CN-122025061-A - Asynchronous intention pre-computing method and system based on diagnosis and treatment context awareness

CN122025061ACN 122025061 ACN122025061 ACN 122025061ACN-122025061-A

Abstract

The application relates to the technical field of medical informatization, and discloses a method and a system for asynchronous intention pre-calculation based on diagnosis and treatment context awareness. The method comprises the steps of collecting multi-mode context data including diagnosis and treatment environment perception data, user interface interaction state data and clinical knowledge data in a diagnosis and treatment scene in real time, comprehensively analyzing the multi-mode context data to generate corresponding context feature characterization, dynamically calculating respective confidence degrees of a plurality of candidate operation intents based on the context feature characterization in an asynchronous process independent of a user interaction main process, and marking the corresponding candidate operation intents as target intents to be executed when the confidence degrees meet preset conditions. According to the application, through multi-mode context sensing and asynchronous intention pre-calculation, the intention pre-judging and pre-processing can be completed before a user sends out an explicit instruction, the response delay of the system is reduced, and the diagnosis and treatment interaction efficiency is improved.

Inventors

QIAN DAHONG
WEN YAOFENG

Assignees

聚交芯创医疗电子(上海)有限公司
聚交元一人工智能科技(武汉)有限公司

Dates

Publication Date: 20260512
Application Date: 20260304

Claims (10)

1. An asynchronous intent pre-computing method based on diagnosis and treatment context awareness, comprising: Acquiring multi-mode context data in a diagnosis and treatment scene in real time, wherein the multi-mode context data at least comprises at least two of diagnosis and treatment environment perception data, user interface interaction state data and clinical knowledge data associated with a current patient; Comprehensively analyzing the multi-mode context data to generate corresponding context feature characterization; in an asynchronous process independent of a user interaction main flow, based on the contextual characteristic characterization, the respective confidence degrees of a plurality of candidate operation intentions in the current diagnosis and treatment scene are dynamically calculated; And when the confidence coefficient of the candidate operation intention meets a preset condition, marking the corresponding candidate operation intention as a target intention to be executed.
2. The method of claim 1, wherein the context awareness data comprises voice-of-diagnosis semantic data, and wherein the performing the comprehensive analysis on the multi-modal context data to generate the corresponding context feature characterization comprises performing a fusion process on the multi-modal context data to generate the unified context feature characterization.
3. The method as recited in claim 1, further comprising: According to the data influence attribute of the operation corresponding to the target intention to be executed, dividing the target intention to be executed into read-only operation and write-in operation, and respectively executing data inquiry on the read-only operation in a background, storing an inquiry result in a buffer area, and simulating execution on the write-in operation in a virtual execution environment isolated from a real service system to generate an operation draft object, wherein the operation draft object is not submitted to the real service system; performing medical rule verification on the operation draft object in the virtual execution environment, and associating a verification result to the operation draft object; When the user's right confirming operation is detected, the right confirming operation is matched with the query result in the buffer area or the operation draft object in the virtual execution environment, wherein the right confirming operation is a confirmatory instruction which is sent by the user through a voice instruction or interface interaction and is pointed to specific business operation, when the matching is successful, the corresponding pre-calculation result is delivered to a user interface for presentation, and when the matching is unsuccessful, the corresponding pre-calculation result is discarded.
4. The method of claim 1, wherein the dynamically calculating comprises reading the latest context feature characterization in each sliding window period with a preset time step as a sliding window period, and outputting respective confidence levels of the plurality of candidate operation intents through an intent recognition model.
5. The method of claim 1, wherein the comprehensively analyzing the multi-modal context data comprises vectorizing the multi-modal context data, performing cross-modal feature fusion through an attention mechanism, and generating a unified high-dimensional context feature tensor as the context feature characterization.
6. The method of claim 1, wherein the multi-modal context data further includes physical environmental acoustic feature data obtained from analysis of room environmental audio for identifying a current type of medical procedure.
7. The method of claim 3, wherein the medical rule check comprises at least one of a medication interaction check, a medication contraindication check, and a medical compliance check, wherein when the matching is successful, the user interface presents the pre-calculation result and simultaneously presents quality control warning information in the check result, and when the matching is unsuccessful, the corresponding pre-calculation result is discarded and returned to a synchronous execution flow.
8. The method of claim 3, further comprising recording interactive feedback actions of a user on pre-computed results generated based on the target intent to be performed, the interactive feedback actions including direct validation, post-modification validation, rejection, or ignore, converting the interactive feedback actions into reinforcement learning signals, and performing parameter updates on an intent recognition model used in the asynchronous process.
9. An asynchronous intent pre-computing system based on clinical context awareness, comprising: a memory for storing computer executable instructions, and A processor coupled to the memory for implementing the steps of the method of any one of claims 1 to 8 when the computer executable instructions are executed.
10. A non-transitory computer readable storage medium having stored therein computer executable instructions which when executed by a processor implement the steps in the method of any one of claims 1 to 8.

