as entry information to automatically conduct intelligent consultations and risk assessments. Patients upload information such as their current discomfort symptoms, changes at the puncture site, and body temperature data daily via the mobile application after surgery. The system first performs text parsing processing on this input. Specifically, the system uses a semantic analysis model based on the Transformer architecture to perform deep semantic understanding and extract key symptom information from natural language descriptions. The model can uniformly map non-standardized expressions such as "arm is a little swollen," "puncture site is red and itchy," and "feels a little feverish today" into structured symptom labels, such as "arm swelling," "puncture site redness and swelling," and "fever." It can also identify supplementary features such as time and severity dimensions, thereby constructing a standardized symptom feature vector. Next, the system performs real-time matching and path reasoning on the identified symptom information based on a pre-constructed central venous access complication knowledge graph database. This knowledge graph is built upon a triplet (symptom-relationship-complication) framework, covering typical and atypical manifestations of common complications such as catheter-related infections, thrombosis, and catheter occlusion. The graph embeds the clinical sequence, dependencies, and probability weights between symptoms. For example, when the co-occurrence of "redness and swelling at the puncture site" and "elevated body temperature" is detected, the system triggers the infection risk path in the graph and further calculates the salience weights of these symptoms within the current graph context using an attention mechanism. This prioritizes guiding the patient to answer subsequent questions most influential on the risk assessment of the complication, such as "Is there purulent discharge?", "Has the area of ​​redness and swelling expanded?", and "Are there chills or other systemic symptoms?", thus dynamically advancing a personalized consultation path under system control. After collecting complete key symptom information, the system feeds these symptom features as input vectors into a pre-trained XGBoost gradient boosting tree model. This model, trained on large-sample clinical follow-up data (such as PICC patient cases with risk levels already labeled), possesses strong nonlinear modeling capabilities, effectively fusing symptom features from different dimensions and uncovering their combination patterns. For example, "redness, swelling, fever, and exudate" have a high-weight combination in the model, often indicating a high-risk level; while "mild aches and pains and no redness or swelling" are classified as low-risk. The model outputs multi-level classification labels, such as "low risk," "medium risk," "high risk," or "requires immediate medical attention." Based on the model output, the system automatically triggers subsequent management procedures, such as generating doctor-side reminders for medium- and high-risk patients, recommending in-person follow-up visits; directly pushing emergency room guidance to high-risk individuals; and allowing low-risk patients to continue with the daily monitoring process to avoid unnecessary intervention.

[0049] In summary, the central venous access complication risk management method provided in this application extracts key symptom information associated with central venous access complications from patient feedback messages regarding pre-emptive issues; adjusts the consultation path in real time based on the key symptom information and a central venous access complication knowledge graph database until multiple key symptom information meets the risk prediction conditions of the knowledge graph database; inputs the obtained multiple key symptom information into an XGBoost gradient boosting tree model to classify and output risk prediction levels, triggering associated treatment management processes based on the risk prediction levels. This significantly improves real-time performance, eliminating reliance on manual intervention for consultation and assessment, allowing for immediate identification and response to symptom changes; enhances personalization, as the consultation path is no longer a fixed template but dynamically controlled based on the knowledge graph, ensuring complete acquisition of key information; and strengthens objectivity in assessment, as machine learning-based model judgment replaces subjective experience, resulting in greater repeatability and accuracy. Furthermore, this solution supports multimodal input in practical deployment, such as combining uploaded puncture site images for image recognition to assist in judging infection signs, further enhancing the system's ability to identify heterogeneous symptoms and truly achieving data-driven complication risk prediction and early intervention.

[0050] In some examples, the extraction of key symptom information associated with complications of the central venous access from patient user feedback messages on pre-existing issues includes:

[0051] The system calls the BERT-BiLSTM model to convert the words in the patient user's feedback message to the preceding question into context-related semantic vectors and models word order and dependency relations to generate a word vector sequence.

[0052] Key symptom information associated with complications of the central venous access can be obtained by label classification based on word vector sequences or by similarity matching with a symptom word vector library.

