Smart medical record generation method and system for realizing non-sensitive experience

Abstract

The invention relates to the field of medical informatization, and discloses an intelligent medical record generation method and system for realizing non-sensitive experience, and the method comprises the steps: carrying out the real-time dynamic capturing of a whole patient reception process, and extracting a patient physiological parameter boundary, a doctor seeing interaction node and a diagnosis and treatment environment noise region based on a medical scene semantic segmentation algorithm; acquiring a diagnosis and treatment state multi-source data stream sequence; performing cross-modal time sequence deep association matching on the diagnosis and treatment state multi-source data stream sequence; based on the multi-dimensional dynamic feature matrix, utilizing an adaptive semantic evolution clustering algorithm to extract potential semantic offset paths in a medical record generation process, and identifying high-risk information distortion candidate nodes by detecting distribution abnormity of semantic evolution trajectories; dividing the diagnosis and treatment semantic units and the influence intervals corresponding to the high-risk information distortion candidate nodes as candidate correction areas; and generating a map based on the local medical record, and performing intervention decision on the information evolution link. The method has the advantage of improving the experience feeling of the patient.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical informatization, in particular to a smart medical record generation method and system for realizing a non-sensory experience. BACKGROUND

[0002] With the development of medical informatization, electronic medical record systems have been widely used in most hospitals, becoming an important tool for recording patient information, assisting in clinical decision-making and improving medical quality. Traditional electronic medical record systems mainly rely on doctors to manually enter patient complaints, diagnoses, test results, treatment suggestions and other clinical information through keyboards, mice and other means. This process not only consumes time and effort, but also may cause doctors to be distracted, affecting the quality of doctor-patient communication. Currently, some smart medical record systems have introduced speech recognition, template reuse and other technologies to improve medical record generation efficiency. However, these technologies still have obvious shortcomings in actual application: on the one hand, the accuracy of speech recognition is affected by environmental noise, pronunciation of professional terms and other factors, resulting in the need for frequent corrections and increasing the burden on doctors; on the other hand, template generation is difficult to adapt to complex and variable clinical scenarios, resulting in semantic expression rigidity and lack of individualization. More importantly, existing systems are still fragmented from the natural diagnosis and treatment process of doctors, resulting in frequent switching between diagnosis and medical record entry for doctors, leading to poor patient experience. Therefore, it is necessary to design a smart medical record generation method and system for realizing a non-sensory experience that improves patient experience. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides a smart medical record generation method and system for realizing a non-sensory experience, which has the advantage of improving patient experience and solving the problems in the above background technology.

[0004] To achieve the above purpose of improving patient experience, the present application provides the following technical solutions: a smart medical record generation method for realizing a non-sensory experience, comprising the following steps: Real-time dynamic capture of the entire patient treatment process, and extraction of patient physiological parameter boundaries, treatment interaction nodes and diagnosis and treatment environment noise zones based on a medical scene semantic segmentation algorithm, to obtain a diagnosis and treatment state multi-source data stream sequence; Cross-modal time series deep correlation matching of the diagnosis and treatment state multi-source data stream sequence, and construction of a multi-dimensional dynamic feature matrix of the medical record generation context in combination with patient historical health trajectories, treatment behavior frequency and medical operation mode; Based on the multi-dimensional dynamic feature matrix, use of an adaptive semantic evolution clustering algorithm to extract potential semantic deviation paths in the medical record generation process, and identification of high-risk information distortion candidate nodes by detecting the distribution of semantic evolution trajectories; The diagnosis and treatment semantic unit and the influence interval corresponding to the high-risk information distortion candidate node are divided as a candidate rectification region, and patient physiological nodes, medical staff interaction nodes and environmental interference nodes in the candidate region are fused to construct a local medical record generation graph; Based on the local medical record generation graph, intervention decisions are made on the information evolution link, and a smart medical record generation result is generated.

[0005] Preferably, the process of acquiring the diagnosis and treatment state multi-source data stream sequence is: The original data of various collection sources are time-aligned through a unified timestamp synchronization mechanism, and the signal-to-noise ratio of images and speech is enhanced, and the dynamic range of illumination and audio is corrected; Based on a medical scene semantic segmentation algorithm, patient entities, medical staff, key operation devices and surrounding interference sources in continuous frame images and speech segments are labeled and semantically recognized frame by frame, and core information segments related to diagnosis and treatment behaviors are extracted; The processed image frames, speech segments and medical device parameter sequences are spliced in multiple channels, and are combined according to the reception process stage to form a diagnosis and treatment state multi-source data stream sequence.

