CN-122025174-A - Intraoperative information prompting method, device, equipment and medium

Abstract

The invention discloses an intraoperative information prompting method, device, system and medium, and belongs to the technical field of medical assistance. The method comprises the steps of obtaining various input features, carrying out statistics feature aggregation on data in a time window according to window refreshing periods respectively corresponding to the input features aiming at each type of input features so as to update a target context, carrying out semantic processing on the target context when receiving triggering operation of information prompt, obtaining standard semantic representation, carrying out semantic routing based on the standard semantic representation, and activating a semantic prompt submodel matched with the current operation intention to obtain an information prompt result in operation. The scheme ensures that the neural network model can more easily identify the truly effective information, and improves the accuracy of information prompt of the operation auxiliary system.

Inventors

• YAO ZHIYUAN
• WANG QIUGE

Assignees

• 江苏圆和医疗科技有限公司

Dates

Publication Date

20260512

Application Date

20260130

Claims (10)

1. 1. An intraoperative information prompting method, the method comprising: The method comprises the steps of acquiring various input features, wherein the input features are obtained by extracting features of data to be identified according to the data type of the data to be identified, and the data to be identified comprises at least one of historical case data of a target object and sensor data, wherein the sensor data are acquired through a sensor in operation; aiming at each type of input features, according to window refreshing periods respectively corresponding to the input features, carrying out statistical feature aggregation on data in a time window so as to update a target context; when receiving triggering operation of information prompt, carrying out semantic processing on the target context to obtain standard semantic representation; And carrying out semantic routing based on the standard semantic representation, and activating a semantic prompt sub-model matched with the current surgical intention to obtain an intraoperative information prompt result.
2. 2. The method of claim 1, wherein before the statistics feature aggregation is performed on the data in the window according to the window refresh periods respectively corresponding to the input features, the method further comprises: And performing autoregressive analysis according to the sampling interval and the historical change sequence of the input feature, and determining the window refreshing period of the input feature.
3. 3. The method according to claim 2, wherein the step of performing statistical feature aggregation on the data in the window according to the window refresh periods respectively corresponding to the input features to update the target context includes: performing time aggregation processing on the input features in a time window to obtain a statistical aggregation amount, wherein the statistical aggregation amount comprises at least one of an average value, a standard deviation, a change slope and inflection point times of the input features; If the update time difference of the input feature is greater than or equal to the window refreshing period, determining a time mask of the input feature as 1, otherwise, determining the time mask as 0; If the time mask of the input feature is 1, updating the target context based on the statistical aggregation amount; if the time mask of the input feature is 0, the target context is not updated.
4. 4. A method according to claim 3, wherein said performing semantic processing on said target context upon receipt of a triggering operation of an information prompt to obtain a standard semantic representation comprises: when receiving triggering operation of information prompt, generating a query vector according to the triggering operation; Generating a key vector and a value vector according to the target context; Obtaining a scaling matrix through scaling processing after carrying out operation according to the key vector and the value vector; determining a time mask corresponding to each element in the scaling matrix according to the input characteristic type corresponding to each element in the scaling matrix; Processing the scaling matrix according to the time masks corresponding to the elements to obtain a masked matrix so as to determine whether the elements corresponding to the input features in the scaling matrix participate in operation or not; And carrying out operation according to the masked matrix and the value vector to obtain the standard semantic representation.
5. 5. The method according to claim 4, wherein said processing the scaling matrix according to the time mask corresponding to each element to obtain a masked matrix comprises: adding each element in the scaling matrix with a corresponding time mask value to obtain a masked matrix; If the time mask of the input feature corresponding to the element is 0, the time mask value corresponding to the element is minus infinity; and if the time mask of the input feature corresponding to the element is 1, the time mask value corresponding to the element is 0.
6. 6. The method according to claim 4, wherein said processing the scaling matrix according to the time mask corresponding to each element to obtain a mask matrix comprises: if the corresponding time mask of the element in the scaling matrix is 1, reserving the element; And if the time mask corresponding to the element in the scaling matrix is 0, masking the element so that the element does not participate in attention calculation.
7. 7. The method according to claim 4, wherein said processing the scaling matrix according to the mask corresponding to each element to obtain a mask matrix includes: acquiring a row index and a column index of the scaling matrix, wherein the row index of the scaling matrix corresponds to each query vector; Mapping the row index and the column index to each node in a predefined graph structure, wherein the predefined graph structure comprises query vectors of different types and nodes corresponding to key vectors, and the connection relation of each node is matched with the association relation of the query vectors and the key vectors in an input feature level; generating a graph constraint mask corresponding to each element according to the connection relation among the nodes; and determining whether the elements in the scaling matrix are shielded according to the time masks and the graph constraint masks corresponding to the elements so as to obtain the mask matrix.
8. 8. An intraoperative information prompting system, the system comprising: The sensing module is used for acquiring various input features, wherein the input features are obtained by extracting features of data to be identified according to the data type of the data to be identified, and the data to be identified comprises at least one of historical case data of a target object and sensor data, and the sensor data is acquired through a sensor in operation; The semantic processing module is used for carrying out statistic feature aggregation on data in a time window according to window refreshing periods respectively corresponding to the input features aiming at each type of input features so as to update a target context; the semantic processing module is used for carrying out semantic processing on the target context when receiving the triggering operation of the information prompt to obtain standard semantic representation; And the decision module is used for carrying out semantic routing based on the standard semantic representation, activating a semantic prompt sub-model matched with the current surgical intention, and obtaining an intraoperative information prompt result.
9. 9. An electronic device, comprising: a memory and a processor, the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, thereby executing the intra-operative information prompting method according to any one of claims 1 to 7.
10. 10. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the intra-operative information prompting method according to any one of claims 1 to 7.