[0053] For example, to accurately extract key symptom information related to central venous access complications from patients' natural language feedback, the system first processes the input text using a pre-trained and fine-tuned BERT-BiLSTM joint model. Specifically, during postoperative follow-up, patients answer pre-questions (such as "Please describe your discomfort today" or "What changes have occurred at the puncture site?") through a mobile consultation system. The system uses this text as input and first performs semantic encoding via the BERT model. BERT, as a deep bidirectional language model, can generate semantically rich contextual vector representations for each word in the input text based on global semantic information. Compared to traditional static word vectors (such as Word2Vec or GloVe), BERT vectors can dynamically reflect changes in word meaning under different contexts. For example, the semantic emphasis of "redness and swelling" differs between "redness and swelling at the puncture site" and "mild redness and swelling of the skin," and BERT can effectively capture this semantic difference. Subsequently, the system inputs the word vector sequence output by BERT into a BiLSTM (Bidirectional Long Short-Term Memory network) to model the order dependency and syntactic relationship of words in semantic expression. BiLSTM structures can learn the contribution of word order to semantics from both forward and backward directions, thereby better understanding the multi-layered meanings and potential symptom association structures in text. For example, for the sentence "The arm puncture site started to feel a little hot and sore yesterday," BiLSTM can effectively identify "hot" and "sore" as symptom information associated with the puncture site and retain the time feature "started yesterday," providing support for subsequent reasoning. After obtaining the complete word vector sequence, the system will perform one of two symptom recognition operations: (1) Using a sequence labeling model, such as the BiLSTM-CRF structure, to classify words using BIO tags and directly output entity information of types such as symptoms, location, and time; (2) Or, the context vector of each word or phrase will be used to calculate the cosine similarity with the pre-built "Central Venous Access Complication Symptom Word Vector Library," and the key symptoms that match the standard symptom words will be identified according to the set similarity threshold. This word vector library can be constructed by collecting data such as clinical guidelines, electronic medical records, and follow-up questionnaires, including high-frequency symptoms such as redness, swelling, fever, effusion, and pain at the puncture site and their semantic variations.

[0054] Understandably, through the aforementioned joint modeling and matching process, the system ultimately outputs a set of structured key symptom information, which serves as input for subsequent dynamic consultation path adjustments and risk assessment modeling. On one hand, the contextual modeling capability based on the BERT-BiLSTM model can significantly improve the accuracy of symptom recognition, especially when patients freely describe their symptoms with semantic ambiguity and word order variations, maintaining strong robustness. On the other hand, by introducing a symptom word vector library for semantic matching, the system possesses a certain degree of adaptability, continuously expanding its understanding of new expressions, making it suitable for patient language input from multiple regions and different cultural backgrounds, thus enhancing the system's versatility and scalability.

[0055] In some examples, it also includes:

[0056] A multi-head attention mechanism is used to calculate the relevance of all key symptom information extracted from patient user feedback messages on pre-existing questions.

[0057] The weights of each key symptom information are obtained through inner product weighting or attention matrix.

[0058] For example, to further enhance the system's decision-making ability in early complication identification and dynamic adjustment of the consultation path, this implementation also includes: introducing a multi-head attention mechanism to calculate the correlation of all extracted key symptom information, and thereby assessing the relative contribution of each key symptom to the current central venous access complication risk. The main goal of this step is to assign weights to different symptoms, enabling the system to prioritize high-weight symptoms to guide consultation, drive model inference, or serve as feature enhancement inputs, thereby improving the system's personalization and intelligence. In specific implementation, after completing the key symptom identification based on the BERT-BiLSTM model, the system obtains a set of structured symptom vector representations, denoted as {s1,s2,...,s...}. n}, each s i This represents a symptom word or phrase within its context as a vector. These vectors are input into a multi-head attention module to model the correlations between symptoms. The multi-head attention mechanism uses multiple parallel attention subspaces (i.e., multiple attention heads) to perform a linear transformation on the same set of inputs and compute the Query(Q), Key(K), and Value(V) matrices separately. Each attention head computes an attention score:

[0059]