[0006] Preferably, the process of cross-modal time series deep correlation matching of the diagnosis and treatment state multi-source data stream sequence is: From the diagnosis and treatment state multi-source data stream sequence, multi-modal information segments within the corresponding time window are extracted, and based on an improved cross-modal deep correlation algorithm, the features of video frames, speech segments and medical instrument data are aligned and jointly embedded; A patient historical health trajectory backtracking mechanism is introduced to associate model the time series diagnosis and treatment nodes, behavior interaction patterns and environmental backgrounds in the past electronic medical records to form a reference trajectory feature set; An interaction frequency and semantic consistency statistical model is used to quantitatively score the interaction behaviors between patients, medical staff and medical devices, and to identify potential fuzzy matching segments in the semantics; In the case of missing segments, noise interference or incomplete modalities in the data, a multi-source Bayesian inference filter is used to dynamically complete and predict the missing modalities; Finally, the correlation matching result is output.

[0007] Preferably, the process of constructing a multi-dimensional dynamic feature matrix of the medical record generation context is: The multi-modal consistency features output in the cross-modal time series deep correlation matching are one-to-one mapped with the entity categories identified by the medical scene semantic segmentation algorithm to establish a correspondence between the patient behavior semantic segments and the specific diagnosis and treatment nodes; The behavior dynamic parameters within the time window are extracted, the auxiliary information recorded by the multi-modal perception collection terminal is fused, and through normalization processing, the data of different modalities are mapped to a unified feature space; Adopting sliding window statistical mechanism, the behavior dynamic parameters and auxiliary information are time-aggregated to calculate the average, variance, change rate and semantic confidence of the features in the window, and local semantic state fragments are generated; The local semantic state fragments are spliced according to the order of diagnosis and treatment process to form a multi-dimensional dynamic feature matrix.

[0008] Preferably, the potential semantic drift path process in the medical record generation process is extracted by using an adaptive semantic evolution clustering algorithm, and the process is as follows: Based on the multi-dimensional dynamic feature matrix, an improved adaptive density peak value semantic clustering algorithm is adopted to identify the diagnosis and treatment semantic mode; Each clustering result is given a context semantic label, the migration trajectory of the clustering center over time is analyzed, a diagnosis and treatment semantic state transition graph is constructed, and high-frequency transition nodes and low-probability mutation nodes are marked; The transition frequency and transition probability between semantic states are counted to identify the formation path of high-risk semantic drift, and the semantic drift path is output.

[0009] Preferably, the process of identifying high-risk information distortion candidate nodes is as follows: The medical record generation semantic evolution trajectory is compared with the historical standard medical record trajectory database to construct a reference semantic trajectory set; A semantic graph similarity comparison algorithm is introduced to structure match the current diagnosis and treatment semantic state transition graph with the historical standard trajectory graph, and locate the node section with decreased similarity; Based on the semantic graph structure comparison result, the key state nodes in the current trajectory that have semantic drift, link breakage or reverse jump compared with the standard path are identified; Combined with the semantic deviation threshold and the context confidence scoring mechanism, the trajectory fragments containing semantic drift nodes are marked; If the semantic deviation is greater than the preset threshold, the trajectory fragment is determined as an abnormal semantic section, and is included in the high-risk information distortion candidate node.

[0010] Preferably, the process of delimiting the diagnosis and treatment semantic unit and the influence interval corresponding to the high-risk information distortion candidate node as the candidate correction area is as follows: According to the abnormal semantic fragments marked in the candidate nodes, the core position and duration of semantic anomaly are determined; The core time position is mapped to the diagnosis and treatment process structured model to identify the directly related diagnosis and treatment links, medical staff interaction units and corresponding medical documents fields; Using the node number mapping mechanism of the electronic medical record template and the diagnosis and treatment path model, the context information and logical dependency relationship near the abnormal semantic nodes are extracted to preliminarily delimit the semantic boundary containing the distortion nodes; Combine the patient real-time physiological parameter fluctuation with the cooperation behavior track of medical staff in the time window, analyze the potential semantic distortion propagation path and the case field range affected by the potential semantic distortion propagation path; The boundary range of the candidate rectification region is dynamically adjusted to form a final candidate rectification region.

[0011] Preferably, the local medical record generation graph construction process is: From the candidate rectification region, patient physiological parameters, medical care interaction records and equipment data that are spatially and temporally coincident with the high-risk information distortion candidate node are extracted, and are respectively identified as patient physiological nodes, medical care interaction nodes and environmental interference nodes, and are summarized to form a multi-source node pool with heterogeneous attributes; Based on the multi-source node pool, a directed edge set between nodes is constructed according to historical association records between nodes, diagnosis and treatment logic dependency relationships and medical record field mapping constraints, and a multi-modal medical record graph structure is formed; A semantic stability label is added to each type of node, the constructed multi-modal medical record graph structure is input into a graph neural network, key topological patterns in the graph are mined through an embedded learning algorithm, semantic drift paths and information distortion trigger points with high risk influence degree are identified, and finally a local medical record generation graph is generated.

