CN-122025069-A - Multi-mode sensing and controllable executing isolated liver perfusion system and control method

CN122025069ACN 122025069 ACN122025069 ACN 122025069ACN-122025069-A

Abstract

The application relates to the technical field of medical equipment control, and discloses an isolated liver perfusion system and a control method for multi-mode sensing and controllable execution, which are characterized in that by collecting and preprocessing monitoring data of at least two heterogeneous modes, and (3) reasoning and outputting regulation parameters and uncertainty assessment by using the first machine learning model through the edge intelligent terminal, and uploading the data to a cloud for further processing. The cloud end generates a regulation and control result through the second machine learning model, adjusts the confidence coefficient of the cloud end model based on the communication quality, generates a regulation and control instruction to be executed by combining the local and cloud end results, ensures the safety of the perfusion process by adopting a conserved regulation and control instruction when the network quality is lower than a threshold value, ensures the executable of the regulation and control instruction through safety constraint verification, and records manual intervention information. The method can effectively cope with the emergency such as unstable network, cloud failure and the like while ensuring the stable execution of the isolated liver perfusion process, and has stronger robustness and reliability.

Inventors

ZHANG LEI
HAO HAIZHOU

Assignees

天津市第一中心医院

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (8)

1. The multi-mode sensing and controllable executing in-vitro liver perfusion control method is applied to a system comprising an edge intelligent terminal, a cloud intelligent service platform and perfusion equipment, and is characterized by comprising the following steps: Step S1, at least two heterogeneous mode monitoring data of an isolated liver are collected, and the monitoring data are preprocessed; s2, in the edge intelligent terminal, reasoning the monitoring data by using a first machine learning model, and outputting a local regulation and control parameter result comprising a perfusion state evaluation parameter and a corresponding uncertainty evaluation value; Step S3, the monitoring data are sent to a cloud intelligent service platform, and the cloud intelligent service platform invokes a second machine learning model based on multi-center isolated liver perfusion data training to conduct reasoning, so that a cloud regulation parameter result is generated; Step S4, calculating local confidence coefficient based on the uncertainty evaluation value according to a preset mapping relation, calculating cloud model confidence coefficient based on performance evaluation parameters of the second machine learning model, adjusting the cloud model confidence coefficient according to communication quality parameters of current network connection to obtain cloud confidence coefficient, normalizing the local confidence coefficient and the cloud confidence coefficient to obtain dynamic fusion weight, carrying out weighted fusion on the local regulation parameter result and the cloud regulation parameter result based on the dynamic fusion weight when the communication quality is not lower than a preset threshold value, and generating a regulation instruction to be executed; S5, carrying out safety constraint verification on the to-be-executed regulation instruction or the conservative regulation instruction, and generating a final regulation instruction after verification is passed; And S6, controlling the perfusion equipment to execute the final regulation and control instruction, and recording manual intervention information.
2. The multi-modal sensing and controllably performing ex vivo liver perfusion control method of claim 1, further comprising: And based on multi-center isolated liver perfusion data, when a preset time interval arrives or an uncertainty evaluation value output by the first machine learning model is continuously higher than a preset threshold value in a set time, carrying out parameter updating on the second machine learning model by adopting a federal learning framework, re-executing knowledge distillation based on the updated second machine learning model, generating an updated first machine learning model, and deploying the updated first machine learning model to the edge intelligent terminal.
3. The multi-modal sensing and controllably executing ex vivo liver perfusion control method of claim 1, wherein the heterogeneous modal monitoring data includes imaging modal data and physicochemical parameter modal data; The imaging mode data is a liver microcirculation perfusion image sequence obtained through optical coherence tomography or laser speckle contrast imaging; The physicochemical parameter mode data comprises at least two parameters selected from pH value, dissolved oxygen concentration, glucose concentration, lactic acid concentration, potassium ion concentration, perfusate temperature and perfusate pressure.
4. The multi-modal sensing and controllably executing ex-vivo liver perfusion control method according to claim 1, wherein the first machine learning model is a lightweight neural network model obtained by knowledge distillation of the second machine learning model; And the knowledge distillation process is executed on a cloud intelligent service platform, so that the first machine learning model is simultaneously fitted with the regulation and control parameter distribution and the intermediate characteristic representation output by the second machine learning model.
5. The multi-modal sensing and controllably performed ex-vivo liver perfusion control method according to claim 1, wherein the uncertainty evaluation value is a confidence interval width corresponding to a regulatory parameter output by the first machine learning model; The local confidence is calculated according to a preset mapping relation based on the uncertainty evaluation value and is inversely related to the uncertainty evaluation value; the cloud model confidence is calculated based on performance evaluation parameters of the second machine learning model; adjusting the cloud model confidence based on network round trip delay and data transmission success rate to obtain cloud confidence; and respectively taking the normalized local confidence coefficient and the normalized cloud confidence coefficient as dynamic fusion weights of the local regulation parameter result and the cloud regulation parameter result.
6. An isolated liver perfusion system with multi-modal sensing and controllable execution, comprising: A perfusion apparatus for performing a perfusion operation on the isolated liver; The edge intelligent terminal is configured to acquire at least two heterogeneous mode monitoring data of an isolated liver and perform preprocessing, call a first machine learning model to infer the monitoring data, output a local regulation and control parameter result comprising a perfusion state evaluation parameter and a corresponding uncertainty evaluation value, and calculate local confidence coefficient based on the uncertainty evaluation value according to a preset mapping relation; The cloud intelligent service platform is configured to call a second machine learning model trained based on multi-center isolated liver perfusion data to infer the monitoring data and generate a cloud regulation parameter result; calculating cloud model confidence coefficient based on the performance evaluation parameters of the second machine learning model, and adjusting the cloud model confidence coefficient according to the communication quality parameters of the current network connection to obtain cloud confidence coefficient; The fusion and control module is used for carrying out normalization processing on the local confidence coefficient and the cloud confidence coefficient to obtain a dynamic fusion weight, carrying out weighted fusion on the local regulation parameter result and the cloud regulation parameter result based on the dynamic fusion weight when the communication quality is not lower than a preset threshold value to generate a regulation instruction to be executed, and generating a preset conservation regulation instruction based on the local regulation parameter result when the communication quality is lower than the preset threshold value; The safety verification module is used for carrying out safety constraint verification on the regulation and control instruction to be executed or the conservative regulation and control instruction, and generating a final regulation and control instruction after verification is passed; the edge intelligent terminal is configured to control the perfusion equipment to execute the final regulation and control instruction after the operator confirms the edge intelligent terminal, and records manual intervention information.
7. An electronic device comprising a processor and a memory, the processor being electrically connected to the memory, the memory having stored thereon a computer program which when executed by the processor implements the multi-modal aware and controllable ex vivo liver perfusion control method according to any one of claims 1 to 5.
8. A computer readable storage medium, wherein a program is stored on the computer readable storage medium, and when the program is executed on a computer, the program causes the computer to execute the multi-modal sensing and controllable ex-vivo liver perfusion control method according to any one of claims 1 to 5.