[0060] The semantic similarity between symptoms is calculated using QKT (QK^T), representing the strength of the relationship between each symptom and all other symptoms. This mechanism allows the model to capture potential dependencies between symptoms from multiple perspectives across different attention subspaces. For example, "redness and swelling at the puncture site" may have a strong correlation with "elevated body temperature" and "local exudation" in the "infection dimension," while "arm soreness" may be more correlated with "thrombosis." After calculating all attention heads, the system concatenates and linearly transforms all attention outputs to form a final unified attention representation. Based on this, the system calculates the importance score of each key symptom according to its average attention weight in the attention matrix, or by weighting it with the inner product of its attention vectors with other symptoms. These scores represent the importance of each symptom in the current context for inferring potential complications, and the system can generate a weight vector {w1, w2, ..., w...} accordingly. n The system then sorts the symptom information to guide subsequent actions. For example, when the system identifies multiple symptoms—redness and swelling at the puncture site, fever, arm soreness, and oozing at the puncture site—the attention mechanism calculation results may show that fever and oozing form a highly concentrated attentional relationship with redness and swelling, while soreness has a lower attentional correlation with other symptoms. In this case, the system assigns higher weights to fever and oozing to prioritize matching the atlas path, trigger supplementary consultation, or assign higher feature values ​​in the risk model.

[0061] Understandably, multi-head attention mechanisms can fully utilize the multi-dimensional features in the deep semantic space to effectively model the collaborative patterns among symptoms, thereby enhancing the system's early perception of potential complications. By calculating symptom weights, intelligent consultation path control can be achieved, avoiding redundant questioning and improving user interaction efficiency. As a dynamic variable input into the subsequent risk prediction model, attention weights can enhance the model's ability to express complex symptom combinations, improve the accuracy and interpretability of predictions, and are particularly suitable for complex patient scenarios with multiple but atypical symptoms.

[0062] In some examples, the real-time adjustment of the consultation path based on the key symptom information and the central venous access complication knowledge graph database, until multiple key symptom information obtained meets the risk prediction conditions of the complication knowledge graph database, includes:

[0063] Select the key symptom information with the highest weight and query the complication knowledge graph database for complications associated with the key symptom information with the highest weight;

[0064] The risk assessment of the complication requires matching at least one symptom condition group. Each symptom condition group includes at least one key symptom information. Meeting any of the symptom condition groups satisfies the risk prediction conditions for the complication.

[0065] Based on the symptom condition group, candidate questions are generated that are associated with the key symptom information with the highest weight;

[0066] Priority scores will be assigned based on the importance of the corresponding complication paths in the knowledge graph for each candidate question;

[0067] Based on prioritizing inquiries, the process continues until multiple key symptom information obtained meets the risk prediction conditions of the complication knowledge graph database.

[0068] Understandably, to achieve personalized identification and dynamic consultation control of postoperative complications after central venous access, the system further introduces a real-time adjustment mechanism for the consultation path based on the linkage between key symptom information and a complication knowledge graph database. This mechanism effectively addresses the problems of low coverage and insufficient handling of individual differences in traditional static questionnaires by combining semantic modeling and knowledge reasoning. It can dynamically identify the symptom with the greatest impact on risk assessment among the key symptoms currently reported by the user as the guiding center, and gradually unfold supplementary intelligent consultation around its corresponding knowledge graph path until the risk prediction conditions for a certain complication are met.

[0069] For example, the system first assigns weights to multiple extracted key symptom information. These weights are comprehensively evaluated based on indicators such as multi-head attention mechanism output, the correlation between symptoms and high-risk paths in the knowledge graph, and the frequency of symptom co-occurrence. Then, the key symptom information with the highest current weight is selected as the initial driving point and used for semantic querying in the central venous access complication knowledge graph database. This graph constructs a multi-level association network between symptoms, symptom groups, and complications, including not only the direct relationship between symptoms and complications but also the symptom condition groups (i.e., a set of symptoms sufficient to form the basis for inferring the complication) required for risk assessment of each complication. After the system queries all potential complications associated with the high-weight symptom, it continues to analyze the risk prediction conditions of these complications, specifically identifying their corresponding symptom condition groups. These condition groups are constructed and structured by clinical experts, indicating that the combined occurrence of one or more sets of key symptoms will trigger a risk warning for the complication. Each symptom condition group may contain 1 to N key symptom information, and satisfying any one set is sufficient for a risk assessment. For example, in cases of catheter-related infections, there may be multiple parallel symptom condition groups, including: "redness and swelling at the puncture site and fever," "redness and swelling at the puncture site and effusion," and "chills, fever, and puncture pain." Based on this structure, the system will automatically generate a set of supplementary questions for the complication pathways associated with the current high-weight symptoms. These questions are constructed from the semantic labels of the corresponding symptom nodes in the graph, and are differentiated by combining information not currently provided by the user and information needed for prediction. For example, if redness and swelling at the puncture site have been identified, and if a complication risk condition group also requires fever or effusion, the system will generate precise questions such as "Do you have a fever today?" and "Is there fluid effusion at the puncture site?" as candidates. When multiple complication pathways exist, the system will also assign a priority score to each candidate question based on the importance of each complication pathway in the graph. This score is derived from factors such as the confidence of the symptom node in the pathway, the high probability of the complication at this stage, and the pathway hit rate in historical consultation data, ensuring that high-risk, high-value questions are asked first. Subsequently, the system pushes supplementary consultation questions one by one according to priority and updates the key symptom information set in real time. Each time a new symptom is confirmed, the system re-verifies whether all current symptom information meets any symptom condition group in a certain complication pathway. If it does, the current inquiry process is immediately stopped, and the process proceeds to the risk level prediction and intervention suggestion process; if it does not meet, the next higher priority question is then addressed.