[0012] The intelligent medical record generation system realizes a non-sensory experience, and comprises: A perception acquisition module: real-time dynamic capture is performed on the whole process of patient reception, and multi-source diagnosis and treatment state data acquisition and semantic segmentation are completed; An association matching module: cross-modal time series deep association matching is performed on the collected diagnosis and treatment state multi-source data stream sequence, and a multi-dimensional dynamic feature matrix of medical record generation is constructed in combination with patient historical health trajectories and medical care operation modes; A semantic analysis module: based on the multi-dimensional dynamic feature matrix, adaptive semantic evolution clustering analysis is performed, semantic shift paths in the medical record generation process are extracted, distribution abnormalities are detected, and high-risk information distortion candidate nodes are identified; A region construction module: the high-risk information distortion candidate nodes are mapped to diagnosis and treatment semantic units and influence intervals, candidate rectification regions are delineated, and patient physiological nodes, medical care interaction nodes and environmental interference nodes are fused to construct a local medical record generation graph; A result generation module: based on the local medical record generation graph, information evolution link intervention and rectification decision are performed, and an intelligent medical record generation result is output.

[0013] Compared with the prior art, the intelligent medical record generation method and system realizing a non-sensory experience have the following beneficial effects: The application can comprehensively retain physiological parameters of patients, medical interaction information and environmental background factors in the medical record generation process, ensure the integrity and spatio-temporal consistency of diagnosis and treatment data, and realize the generation of intelligent medical records with consistent semantics and coherent context by constructing a multi-dimensional dynamic feature matrix of the medical record generation context and using an adaptive semantic evolution clustering algorithm to identify semantic drift paths, so as to timely find abnormal nodes such as potential information distortion, semantic drift or diagnosis and treatment record breakage, improve the accuracy and semantic continuity of medical record content, realize directional intervention and semantic correction of abnormal information links by automatic demarcation of candidate correction areas and construction of local medical record generation atlas, effectively avoid record deviation caused by environmental noise, operation omission or data loss, and finally generate intelligent medical records with consistent semantics and coherent context, which can dynamically reflect the real diagnosis and treatment process and state change of patients, significantly improve the automation degree of medical record generation, information credibility and the 'non-invasive' level of patient experience, and provide high-precision, traceable semantic medical record data basis for subsequent intelligent analysis, auxiliary diagnosis and medical quality evaluation. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A schematic diagram of the method of the application is shown in the figure. Figure 2 A schematic diagram of the system of the application is shown in the figure. DETAILED DESCRIPTION

[0015] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0016] Embodiment 1: Please refer to Figure 1 The intelligent medical record generation method for realizing non-invasive experience described in the embodiments of the application includes the following steps: S1: Real-time dynamic capture is performed on the whole process of patient reception, and based on a medical scene semantic segmentation algorithm, patient physiological parameter boundaries, medical interaction nodes and diagnosis and treatment environment noise areas are extracted to obtain a diagnosis and treatment state multi-source data stream sequence.

[0017] The process of obtaining the diagnosis and treatment state multi-source data stream sequence in S1 is as follows: A multi-modal perception acquisition terminal is arranged in the reception scene, including a high-resolution video acquisition device, a voice interaction acquisition module, a medical instrument data interface and an environmental noise monitoring unit, which are respectively used to acquire physiological characteristics of patients, medical language communication information, real-time diagnosis and treatment data and environmental background interference signals. The original data of various types of acquisition sources are time-aligned through a unified timestamp synchronization mechanism, and the image and voice are enhanced in signal-to-noise ratio, and the light and audio dynamic range are corrected, to ensure the consistency of multi-source information in time and space dimensions. Based on a medical scene semantic segmentation algorithm, the patient entity, medical staff, key operation equipment and surrounding interference source in the continuous frame image and voice segment are frame-by-frame labeled and semantically recognized, and the core information segment related to the diagnosis and treatment behavior is extracted. The processed image frames, voice segments and medical equipment parameter sequences are multi-channel spliced and combined according to the reception process stages to form a diagnosis and treatment state multi-source data stream sequence.