Description

Multi-mode sensing and controllable executing isolated liver perfusion system and control method Technical Field The application relates to the technical field of medical equipment control, in particular to an isolated liver perfusion system capable of being performed in a multi-mode sensing and controllable mode and a control method. Background The isolated liver mechanical perfusion technology can continuously and controllably perfuse and maintain the liver supply in vitro by simulating the physiological environment, can effectively evaluate the vitality of the liver, repair damaged liver cells, and remarkably prolong the safe preservation time of the liver supply, thereby relieving the crisis of organ shortage. Related ex vivo liver perfusion systems often rely on manual regulation empirically by the operator based on limited monitoring parameters such as perfusion pressure, flow, pH, or employ fixed rule based automated controls such as PID controllers. The method is simple to realize, but is difficult to cope with the high dynamic property and individual variability of the liver complex biological system, cannot realize the optimal regulation and control based on the liver real-time physiological state, and has limited control precision and adaptability. With the application of artificial intelligence technology in isolated organ perfusion control, a machine learning model trained based on historical data can learn a complex regulation strategy. The current mainstream AI scheme has two types of limitations, namely, cloud centralized control, namely, high-precision regulation and control can be realized by utilizing strong calculation power and multi-center data, but the current mainstream AI scheme is completely dependent on a network, and has safety risks caused by delay or interruption in a real-time life support scene, and edge-end local control, which has quick response and high reliability, is limited by local resources, has insufficient model performance and generalization capability, and is difficult to continuously utilize the multi-center data for optimization. There is an inherent conflict between accuracy and reliability. Disclosure of Invention The application provides a multi-mode sensing and controllable executing isolated liver perfusion system and a control method, which realize accurate regulation and robust execution of an isolated liver perfusion process on the premise of ensuring perfusion safety by an intelligent control mechanism of multi-mode sensing and cloud edge cooperation and maintain continuity, reliability and adaptability of the system in a complex network environment. In order to achieve the aim, the application adopts the following technical scheme that the isolated liver perfusion system and the control method are performed in a multi-mode sensing and controllable way. A multi-mode sensing and controllable executing isolated liver perfusion control method is applied to a system comprising an edge intelligent terminal, a cloud intelligent service platform and perfusion equipment, and comprises the following steps: Step S1, at least two heterogeneous mode monitoring data of an isolated liver are collected, and the monitoring data are preprocessed; s2, in the edge intelligent terminal, reasoning the monitoring data by using a first machine learning model, and outputting a local regulation and control parameter result comprising a perfusion state evaluation parameter and a corresponding uncertainty evaluation value; Step S3, the monitoring data are sent to a cloud intelligent service platform, and the cloud intelligent service platform invokes a second machine learning model based on multi-center isolated liver perfusion data training to conduct reasoning, so that a cloud regulation parameter result is generated; Step S4, calculating local confidence coefficient based on the uncertainty evaluation value according to a preset mapping relation, calculating cloud model confidence coefficient based on performance evaluation parameters of the second machine learning model, adjusting the cloud model confidence coefficient according to communication quality parameters of current network connection to obtain cloud confidence coefficient, normalizing the local confidence coefficient and the cloud confidence coefficient to obtain dynamic fusion weight, carrying out weighted fusion on the local regulation parameter result and the cloud regulation parameter result based on the dynamic fusion weight when the communication quality is not lower than a preset threshold value, and generating a regulation instruction to be executed; S5, carrying out safety constraint verification on the to-be-executed regulation instruction or the conservative regulation instruction, and generating a final regulation instruction after verification is passed; And S6, controlling the perfusion equipment to execute the final regulation and control instruction, and recording manual intervention information. Fur