[0070] Understandably, by using high-weighted symptom-driven path selection, redundant full-map traversal is avoided, significantly improving consultation efficiency. The combination of symptom condition group structure makes the consultation process structurally convergent, enabling rapid collection of diagnostic information within the shortest question path. The priority ranking of candidate questions and real-time graph feedback mechanism ensure that key risks are not overlooked, improving the completeness, directionality, and clinical credibility of the system's consultation. More importantly, this mechanism supports dynamic adaptation across individual differences, truly realizing a new intelligent remote health management model that tailors consultations to the individual and the disease, particularly suitable for the early intervention needs of postoperative complications of high-risk devices such as central venous catheters.

[0071] In some examples, the extraction of key symptom information associated with complications of the central venous access from patient user feedback messages on pre-existing issues includes:

[0072] Based on the ResNet50 convolutional neural network, image features are extracted from the image information in the patient's feedback message to the previous question, and image feature vectors including at least one type of information such as color distribution, redness and swelling area outline and effusion trace are obtained.

[0073] Based on the symptom evolution information in the patient's feedback messages on the preceding issues, a time-stamped symptom sequence is constructed, the symptom development trajectory pattern is identified, and a time-series state feature vector is output.

[0074] The key symptom information is obtained by fusing the image feature vector, the time series state feature vector, and the text information feature vector in the feedback message.

[0075] Understandably, in order to improve the sensitivity and accuracy of identifying the risk of complications in central venous access, this implementation method introduces a multimodal information fusion strategy in the extraction of key symptom information. It is no longer limited to structured text, but combines image information, time series information and natural language description to construct a more clinically realistic and predictive expression of symptom features.

[0076] For example, after receiving feedback from patients regarding pre-existing issues, the system first identifies and analyzes the image content, then extracts time-series symptom evolution patterns, and finally fuses the image feature vector, time-series state feature vector, and text semantic vector to generate a unified format representation of key symptom information. In terms of image processing, the system calls an image feature extraction module based on a ResNet50 convolutional neural network to automatically analyze user-uploaded photos. Users often upload images related to puncture sites via mobile devices, such as puncture sites, infusion port areas, and swollen areas of the limbs. After this image is input into the ResNet50 network, its residual connection structure effectively avoids the gradient vanishing problem while maintaining deep feature representation capabilities, thereby extracting multi-scale semantic features of the image. The feature map output by the network is processed through global average pooling and high-order feature encoding to form a set of image feature vectors, specifically including but not limited to: the color distribution of the swollen area, the edge contour morphology (e.g., whether there is symmetry or diffusion), and the density and area of ​​exudate traces. These features are aggregated through feature channels to form an image representation used for subsequent judgment. The system also performs structured analysis on the symptom evolution information actively or indirectly provided by patients in their feedback messages to construct a timestamped symptom sequence. This sequence can originate from consultation information over several consecutive days or time-sensitive descriptions in a single question-and-answer session (such as "it started turning red yesterday," "it's noticeably swollen today," "it was a little itchy before"). The system uses a time analysis module to identify the start time, duration, and evolution trend of each symptom and constructs a symptom evolution timeline. The system also employs a Transformer-based temporal modeling network to dynamically encode the symptom sequence, extracting dynamic elements such as the rate of symptom change, stage-specific high-incidence features, and increasing trends, ultimately generating a set of time-series state feature vectors to characterize the potential evolutionary risk of current complications. After the above image feature vectors and time-series feature vectors are extracted, the system also uses text feature vectors extracted from natural language feedback (such as key symptom words and their contextual semantic vectors obtained through BERT or BiLSTM models) as a third modality input, which is then fed into the multimodal feature fusion module for fusion processing along with the first two. This fusion can be based on feature concatenation followed by dimensionality unification using a multilayer perceptron (MLP), or it can employ an attention fusion mechanism to dynamically assign importance weights to different modalities within the current sample. For example, in cases of poor image quality, the system can automatically increase the weights of textual and time-series information in the fused vector. Ultimately, this fused vector serves as a complete representation of key symptom information and is then used in subsequent knowledge graph matching and risk prediction processes.