[0018] In the reception scene, multi-modal perception acquisition terminals are pre-deployed, including high-resolution video acquisition equipment for capturing patient expressions and actions, voice interaction acquisition modules for acquiring medical language communication and patient voice feedback, medical instrument data interfaces for recording vital signs and real-time diagnosis and treatment parameters, and environmental noise monitoring units for detecting background noise and environmental light changes, to ensure the coverage of patient physiological, behavioral and environmental information. Through a unified timestamp synchronization mechanism, the original acquisition data of different perception sources are time reference aligned, and image enhancement algorithms are used for light correction, contrast equalization and frame jitter compensation, and voice signal noise reduction and dynamic range compression techniques are used to improve voice clarity and stability, to ensure the consistency of multi-source data in time, space and quality dimensions. Using a medical scene semantic segmentation algorithm, the continuously acquired video frames and voice segments are frame-by-frame labeled to identify semantic objects such as patient entities, medical staff identities, key operation equipment and environmental interference sources, and through semantic filtering and context association, core information segments directly related to the current diagnosis and treatment behavior are extracted, and redundant background information is filtered out. The processed multi-source information is multi-channel spliced in time sequence, the video frames, voice segments and medical instrument parameter sequences are mapped into a unified data structure, and according to the stage division of the reception process, a structured and traceable diagnosis and treatment state multi-source data stream sequence is generated.

[0019] S2: Cross-modal time sequence deep correlation matching of the diagnosis and treatment state multi-source data stream sequence, combined with the patient's historical health trajectory, the frequency of the visit behavior and the operation mode of the medical staff, to construct a multi-dimensional dynamic feature matrix of the medical record generation context.

[0020] The process of cross-modal time sequence deep correlation matching of the diagnosis and treatment state multi-source data stream sequence in S2 is as follows: From the diagnosis and treatment state multi-source data stream sequence, the multi-modal information segments within the corresponding time window are extracted, and based on an improved cross-modal deep correlation algorithm, the video frames, voice segments and medical instrument data are aligned and jointly embedded at the feature layer. A patient history health trajectory backtracking mechanism is introduced to associate and model the time sequence diagnosis and treatment nodes, behavior interaction patterns and environmental background in the past electronic medical records to form a reference trajectory feature set; An interaction frequency and semantic consistency statistical model is used to quantitatively score the interaction behaviors between patients, medical staff and medical equipment, and identify potential fuzzy matching fragments in semantics; In view of the missing fragments, noise interference or incomplete modalities in the data, a multi-source Bayesian inference filter is used to dynamically complete and predict the missing modalities, thereby improving the stability of cross-modality time sequence matching; Finally, the associated matching results containing multi-modality consistency features, time sequence alignment information and interaction semantic relationships are output.

[0021] From the diagnosis and treatment state multi-source data stream sequence, multi-modality information fragments containing video frames, voice fragments and medical instrument parameters are extracted according to the set time window, and the consistency of different modalities in the time dimension is ensured through timestamp alignment; an improved cross-modality deep association algorithm is used to align the features of each modality, and multi-modality feature coding networks are used to extract the spatial semantic features of video, the text semantic features of voice and the numerical dynamic features of medical data, and then the cross-modality feature space is constructed through joint embedding; subsequently, a patient history health trajectory backtracking mechanism is introduced to compare and model the features of the current modality information and the time sequence diagnosis and treatment nodes, behavior interaction patterns and environmental background information in the past electronic medical records to form a trajectory feature set for reference; an interaction frequency and semantic consistency statistical model is used to quantitatively score the interaction behaviors between patients, medical staff and medical equipment, to determine the semantic consistency level in the interaction fragments, and to identify potential fuzzy matching or semantic conflict fragments; in view of the missing fragments in the multi-source data caused by incomplete collection or noise interference, a multi-source Bayesian inference filter is used in combination with a time sequence prediction model to dynamically complete the missing modalities or correct abnormal modality information, thereby improving the continuity and accuracy of time sequence matching; finally, the associated matching results containing cross-modality consistency features, time sequence alignment relationships, interaction semantic association structures and abnormal matching identifiers are output.

[0022] The process of constructing a multi-dimensional dynamic feature matrix of medical record generation context in S2 is as follows: The multi-modality consistency features output in the cross-modality time sequence deep association matching are one-to-one mapped with the entity categories identified by the medical scene semantic segmentation algorithm to establish the correspondence between the patient behavior semantic fragments and specific diagnosis and treatment nodes; The behavior dynamic parameters within the time window are extracted, including voice speed, interaction frequency, expression change rate, environmental noise fluctuation amplitude and medical instrument reading change gradient; The auxiliary information recorded by the multi-modal perception collection terminal is fused, including the doctor-patient interaction rhythm, standard diagnosis and treatment operation steps, and patient physiological response parameters, and through normalization processing, the data of different modalities are mapped to a unified feature space; The sliding window statistical mechanism is adopted to perform time sequence aggregation on the behavior dynamic parameters and the auxiliary information, the average value, variance, change rate and semantic confidence of the features in the window are calculated, and a local semantic state fragment is generated. The local semantic state fragments are spliced according to the order of the diagnosis and treatment process to form a multi-dimensional dynamic feature matrix capable of describing the semantic evolution path of the medical record generation context and the behavior logic relationship.