Description

Intraoperative information prompting method, device, equipment and medium Technical Field The present invention relates to the technical field of medical assistance, and in particular, to a method, an apparatus, a device, and a medium for prompting information in surgery. Background At present, as the application of the neural network model in medical scenes is continuously expanding, the neural network model system gradually becomes an important tool for assisting doctors in diagnosis, preoperative evaluation and postoperative follow-up. Information interaction in the surgical operating room is also developing in an intelligent direction. The prior augmented reality operation navigation system has been tried to embed the neural network model into the operation auxiliary system, so that a surgeon can realize the processing identification of the information in the operation through the operation auxiliary system constructed by the neural network model, even predict the operation steps of a follow-up operator, and assist the surgeon in the operation process. For example, the operation assisting system can acquire relevant data of a patient in real time through a sensor and comprehensively process the relevant data and the historical cases of the patient, so that prompt information including patient information or operation advice and the like is output. However, in the surgical procedure, the surgical assistance system may collect a large amount of patient data of different types, and in addition to the historical cases of the patient, the calculation amount is too large, and the neural network model has difficulty in identifying truly effective information in a large amount of data, resulting in poor effect of the surgical assistance system. Disclosure of Invention In view of the foregoing, it is desirable to provide a method, apparatus, device, and medium for information presentation during surgery, which improve the accuracy of information presentation for the surgery assistance system. The application provides an intraoperative information prompting method, which comprises the following steps: The method comprises the steps of acquiring various input features, wherein the input features are obtained by extracting features of data to be identified according to the data type of the data to be identified, and the data to be identified comprises at least one of historical case data of a target object and sensor data, wherein the sensor data are acquired through a sensor in operation; aiming at each type of input features, according to window refreshing periods respectively corresponding to the input features, carrying out statistical feature aggregation on data in a time window so as to update a target context; when receiving triggering operation of information prompt, carrying out semantic processing on the target context to obtain standard semantic representation; And carrying out semantic routing based on the standard semantic representation, and activating a semantic prompt sub-model matched with the current surgical intention to obtain an intraoperative information prompt result. In an optional implementation manner, before the data in the window is aggregated according to the window refresh periods corresponding to the input features, the method further includes: And performing autoregressive analysis according to the sampling interval and the historical change sequence of the input feature, and determining the window refreshing period of the input feature. In an optional implementation manner, the step of performing statistics feature aggregation on the data in the window according to the window refresh periods respectively corresponding to the input features to update the target context includes: performing time aggregation processing on the input features in a time window to obtain a statistical aggregation amount, wherein the statistical aggregation amount comprises at least one of an average value, a standard deviation, a change slope and inflection point times of the input features; If the update time difference of the input feature is greater than or equal to the window refreshing period, determining a time mask of the input feature as 1, otherwise, determining the time mask as 0; If the time mask of the input feature is 1, updating the target context based on the statistical aggregation amount; if the time mask of the input feature is 0, the target context is not updated. In an optional implementation manner, when receiving the triggering operation of the information prompt, performing semantic processing on the target context to obtain a standard semantic representation, including: when receiving triggering operation of information prompt, generating a query vector according to the triggering operation; Generating a key vector and a value vector according to the target context; Obtaining a scaling matrix through scaling processing after carrying out operation according to the key vector and the