[0077] Understandably, image features address the difficulty patients face in accurately describing visually apparent symptoms such as redness and effusion, enhancing objectivity. Time-series modeling enables the system to identify symptom aggravation trends or recurring fluctuations, supporting dynamic evolutionary judgments rather than static snapshots. Textual information preserves the richness of subjective feelings and linguistic expressions, strengthening the semantic link between the system and the user. The key symptom information vector structure formed by the fusion of these three elements not only enriches the symptom dimensions but also improves the system's ability to identify atypical symptom combinations. This is particularly suitable for complex case scenarios after surgery where multiple sources of information are inconsistent, descriptions are vague, but the condition is real, thus better supporting subsequent personalized path decisions and high-confidence risk predictions.

[0078] In some examples, a silent symptom prediction sub-model is used to infer potential symptoms that are not actively mentioned by the patient in the current consultation context, so as to infer abnormal silent symptoms when no key symptoms are directly expressed, and to add the abnormal silent symptoms to subsequent consultation questions.

[0079] According to some embodiments, the method of using a silent symptom prediction sub-model to infer potential symptoms not actively mentioned by the patient in the current consultation context, so as to infer abnormal silent symptoms even without receiving direct expression of key symptoms, and to add the abnormal silent symptoms to subsequent consultation questions, includes:

[0080] Construct a semantic feature representation of the current patient based on the acquired explicit symptom features;

[0081] By comparing with the pre-defined symptom co-occurrence probability distribution, identify missing symptoms that are not present in the current conversation but are highly related to existing symptoms;

[0082] The missing symptom is used as a mask position and input into the mask language model for probability prediction to obtain the significance score of the missing symptom in the current context.

[0083] When the significance score exceeds the set threshold, the missing symptom is marked as an abnormal silent symptom, and the abnormal silent symptom is added to subsequent consultation questions.

[0084] Understandably, to address the issue of missed reports of important symptoms during remote self-consultation, which may be due to patients' lack of subjective awareness, insufficient expression, or insufficient understanding of key symptoms of complications, the system also introduces a silent symptom prediction sub-model. This sub-model is used to identify and proactively infer abnormal silent symptoms that are not mentioned by the patient but have a high probability of co-occurrence in medicine. Based on comorbidity modeling and contextual language prediction principles, this mechanism can supplement inferences for missing but potentially crucial symptoms, thereby improving the completeness of consultation information and the sensitivity of risk identification.

