Search

CN-122024713-A - Medical scene multi-mode message recognition processing method and system

CN122024713ACN 122024713 ACN122024713 ACN 122024713ACN-122024713-A

Abstract

The invention relates to a medical scene multi-mode message recognition processing method and system, wherein the method comprises the steps of dynamically adjusting voice recognition parameters according to collected voice signals of doctors, environmental noise level, personnel position and activity intensity to obtain voice input data; the method comprises the steps of obtaining corrected text information according to voice input data, generating structured display content according to the corrected text information, calculating an emergency index based on an emergency degree label to obtain priority display information, obtaining dynamically updated execution feedback information according to the priority display information, recording a complete event chain according to the dynamically updated execution feedback information, generating an electronic audit log with a time stamp, generating clinical decision assistance advice by combining the log to obtain a decision support result, and triggering a rescheduling mechanism if the emergency event corresponding to the priority display information is detected not to be executed within preset time according to the decision support result. The invention can improve the accuracy of medical scene information transmission.

Inventors

  • XIONG LING

Assignees

  • 南方医科大学南方医院

Dates

Publication Date
20260512
Application Date
20251110

Claims (10)

  1. 1. A medical scene multi-mode message recognition processing method, which is characterized by comprising the following steps: according to the collected voice signals of doctors, the ambient noise level, the personnel position and the activity strength, dynamically adjusting voice recognition parameters to obtain voice input data; According to the voice input data, converting the voice input data into text information, carrying out semantic alignment on the text information by combining a medical knowledge graph, detecting context semantic conflict and outputting an error correction prompt to obtain corrected text information; generating structured display content according to the error-corrected text information, and calculating an emergency index based on an emergency label to obtain priority display information; receiving execution feedback provided by medical staff through touch screen operation or voice instructions according to the priority display information, and if the execution feedback contains abnormal description, converting the execution feedback into characters and marking the characters as the execution abnormality, and generating an abnormality notification to obtain dynamically updated execution feedback information; recording a complete event chain according to the dynamically updated execution feedback information, generating an electronic audit log with a time stamp, and generating clinical decision assistance suggestions by combining the log to obtain a decision support result; And according to the decision support result, if the fact that the emergency event corresponding to the priority display information is not executed within the preset time is detected, triggering a rescheduling mechanism to set the emergency event to be displayed on top, and ensuring that the emergency information is transmitted and executed preferentially.
  2. 2. The method for multi-modal message recognition processing in a medical scenario according to claim 1, wherein dynamically adjusting the voice recognition parameters according to the collected voice signal of the doctor and the environmental noise level, the personnel position and the activity intensity to obtain the voice input data comprises: Generating a noise suppression factor according to the environmental noise level to obtain a noise reduction voice signal; and generating a voice weight parameter according to the personnel position and the activity intensity, and applying the weight parameter to the noise reduction voice signal to obtain weighted voice input data.
  3. 3. The method for recognizing and processing the multi-modal messages in the medical scene according to claim 2, wherein the steps of converting the voice input data into text information, performing semantic alignment on the text information in combination with a medical knowledge graph, detecting context semantic conflict and outputting an error correction prompt to obtain the text information after error correction, include: Generating an initial text sequence according to the voice input data; performing context comparison on the initial text sequence according to the medical knowledge graph to obtain a candidate error correction result; And outputting the character information subjected to error correction according to the candidate error correction result and the context consistency score.
  4. 4. The method for multi-modal message recognition processing in a medical scenario according to claim 3, wherein the generating structural display content according to the corrected text information, and calculating an emergency index based on an emergency label, to obtain priority display information, includes: extracting key medical elements according to the corrected text information to obtain an element set; and generating an emergency label according to the element set and combining with a preset rule, and calculating an emergency index to obtain the corresponding priority display information.
  5. 5. The method for recognizing and processing the multi-modal messages in a medical scene according to claim 4, wherein the step of receiving the execution feedback provided by the medical staff through the touch screen operation or the voice command according to the priority display information, and if the execution feedback contains the abnormal description, converting the execution feedback into text and marking the execution feedback as the execution abnormality, and generating the abnormality notification to obtain the dynamically updated execution feedback information comprises the following steps: Generating an execution confirmation record according to the operation of the medical staff; And detecting whether an abnormal description exists according to the execution confirmation record, if so, converting the abnormal description into characters and generating an abnormal label to obtain the dynamically updated execution feedback information.
  6. 6. The method for multi-modal message recognition processing in a medical scenario according to claim 5, wherein the recording of a complete event chain according to the dynamically updated execution feedback information, generating a time-stamped electronic audit log, and generating clinical decision assistance advice in combination with the log, and obtaining a decision support result, comprises: generating a time-series event record according to the dynamically updated execution feedback information; Generating an auxiliary suggestion result according to the event record and comparison of the event record with a preset medical guideline; And combining the auxiliary suggestion result with a time stamp to obtain the clinical decision support result.
  7. 7. The method for multi-modal message recognition processing in a medical scenario according to claim 6, wherein the step of triggering a rescheduling mechanism to display the emergency event on top to ensure the priority transmission and execution of the emergency information if the emergency event corresponding to the priority display information is detected not to be completed within a preset time according to the decision support result comprises: Judging whether the emergency event has a delay state according to the decision support result to obtain a delay mark; Triggering a full-screen flashing, voice broadcasting and task suspending mechanism according to the delay mark to obtain a rescheduling execution result.
  8. 8. The method for multi-mode message recognition processing in a medical scenario according to claim 7, wherein the obtaining a rescheduling execution result comprises: Generating a visual warning record according to the full-screen flickering prompt; generating an audible warning record according to the voice broadcasting prompt; And generating a task suspension record according to a task suspension mechanism, and forming a rescheduling event chain by the visual warning record, the audible warning record and the task suspension record together.
  9. 9. The method for multi-modal message recognition processing in a medical scenario according to claim 8, wherein the combining the auxiliary suggestion result with a timestamp to obtain the clinical decision support result includes: Generating a risk score according to the event record; generating operation advice according to the risk score and the preset medical guideline; and outputting final clinical intervention advice according to the combination of the operation advice and the feedback execution result.
  10. 10. A medical scene multimodal message recognition processing system, the system comprising: The voice input module is used for dynamically adjusting voice recognition parameters according to the collected voice signals of doctors, the ambient noise level, the personnel position and the activity intensity to obtain voice input data; The text error correction module is used for converting the voice input data into text information according to the voice input data, carrying out semantic alignment on the text information by combining a medical knowledge graph, detecting context semantic conflict and outputting an error correction prompt to obtain corrected text information; The information display module is used for generating structured display content according to the error-corrected text information, and calculating an emergency index based on an emergency label to obtain priority display information; The information feedback module is used for receiving the execution feedback provided by medical staff through touch screen operation or voice instructions according to the priority display information, and if the execution feedback contains abnormal description, the execution feedback is transcribed into characters and marked as the execution abnormality, and meanwhile, an abnormality notification is generated to obtain dynamically updated execution feedback information; The decision support module is used for recording a complete event chain according to the dynamically updated execution feedback information, generating an electronic audit log with a time stamp, and generating clinical decision assistance suggestions by combining the log to obtain a decision support result; And the information processing module is used for triggering a rescheduling mechanism to display the emergency event on the top according to the decision support result if the emergency event corresponding to the priority display information is detected not to be executed within the preset time, so as to ensure the priority transmission and execution of the emergency information.