[0023] The multi-modal consistency features output by the cross-modal time sequence deep correlation matching are mapped to the entity categories identified by the medical scene semantic segmentation algorithm one by one, the semantic correspondence between the patient voice fragments, expression actions, doctor-patient interaction sentences, medical device data and specific diagnosis and treatment nodes is clarified, and a mapping table of behavior semantic fragments and diagnosis and treatment process nodes is constructed; the dynamic parameters of the patient behavior are extracted within a set time window, including the speech speed and tone fluctuation, doctor-patient interaction frequency and response time delay, patient facial expression and body movement change rate, environmental noise fluctuation amplitude and medical instrument reading gradient change, forming a dynamic parameter set reflecting the current diagnosis and treatment state change; these dynamic parameters are fused with the auxiliary information recorded by the multi-modal perception collection terminal, the auxiliary information includes the doctor-patient interaction rhythm, standard diagnosis and treatment operation steps and patient physiological response parameters, and through normalization processing, time stamp alignment and feature compression, the data of different modalities are mapped to a unified feature space, ensuring the consistency of feature scale and time reference; then, the sliding window statistical mechanism is adopted to perform time sequence aggregation on the behavior dynamic parameters and the auxiliary information in the time window, calculate the average value, variance, change rate and semantic confidence of the features, and generate semantic state fragments that can reflect the local evolution characteristics of the diagnosis and treatment state; these local semantic state fragments are spliced and sequenced according to the actual order of the diagnosis and treatment process to form a complete multi-dimensional dynamic feature matrix, which is used to describe the semantic evolution path and behavior logic relationship of the medical record generation context.

[0024] S3: Based on the multi-dimensional dynamic feature matrix, an adaptive semantic evolution clustering algorithm is used to extract the potential semantic deviation path in the medical record generation process, and by detecting the distribution anomaly of the semantic evolution trajectory, a high-risk information distortion candidate node is identified.

[0025] The process of extracting the potential semantic deviation path in the medical record generation process in S3 using the adaptive semantic evolution clustering algorithm is: Based on the multi-dimensional dynamic feature matrix, an improved adaptive density peak value semantic clustering algorithm is used to identify the diagnosis and treatment semantic pattern; assigning context semantic labels to each clustering result, including normal semantic link, repeated redundant segment, ambiguous offset segment and potential distortion segment; analyzing the migration trajectory of the clustering center over time, constructing a diagnosis and treatment semantic state transition graph, and marking high-frequency transition nodes and low-probability mutation nodes;

[0026] Taking a multi-dimensional dynamic feature matrix as input, an improved adaptive density peak value semantic clustering algorithm is used to automatically determine the number of clustering centers and form multiple semantic clustering clusters by calculating the local density and relative distance of feature points in a high-dimensional feature space, thereby identifying different semantic patterns existing in the diagnosis and treatment process. Each clustering cluster is annotated according to its feature distribution, context correlation and semantic consistency index, and normal continuous semantic links, areas with repeated information or segment redundancy, ambiguous or unclear boundary offset segments, and significantly abnormal potential distortion segments are distinguished and assigned corresponding context semantic labels. The evolution path of the clustering center over time is tracked, the migration direction, speed and residence time of the clustering center in different time windows are recorded, a diagnosis and treatment semantic state transition graph is constructed, and regular nodes with high-frequency cyclic appearance and mutation nodes with extremely low probability but significant impact are identified in the transition graph. Then, the transition relationship between semantic states is statistically analyzed, the transition frequency, conditional transition probability and entropy value of the transition path between states are calculated to quantify the stability and abnormal burst degree of the semantic pattern. The transition graph structure, transition probability distribution and mutation node information are integrated to identify and output high-risk offset paths from normal semantic links to ambiguous or distorted semantic segments.

[0027] The process of identifying high-risk information distortion candidate nodes in S3 is as follows: Compare the medical record generation semantic evolution trajectory with the historical standard medical record trajectory database to construct a reference semantic trajectory set; Introduce a semantic graph similarity comparison algorithm to structure match the current diagnosis and treatment semantic state transition graph with the historical standard trajectory graph, and locate the node section with decreased similarity; Based on the semantic graph structure comparison result, identify the key state nodes in the current trajectory that have semantic drift, link breakage or reverse jump compared with the standard path; Combine the semantic deviation threshold and the context confidence score mechanism to mark the trajectory segment containing semantic offset nodes; If the semantic deviation is greater than the preset threshold, the trajectory segment is determined to be an abnormal semantic section, and is included in the high-risk information distortion candidate node.