[0085] For example, the system first constructs a semantic feature representation of the patient based on the explicit symptoms reported by the current patient. This representation integrates the contextual semantic vectors of symptom keywords mentioned in the text (e.g., extracted based on BERT / ALBERT), image features (e.g., degree of redness and swelling at the puncture site, morphology of exudate, etc.), and temporal evolution features (e.g., duration of fever, trend of redness and swelling expansion, etc.), and fuses them into a complete patient symptom semantic vector through a unified encoder. Subsequently, the system calls a pre-trained symptom co-occurrence probability distribution model to compare with the explicit symptoms in the current semantic vector and search for symptoms with high medical co-occurrence rates but not mentioned in the current dialogue, i.e., identifying the missing symptom candidate set. For example, if the current patient has reported redness and swelling at the puncture site and elevated body temperature, the system can identify missing symptoms strongly correlated with this combination, such as exudate, chills, and tenderness at the puncture site, from the statistical graph. After identifying the missing symptom candidates, the system inputs these symptoms as mask positions into the mask language model and performs conditional probability prediction based on the current dialogue context. This model generates a probability distribution of each missing symptom as omitted but crucial information in the current context based on contextual information, and outputs a significance score for each candidate missing symptom, reflecting the irrationality of its silence. If the significance score of a missing symptom exceeds a dynamic threshold set by the system (this threshold can be set based on symptom importance, contextual entropy value, or historical misjudgment cost), the system marks the missing symptom as an abnormally silent symptom, considering that although the symptom has not been complained of, it is highly likely to exist and has high clinical value. Based on this, the system immediately incorporates the information marked as an abnormally silent symptom into subsequent consultation questions. For example, if chills are not mentioned in the current context, but their significance score exceeds the threshold in the infection path, the system will automatically add a question: "Have you had chills or feeling cold all over in the last two days?" This questioning mechanism, through reverse mining of abnormal information, supplements the one-way information chain in the traditional consultation process that relies solely on active expression, effectively improving the sensitivity of symptom collection and the contextual completeness of the prediction model. Therefore, without increasing the burden on patients, the system significantly improves the coverage of consultation information and the sensitivity of complication identification by supplementing unexpressed but highly important information through reasoning; the identification of silent symptoms avoids the risk of missing early abnormalities by relying solely on surface complaints, and is particularly suitable for complications with atypical symptom expression such as infectious and thrombotic diseases; the constructed salience prediction mechanism is interpretable, and the system can output the silent score behind each additional question, which helps medical staff understand the basis of the model's reasoning and helps to establish a human-machine collaborative trust mechanism.

[0086] Please see Figure 2 One embodiment of the central venous access complication risk management device in this application may include:

[0087] Extraction unit, used to extract key symptom information associated with complications of the central venous access from the patient user's feedback messages on the preceding questions;

[0088] The analysis unit is used to adjust the consultation path in real time based on the key symptom information and the knowledge graph database of central venous access complications until the obtained key symptom information meets the risk prediction conditions of the knowledge graph database.

[0089] The treatment unit is used to input the obtained multiple key symptom information into the XGBoost gradient boosting tree model to classify and output the risk prediction level, so as to trigger the associated treatment management process based on the risk prediction level.

[0090] In summary, the central venous access complication risk management device provided in this application extracts key symptom information associated with central venous access complications from patient feedback messages regarding pre-emptive issues; adjusts the consultation path in real time based on the key symptom information and a central venous access complication knowledge graph database until multiple key symptom information meets the risk prediction conditions of the knowledge graph database; inputs the obtained multiple key symptom information into an XGBoost gradient boosting tree model to classify and output risk prediction levels, triggering associated treatment management processes based on the risk prediction levels. This significantly improves real-time performance, eliminating reliance on manual intervention for consultation and assessment, allowing for immediate identification and response to symptom changes; enhances personalization, as the consultation path is no longer a fixed template but dynamically controlled based on the knowledge graph, ensuring complete acquisition of key information; and strengthens objectivity in assessment, as machine learning-based model judgment replaces subjective experience, resulting in greater repeatability and accuracy. Furthermore, this solution supports multimodal input in practical deployment, such as combining uploaded puncture site images for image recognition to assist in judging infection signs, further enhancing the system's ability to identify heterogeneous symptoms and truly achieving data-driven complication risk prediction and early intervention.

[0091] like Figure 3 As shown, this application embodiment also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 320 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of any of the above-described methods for managing the risk of complications of central venous access.

[0092] Since the electronic device described in this embodiment is the device used to implement the central venous access complication risk management device in the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiments of this application is within the scope of protection of this application.

[0093] In practical implementation, when the computer program 311 is executed by the processor, it can achieve the following: Figure 1 Any of the corresponding implementation methods in the embodiments.

[0094] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0095] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0096] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0097] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0098] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0099] This application also provides a computer program product, which includes computer software instructions that, when executed on a processing device, cause the processing device to perform actions such as... Figure 1 The corresponding embodiment describes the process for managing the risk of complications from central venous access.

[0100] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0101] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0102] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0103] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0104] Furthermore, the functional units in the various embodiments of this application 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.

[0105] 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. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and 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 this application. 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.