Description

Asynchronous intention pre-computing method and system based on diagnosis and treatment context awareness Technical Field The application relates to the technical field of medical informatization, in particular to an asynchronous intention pre-computing technology based on diagnosis and treatment context awareness. Background The statements in this section are intended to provide background information only for embodiments of the present disclosure and are not to be construed as admissions or implications of prior art. With the continuous deep construction of medical informatization, medical institutions commonly deploy a plurality of business information systems to support daily diagnosis and treatment activities. The hospital information system (Hospital Information System, HIS for short) is used for managing basic business processes of patient registration, charging, hospitalization and the like, the electronic medical record system is used for structurally recording medical record information of the patient, and the image archiving and communication system is used for storing and transmitting medical image data. In the clinical diagnosis and treatment process, a doctor usually needs to access the plurality of service systems through a workstation at the same time so as to finish operations such as medical record review, doctor's advice provision, examination result review and the like. In the prior art, the interaction between doctors and the business systems described above relies mainly on graphical user interfaces. The doctor needs to frequently switch between a multi-level menu and a plurality of windows, manually retrieve the historical diagnosis and treatment data of the patient, enter medical record contents item by item and fill in the medical advice form step by step. This interaction mode requires the physician to initiate each step actively, with the system responding only passively to the physician's explicit instructions. Under the high-frequency and fast-paced diagnosis and treatment scenes such as outpatient service, a large amount of system operation consumes time and energy of doctors, and diagnosis and treatment efficiency is reduced. In view of the above problems, a medical auxiliary interaction scheme based on voice recognition appears in the prior art. The typical processing flow of the scheme is that the system waits for a doctor to send out a voice instruction, and after receiving a voice signal, processing steps such as voice text conversion, semantic analysis, database query and the like are sequentially executed, and finally, the result is returned to the interface. However, the flow is still a synchronous serial processing mechanism in nature, i.e., the system must start subsequent processing after receiving an explicit user instruction, with each processing step being performed in sequence, waiting step by step, thereby creating an accumulated delay from the instruction issue to the presentation of the result. In addition, in the prior art, a macro command scheme based on keyword triggering exists, the operation steps are simplified through the mapping relation between the preset keywords and the preset script, but a doctor needs to memorize a specific triggering instruction, and the system can only respond to the preset fixed instruction and cannot process diversified intentions in natural language expression. The prior system is difficult to prejudge the operation intention before a doctor sends out an explicit instruction, passive processing can be started only after the instruction is received, and idle time windows in the diagnosis and treatment process can not be fully utilized to prepare data or operation results which may be needed in advance. Disclosure of Invention The application provides an asynchronous intention pre-computing method and an asynchronous intention pre-computing system based on diagnosis and treatment context awareness, which solve the technical problem of how to pre-judge the operation intention of a user before the user sends out an explicit instruction. The application discloses an asynchronous intention pre-calculation method based on diagnosis and treatment context awareness, which comprises the following steps: Acquiring multi-mode context data in a diagnosis and treatment scene in real time, wherein the multi-mode context data at least comprises at least two of diagnosis and treatment environment perception data, user interface interaction state data and clinical knowledge data associated with a current patient; Comprehensively analyzing the multi-mode context data to generate corresponding context feature characterization; in an asynchronous process independent of a user interaction main flow, based on the contextual characteristic characterization, the respective confidence degrees of a plurality of candidate operation intentions in the current diagnosis and treatment scene are dynamically calculated; And when the confidence coefficient of