[0028] ​The semantic evolution trajectory extracted in the previous step is compared with the historical standard medical record trajectory database, a reference semantic trajectory set for comparative analysis is constructed by calling various typical diagnosis and treatment semantic link samples stored in the standard trajectory library, a semantic graph similarity comparison algorithm based on node structure and path sequence comprehensive features is introduced, the current diagnosis and treatment semantic state transition graph is matched with the historical standard trajectory graph node by node and link by link, the node matching rate, path overlap rate and context semantic consistency score are calculated, and the node section with significant similarity decrease or path incomplete overlap is located; then, according to the semantic graph structure comparison result, the key state nodes with significant deviation from the standard trajectory are identified, including the drift node (the node with slight deviation from the context) in the semantic evolution path, the link break node (the node with interrupted or missing original normal semantic link) and the reverse jump node (the node with nonlinear rollback or jump of semantic link); combined with the preset semantic deviation threshold and the context confidence score mechanism, the trajectory segment containing the above deviation nodes is marked as risk, and the incidental deviation caused by environmental noise or redundant information is filtered out through multiple rounds of confidence weighted calculation; for the trajectory segment with deviation exceeding the preset threshold and significantly reduced context confidence, it is determined as an abnormal semantic section, and the drift node, break node and jump node in the section are classified as high-risk information distortion candidate nodes.

[0029] S4: The diagnosis and treatment semantic unit and the influence interval corresponding to the high-risk information distortion candidate node are divided into a candidate correction area, and the patient physiological node, medical interaction node and environmental interference node in the candidate area are fused to construct a local medical record generation graph.

[0030] The process of dividing the diagnosis and treatment semantic unit and the influence interval corresponding to the high-risk information distortion candidate node into a candidate correction area in S4 is as follows: According to the abnormal semantic segment marked in the candidate node, combined with the time sequence information of the patient behavior, medical interaction and device data to which the segment belongs, the core position and duration of semantic anomaly occurrence are determined; Map the core time position to the structured model of the diagnosis and treatment process to identify the directly related diagnosis and treatment links, medical interaction units and corresponding medical documents fields; Using the node number mapping mechanism of the electronic medical record template and the diagnosis and treatment path model, the context information and logical dependency near the abnormal semantic node are extracted to preliminarily determine the semantic boundary containing the distortion node; Combined with the real-time physiological parameter fluctuation of the patient and the cooperation behavior trajectory of the medical staff within the time window, the potential semantic distortion propagation path and the affected medical record field range are analyzed; Adjust the boundary range of the candidate correction area dynamically to form the final candidate correction area.

[0031] The local medical record generation graph construction process in S4 is: From the candidate rectification area, extract the patient physiological parameters, medical interaction records and device data that are spatially and temporally coincident with the high-risk information distortion candidate nodes, and identify them as patient physiological nodes, medical interaction nodes and environmental interference nodes, respectively, and form a multi-source node pool with heterogeneous attributes; Based on the multi-source node pool, according to the historical association records between nodes, diagnosis and treatment logic dependency relationship and medical record field mapping constraints, a directed edge set between nodes is constructed to form a multi-modal medical record graph structure reflecting the actual diagnosis and treatment logic and semantic evolution relationship; Add a semantic stability label to each type of node, input the constructed multi-modal medical record graph structure into a graph neural network, and use an embedded learning algorithm to mine key topological patterns in the graph, identify semantic drift paths with high risk influence and information distortion trigger points, and finally generate a local medical record generation graph.

[0032] S5: Based on the local medical record generation graph, make intervention decisions on the information evolution link, and generate a smart medical record generation result.

[0033] Embodiment 2: As shown in Figure 2 , a smart medical record generation system that realizes a non-inductive experience includes: Perception acquisition module: Real-time dynamic capture of the entire patient admission process is performed to complete the acquisition and semantic segmentation of multi-source diagnosis and treatment state data; Correlation matching module: Cross-modal temporal deep correlation matching is performed on the collected diagnosis and treatment state multi-source data stream sequence, and combined with the patient's historical health trajectory and medical operation mode, a multi-dimensional dynamic feature matrix for medical record generation is constructed; Semantic analysis module: Based on the multi-dimensional dynamic feature matrix, perform adaptive semantic evolution clustering analysis, extract the semantic drift path in the medical record generation process and detect the distribution anomaly, and identify high-risk information distortion candidate nodes; Area construction module: Map the high-risk information distortion candidate nodes to the diagnosis and treatment semantic units and influence intervals, delineate the candidate rectification area, and integrate the patient physiological nodes, medical interaction nodes and environmental interference nodes to construct a local medical record generation graph; Result generation module: Based on the local medical record generation graph, make intervention and rectification decisions on the information evolution link, and output a smart medical record generation result.

