CN-122025013-A - Soft tissue traction intelligent decision method, electronic equipment and storage medium

CN122025013ACN 122025013 ACN122025013 ACN 122025013ACN-122025013-A

Abstract

The invention discloses a soft tissue traction intelligent decision method for a sparse small sample, and relates to the technical fields of intelligent decision of surgical robots and medical data processing. The method comprises the steps of S1 constructing a state transition model of soft tissue stripping, establishing an energy conservation relation and a geometric constraint relation, S2 obtaining sparse demonstration data aiming at soft tissues, performing time domain complementation by using an interpolation algorithm to generate an initial continuous demonstration track, S3 constructing an objective function, calculating gradient information by using the state transition model, performing inverse gradient propagation optimization on an action sequence, and S4 performing imitation learning by using optimized track data to obtain an intelligent decision strategy. The invention aims to solve the problems that the existing operation demonstration data are sparse and the sample size is small, so that a high-quality strategy cannot be trained, and high-quality complete data are generated through physical modeling and track optimization, so that the operation robot can stably and softly pull and guide soft tissues.

Inventors

Qin Jiahu
LI YACHEN
LI MAN
LIU QINGCHEN
ZHU ZHIQIANG
TANG QINQING
WU CHUANQING

Assignees

中国科学技术大学

Dates

Publication Date: 20260512
Application Date: 20260120

Claims (12)

1. An intelligent decision-making method for soft tissue traction is characterized by comprising the following steps: the method comprises the steps of S1, constructing a state transfer model of soft tissue stripping, defining an action variable describing the motion of an end effector of a mechanical arm and a state variable describing the interaction state of the soft tissue and the mechanical arm, establishing an energy conservation relation for representing the balance between the traction acting of the mechanical arm and the adhesion breaking energy of the soft tissue, establishing a geometric constraint relation for representing the geometric recursion between the position change of the end effector of the mechanical arm and the stripping progress of the soft tissue, and determining the state transfer model based on the energy conservation relation and the geometric constraint relation; s2, acquiring initial continuous demonstration tracks, acquiring sparse demonstration data aiming at soft tissues, and performing time domain completion on the sparse demonstration data by utilizing an interpolation algorithm to generate initial tracks comprising continuous state sequences and continuous action sequences; S3, generating an optimized track based on reverse gradient propagation, constructing an objective function for guiding the optimization of the traction track, calculating gradient information of the objective function on action variables in the initial track by using the state transition model, and iteratively updating the action sequence by using the gradient information until a preset convergence condition is met to obtain the optimized track; and S4, training a soft tissue traction strategy, and performing imitation learning by utilizing the optimized track to obtain an intelligent decision strategy capable of outputting traction action according to a real-time state.
2. The method according to claim 1, wherein in step S1, the motion variable is And the state variable Is defined as: Wherein, the The time of day is indicated as such, And Respectively the end effectors of the mechanical arm The displacement component of the moment of time, And Is that The position coordinates of the point are grasped at the moment, In order to have a length of tissue that has been stripped, Is a stripped part and Included angle of the coordinate axes.
3. The method according to claim 2, wherein in step S1, the energy conservation relationship is expressed as: Wherein, the Work is done for the traction force of the mechanical arm end effector, The energy required for breaking the adhesion between the attachment points is calculated based on the energy conservation relationship Time-to-time peel length variation The formula of (2) is: Wherein, the In order to be able to draw the force, For the width of the surface of the soft tissue to be peeled in the direction perpendicular to the peeling direction at the current peeling point, The work required for adhesion failure per unit area.
4. A soft tissue traction intelligent decision method according to claim 3 wherein in step S1, the geometric constraint relationship satisfies the following set of equations: Wherein, the Is that The angle of the tangential direction of the soft tissue surface at the position of the peeling point at the moment, For the next time the part is stripped off Included angle of the coordinate axes.
5. The intelligent decision making method for soft tissue traction according to claim 4, wherein the angle of the tangential direction of the soft tissue surface The calculation formula of (2) is as follows: 。
6. The method according to claim 5, wherein in step S1, the state transition model is used for angular state The updated formula of (2) is: the system state transition equation is expressed as In which the position state Updated by linear superposition, length status From claim 3 Updating.
7. The soft tissue traction intelligent decision method according to claim 1, wherein in the step S2, the specific method for performing time domain completion on the sparse demonstration data by using an interpolation algorithm is that a cubic spline interpolation is adopted on a position variable to ensure second-order continuous conduction of an interpolation function at interpolation time, continuous linear interpolation is adopted on a stripped length variable, and continuous linear interpolation after angular expansion is adopted on a direction variable.
8. The method according to claim 6, wherein in step S3, the method comprises the steps of Time objective function Comprising a first sub-target Second sub-target And a third sub-target The calculation formula is as follows: Wherein, the As the weight coefficient of the light-emitting diode, For characterizing the progress of the completion of the stripping task, For characterizing the compliance of the execution, For the reference of the peeling rate, For characterizing the operating space constraints.
9. The method according to claim 8, wherein in step S3, the length is The objective function of the sample trajectory of (a) is: Gradient information of objective function on motion variables The calculation formula of (2) is as follows: Wherein, the As intermediate variables, by time reversal recursive calculation: And 。
10. The intelligent decision-making method according to claim 1, wherein in step S4, the simulated learning adopts a behavior cloning method in the form of supervised learning, and the neural network is trained to approximate a strategy function by minimizing an error function To train, the error function Expressed as: Wherein, the And Respectively the states and actions in the optimized trajectory, Is a policy network.
11. An electronic device comprising a memory for storing a computer program and a processor, characterized in that the processor runs the computer program to cause the electronic device to perform the soft tissue pulling intelligent decision method according to claim 1.
12. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the soft tissue pulling intelligent decision method according to claim 1.