Description

Medical scene multi-mode message recognition processing method and system Technical Field The invention relates to the technical field of data processing, in particular to a medical scene multi-mode message recognition processing method, a system, electronic equipment and a non-transitory computer readable storage medium. Background Today, in clinical healthcare work scenarios, the transfer of information between healthcare workers relies primarily on verbal communication or telephonic notification. When a patient reflects the illness state or requests the nurse, if the doctor is not present, the nurse usually informs the doctor of the related information through oral transfer or telephone, and similarly, the doctor needs to use a long-term doctor's advice or an urgent temporary doctor's advice issued by a computer, and also usually reminds the doctor to the office nurse through oral or telephone mode, and then prints the doctor's advice to transfer to the clinical nurse. However, in the oral delivery process, errors of oral delivery, hearing errors or memory deviation are very easy to occur, and telephone communication is not effectively recorded, so that traceability is lacking, and if paper recording is adopted, errors of writing or untimely recording are easy to occur. Particularly in emergency situations, delays or distortions in information delivery may directly affect patient safety and medical quality. Disclosure of Invention Aiming at the technical problems in the prior art, the invention provides a medical scene multi-mode message recognition processing method, a system, electronic equipment and a non-transitory computer readable storage medium, which can improve the accuracy of medical scene information transmission. The technical scheme for solving the technical problems is as follows: the invention provides a medical scene multi-mode message recognition processing method, which comprises the following steps: according to the collected voice signals of doctors, the ambient noise level, the personnel position and the activity strength, dynamically adjusting voice recognition parameters to obtain voice input data; According to the voice input data, converting the voice input data into text information, carrying out semantic alignment on the text information by combining a medical knowledge graph, detecting context semantic conflict and outputting an error correction prompt to obtain corrected text information; generating structured display content according to the error-corrected text information, and calculating an emergency index based on an emergency label to obtain priority display information; receiving execution feedback provided by medical staff through touch screen operation or voice instructions according to the priority display information, and if the execution feedback contains abnormal description, converting the execution feedback into characters and marking the characters as the execution abnormality, and generating an abnormality notification to obtain dynamically updated execution feedback information; recording a complete event chain according to the dynamically updated execution feedback information, generating an electronic audit log with a time stamp, and generating clinical decision assistance suggestions by combining the log to obtain a decision support result; And according to the decision support result, if the fact that the emergency event corresponding to the priority display information is not executed within the preset time is detected, triggering a rescheduling mechanism to set the emergency event to be displayed on top, and ensuring that the emergency information is transmitted and executed preferentially. Optionally, the step of dynamically adjusting the voice recognition parameters according to the collected voice signal of the doctor, the ambient noise level, the personnel position and the activity intensity to obtain voice input data includes: Generating a noise suppression factor according to the environmental noise level to obtain a noise reduction voice signal; and generating a voice weight parameter according to the personnel position and the activity intensity, and applying the weight parameter to the noise reduction voice signal to obtain weighted voice input data. Optionally, the step of converting the voice input data into text information according to the voice input data, and performing semantic alignment on the text information by combining a medical knowledge graph, detecting context semantic conflict, and outputting an error correction prompt to obtain the text information subjected to error correction includes: Generating an initial text sequence according to the voice input data; performing context comparison on the initial text sequence according to the medical knowledge graph to obtain a candidate error correction result; And outputting the character information subjected to error correction according to the candidate error correction result and t