[0034] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting; it is not intended to exclude myriad other embodiments of the present application that other inventors can develop based on the same general inventive concepts embodied by the described embodiments. That is, although the present application is described in terms of particular embodiments and illustrative figures, it should be apparent that the scope of the present application is not limited to these specific embodiments.

[0035] While the embodiments of the application have been shown and described herein, it will be understood by those skilled in the art that many changes, modifications, substitutions and alterations to these embodiments can be made without departing from the principles and spirits of the application, and it is intended that the scope of the application be limited solely by the scope of the appended claims and the equivalents thereof.

Claims

1. A smart medical record generation method for achieving a non-inductive experience, characterized by, The method comprises the following steps: Real-time dynamic capture is performed on the whole process of patient reception, and patient physiological parameter boundaries, medical treatment interaction nodes and diagnosis and treatment environment noise zones are extracted based on a medical scene semantic segmentation algorithm to obtain a diagnosis and treatment state multi-source data stream sequence; Cross-modal time sequence deep correlation matching is performed on the diagnosis and treatment state multi-source data stream sequence, and a multi-dimensional dynamic feature matrix of a medical record generation context is constructed in combination with a patient historical health trajectory, a medical treatment behavior frequency and a medical operation mode; Based on the multi-dimensional dynamic feature matrix, an adaptive semantic evolution clustering algorithm is used to extract a potential semantic deviation path in a medical record generation process, high-risk information distortion candidate nodes are identified by detecting distribution abnormalities of semantic evolution trajectories, and a diagnosis and treatment semantic unit and an influence interval corresponding to the high-risk information distortion candidate nodes are delimited as a candidate rectification region; A local medical record generation graph is constructed by fusing patient physiological nodes, medical staff interaction nodes and environmental interference nodes in the candidate region; Intervention decisions are made on information evolution links based on the local medical record generation graph, and a smart medical record generation result is generated.

2. The smart medical record generation method for achieving a no-touch experience according to claim 1, wherein, The process of obtaining the diagnosis and treatment state multi-source data stream sequence comprises the following steps: Raw data of various collection sources are time-aligned through a unified timestamp synchronization mechanism, and the signal-to-noise ratio of images and voices is enhanced, and the dynamic range of light and audio is corrected; Based on the medical scene semantic segmentation algorithm, patient entities, medical staff, key operation devices and surrounding interference sources in continuous frame images and voice segments are frame-by-frame labeled and semantically recognized to extract core information segments associated with diagnosis and treatment behaviors; The processed image frames, voice segments and medical device parameter sequences are multi-channel spliced and combined according to the reception process stages to form the diagnosis and treatment state multi-source data stream sequence.

3. The smart medical record generation method for achieving a no-touch experience according to claim 2, wherein, The process of cross-modal time sequence deep correlation matching of the diagnosis and treatment state multi-source data stream sequence comprises the following steps: Multi-modal information segments within a corresponding time window are extracted from the diagnosis and treatment state multi-source data stream sequence, and an improved cross-modal deep correlation algorithm is used to align and jointly embed feature layers of video frames, voice segments and medical instrument data; A patient historical health trajectory backtracking mechanism is introduced to model the time sequence diagnosis and treatment nodes, behavior interaction modes and environmental backgrounds in the past electronic medical records to form a reference trajectory feature set; An interaction frequency and semantic consistency statistical model is used to quantitatively score the interaction behaviors among patients, medical staff and medical devices to identify potential fuzzy matching segments in semantics; In the case of missing segments, noise interference or incomplete modalities in the data, a multi-source Bayesian inference filter is used to dynamically complete and predict the missing modalities; Finally, the correlation matching result is output.

4. The smart medical record generation method for achieving a no-touch experience according to claim 3, wherein, The process of constructing a multi-dimensional dynamic feature matrix of a medical record generation context comprises the following steps: The multi-modal consistency features output in the cross-modal time sequence deep correlation matching are one-to-one mapped with the entity categories recognized by the medical scene semantic segmentation algorithm to establish a correspondence between patient behavior semantic segments and specific diagnosis and treatment nodes; Behavior dynamic parameters within a time window are extracted, auxiliary information recorded by multi-modal perception collection terminals is fused, and different modal data is mapped to a unified feature space through normalization processing; The sliding window statistical mechanism is adopted to perform time sequence aggregation on the behavior dynamic parameters and auxiliary information, to calculate the average, variance, change rate and semantic confidence of the features in the window, and to generate a local semantic state fragment; The local semantic state fragments are spliced according to the order of the diagnosis and treatment process to form a multi-dimensional dynamic feature matrix.