Description

Soft tissue traction intelligent decision method, electronic equipment and storage medium Technical Field The invention relates to the technical field of intelligent decision making and medical data processing of surgical robots, in particular to an intelligent decision making method for soft tissue traction, electronic equipment and a storage medium. Background The surgical robot system plays an important role in complex operations such as tumor excision, digestive tract reconstruction and the like by virtue of the high precision, high flexibility and anti-fatigue characteristics. Although the existing surgical robot can assist doctors to finish fine operations such as tissue separation and suturing, the existing surgical robot still depends on the whole-course planning and real-time control of doctors to a great extent, and the cognition and operation load of the doctors cannot be effectively reduced. With the development of artificial intelligence technology, the simulation learning technology is utilized to enable a robot to learn an operation strategy from the demonstration data of doctors, so that the simulation learning technology becomes an important way for improving the intelligent decision-making capability of the surgical robot. However, in a practical soft tissue surgical scene such as endoscopic mucosal resection, there are many cases where doctor operation data acquired by a visual sensor are discontinuous and have a large amount of noise due to objective factors such as blood exudation, smoke shielding, and endoscopic visual field narrowing. Moreover, the high quality expert demonstration data that can be obtained is extremely scarce, with the sample size generally small, due to limitations of medical ethics, patient privacy protection, and non-repeatability of the surgical procedure. The current situation that the data is sparse and the sample size is small, which results in that the existing imitation learning algorithm is difficult to train a stable and reliable strategy model, and even erroneous decision guidance can be generated due to insufficient data coverage. Therefore, how to extract effective information from sparse and small-scale clinical operation data and construct a high-quality demonstration track to support training of simulating a learning model, so as to realize safe and flexible soft tissue traction intelligent decision, and become a technical problem to be solved urgently. Disclosure of Invention The invention mainly aims to provide an intelligent decision-making method for soft tissue traction, electronic equipment and a storage medium, and aims to extract effective information from sparse and small-scale clinical operation data, construct a high-quality demonstration track to support training imitating a learning model, and thus realize safe and flexible intelligent decision-making for soft tissue traction. In order to achieve the above purpose, the invention provides an intelligent decision method for soft tissue traction, which comprises the following steps: the method comprises the steps of S1, constructing a state transfer model of soft tissue stripping, defining an action variable describing the motion of an end effector of a mechanical arm and a state variable describing the interaction state of the soft tissue and the mechanical arm, establishing an energy conservation relation for representing the balance between the traction acting of the mechanical arm and the adhesion breaking energy of the soft tissue, establishing a geometric constraint relation for representing the geometric recursion between the position change of the end effector of the mechanical arm and the stripping progress of the soft tissue, and determining the state transfer model based on the energy conservation relation and the geometric constraint relation; s2, acquiring initial continuous demonstration tracks, acquiring sparse demonstration data aiming at soft tissues, and performing time domain completion on the sparse demonstration data by utilizing an interpolation algorithm to generate initial tracks comprising continuous state sequences and continuous action sequences; S3, generating an optimized track based on reverse gradient propagation, constructing an objective function for guiding the optimization of the traction track, calculating gradient information of the objective function on action variables in the initial track by using the state transition model, and iteratively updating the action sequence by using the gradient information until a preset convergence condition is met to obtain the optimized track; and S4, training a soft tissue traction strategy, and performing imitation learning by utilizing the optimized track to obtain an intelligent decision strategy capable of outputting traction action according to a real-time state. Preferably, in step S1, the action variableAnd the state variableIs defined as: Wherein, the The time of day is indicated as such,AndRespectively the e