[0106] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for managing the risk of complications from central venous access, characterized in that, include: Extract key symptom information associated with complications of the central venous access from patient user feedback messages regarding pre-existing issues; The consultation path is adjusted in real time based on the key symptom information and the knowledge graph database of central venous access complications until the obtained key symptom information meets the risk prediction conditions of the knowledge graph database. The obtained key symptom information is input into the XGBoost gradient boosting tree model to classify and output the risk prediction level, so as to trigger the associated treatment and management process based on the risk prediction level.

2. The method as described in claim 1, characterized in that, The extraction of key symptom information related to complications of the central venous access from patient feedback messages regarding pre-existing issues includes: The system calls the BERT-BiLSTM model to convert the words in the patient user's feedback message to the preceding question into context-related semantic vectors and models word order and dependency relations to generate a word vector sequence. Key symptom information associated with complications of the central venous access can be obtained by label classification based on word vector sequences or by similarity matching with a symptom word vector library.

3. The method as described in claim 2, characterized in that, Also includes: A multi-head attention mechanism is used to calculate the relevance of all key symptom information extracted from patient user feedback messages on pre-existing questions. The weights of each key symptom information are obtained through inner product weighting or attention matrix.

4. The method as described in claim 2, characterized in that, The process involves adjusting the consultation path in real time based on the key symptom information and the central venous access complication knowledge graph database until multiple key symptom information pieces meet the risk prediction conditions of the complication knowledge graph database, including: Select the key symptom information with the highest weight and query the complication knowledge graph database for complications associated with the key symptom information with the highest weight; The risk assessment of the complication requires matching at least one symptom condition group. Each symptom condition group includes at least one key symptom information. Meeting any of the symptom condition groups satisfies the risk prediction conditions for the complication. Based on the symptom condition group, candidate questions are generated that are associated with the key symptom information with the highest weight; Priority scores will be assigned based on the importance of the corresponding complication paths in the knowledge graph for each candidate question; Based on prioritizing inquiries, the process continues until multiple key symptom information obtained meets the risk prediction conditions of the complication knowledge graph database.

5. The method as described in claim 1, characterized in that, The extraction of key symptom information related to complications of the central venous access from patient feedback messages regarding pre-existing issues includes: Based on the ResNet50 convolutional neural network, image features are extracted from the image information in the patient's feedback message to the previous question, and image feature vectors including at least one type of information such as color distribution, redness and swelling area outline and effusion trace are obtained. Based on the symptom evolution information in the patient's feedback messages on the preceding issues, a time-stamped symptom sequence is constructed, the symptom development trajectory pattern is identified, and a time-series state feature vector is output. The key symptom information is obtained by fusing the image feature vector, the time series state feature vector, and the text information feature vector in the feedback message.

6. The method according to any one of claims 1 to 6, characterized in that, Also includes: The silent symptom prediction sub-model is used to infer potential symptoms that are not actively mentioned by the patient in the current consultation context, so as to infer abnormal silent symptoms when no key symptoms are directly expressed, and to add the abnormal silent symptoms to subsequent consultation questions.

7. The method as described in claim 6, characterized in that, The method of using a silent symptom prediction sub-model to infer potential symptoms not actively mentioned by the patient in the current consultation context, so as to infer abnormal silent symptoms even without receiving direct expression of key symptoms, and to add the abnormal silent symptoms to subsequent consultation questions, includes: Construct a semantic feature representation of the current patient based on the acquired explicit symptom features; By comparing with the pre-defined symptom co-occurrence probability distribution, identify missing symptoms that are not present in the current conversation but are highly related to existing symptoms; The missing symptom is used as a mask position and input into the mask language model for probability prediction to obtain the significance score of the missing symptom in the current context. When the significance score exceeds the set threshold, the missing symptom is marked as an abnormal silent symptom, and the abnormal silent symptom is added to subsequent consultation questions.

8. A central venous access complication risk management device, characterized in that, include: Extraction unit, used to extract key symptom information associated with complications of the central venous access from the patient user's feedback messages on the preceding questions; The analysis unit is used to adjust the consultation path in real time based on the key symptom information and the knowledge graph database of central venous access complications until the obtained key symptom information meets the risk prediction conditions of the knowledge graph database. The treatment unit is used to input the obtained multiple key symptom information into the XGBoost gradient boosting tree model to classify and output the risk prediction level, so as to trigger the associated treatment management process based on the risk prediction level.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program stored in the memory, implements the steps of the central venous access complication risk management method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the central venous access complication risk management method as described in any one of claims 1-7.

Images