5. The smart medical record generation method for achieving a no-touch experience according to claim 4, wherein, The adaptive semantic evolution clustering algorithm is used to extract the potential semantic drift path in the medical record generation process as follows: Based on the multi-dimensional dynamic feature matrix, an improved adaptive density peak value semantic clustering algorithm is used to identify the diagnosis and treatment semantic patterns; Contextual semantic labels are assigned to each clustering result, the migration trajectory of the clustering center over time is analyzed, a diagnosis and treatment semantic state transition graph is constructed, and high-frequency transition nodes and low-probability mutation nodes are marked; The transition frequency and transition probability between semantic states are counted, the formation path of high-risk semantic drift is identified, and the semantic drift path is output.

6. The smart medical record generation method for achieving a no-touch experience according to claim 5, wherein, The process of identifying high-risk information distortion candidate nodes is as follows: The medical record generation semantic evolution trajectory is compared with the historical standard medical record trajectory database to construct a reference semantic trajectory set; The semantic graph similarity comparison algorithm is introduced to perform structural matching between the current diagnosis and treatment semantic state transition graph and the historical standard trajectory graph, and to locate the node section with decreased similarity; Based on the semantic graph structure comparison result, the key state nodes in the current trajectory that have semantic drift, link breakage or reverse jump compared with the standard path are identified; Combined with the semantic deviation threshold and the context confidence scoring mechanism, the trajectory fragment containing the semantic drift node is marked; If the semantic deviation is greater than the preset threshold, the trajectory fragment is determined as an abnormal semantic section, and is included in the high-risk information distortion candidate node.

7. The smart medical record generation method for achieving a no-touch experience according to claim 6, wherein, The process of delimiting the diagnosis and treatment semantic unit corresponding to the high-risk information distortion candidate node and the influence interval as the candidate correction area is as follows: According to the abnormal semantic fragment marked in the candidate node, the core position and duration of the semantic anomaly are determined; The core time position is mapped to the structured model of the diagnosis and treatment process to identify the directly related diagnosis and treatment links, medical staff interaction units and corresponding medical documents fields; Using the node number mapping mechanism of the electronic medical record template and the diagnosis and treatment path model, the context information and logical dependency relationship near the abnormal semantic node are extracted to preliminarily delimit the semantic boundary containing the distortion node; Combined with the real-time physiological parameter fluctuation of the patient and the cooperation behavior trajectory of the medical staff within the time window, the potential semantic distortion propagation path and the affected medical record field range are analyzed; The boundary range of the candidate correction area is dynamically adjusted to form the final candidate correction area.

8. The smart medical record generation method for achieving a no-touch experience according to claim 7, wherein, The process of constructing a local medical record generation graph is as follows: The patient physiological parameters, medical staff interaction records and equipment data that are spatially and temporally coincident with the high-risk information distortion candidate node are extracted from the candidate correction area, and are respectively identified as patient physiological nodes, medical staff interaction nodes and environmental interference nodes to form a multi-source node pool with heterogeneous attributes; Based on the multi-source node pool, a directed edge set between nodes is constructed according to the historical association records between nodes, the diagnosis and treatment logical dependency relationship and the medical record field mapping constraints to form a multi-modal medical record graph structure; The semantic stability label is added to each type of node, the constructed multi-modal medical record graph structure is input into a graph neural network, a key topological pattern in the graph is mined through an embedded learning algorithm, a semantic drift path with high risk influence degree and an information distortion trigger point are identified, and finally a local medical record generation graph is generated.

9. A smart medical record generation system for achieving a non-invasive experience, applied to the method of any one of claims 1-8, characterized in that, Comprise: A perception acquisition module: real-time dynamic capture of the whole process of patient reception, completion of multi-source diagnosis and treatment state data acquisition and semantic segmentation; An association matching module: cross-modal time series deep association matching of the collected diagnosis and treatment state multi-source data stream sequence, construction of a multi-dimensional dynamic feature matrix of medical record generation combined with the patient's historical health trajectory and medical operation mode; A semantic analysis module: based on the multi-dimensional dynamic feature matrix, adaptive semantic evolution clustering analysis is performed, the semantic shift path in the medical record generation process is extracted and the distribution anomaly is detected, and the high-risk information distortion candidate node is identified; A region construction module: mapping the high-risk information distortion candidate node to the diagnosis and treatment semantic unit and the influence interval, delimiting the candidate correction area, and fusing the patient physiological node, medical interaction node and environmental interference node to construct a local medical record generation graph; A result generation module: based on the local medical record generation graph, intervention and correction decision of the information evolution link, and output of the intelligent medical record generation result.

Images