Intelligent control method and device for otitis media

Abstract

The invention relates to an intelligent otitis media regulation and control method and device. The method comprises the steps that a negative electrode needle and a positive electrode needle of otitis media regulation and control equipment are connected to an intelligent regulation and control interface module; the negative electrode needle is guided by an endoscope to enter the nasal cavity of the subject; an endoscopic image is analyzed through an MSFFN positioning algorithm and a DAN model of the intelligent regulation and control interface module, preliminary image recognition is performed through the MSFFN positioning algorithm to capture multi-scale features, and treatment positions of a negative electrode needle and a positive electrode needle are learned through the DAN model; the negative electrode needle and the positive electrode needle are controlled by the intelligent treatment control interface module to be inserted into a predetermined treatment position learned by the DAN model after being in place; connecting the positive electrode needle to otitis media regulation and control equipment; and the electrode needle is positioned and adjusted through an ARNN treatment control model of the intelligent treatment control interface module. The electrode needle is intelligently controlled and adjusted through the intelligent treatment control interface module and the ARNN treatment control model, and treatment individuation and effectiveness are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of auxiliary control of medical devices, in particular to a middle ear inflammation intelligent regulation method and device and computer equipment. BACKGROUND

[0002] Traditional middle ear inflammation treatment methods mainly include drug treatment (antibiotics, anti-inflammatory drugs) and surgical treatment (tympanic membrane puncture, tympanic cavity tube placement). Among them, long-term use of antibiotics in drug treatment may lead to drug resistance and is not effective for some types of middle ear inflammation. Surgical operations such as tympanic membrane puncture and tympanic cavity tube placement have high precision requirements and are prone to cause damage and poor patient experience. In addition, since the operation depends on the experience of doctors, the effect varies from person to person, and there is a certain risk and uncertainty.

[0003] With the development of medical technology, the method of using physical energy regulation for treatment may become a potential means for middle ear inflammation treatment, but this way lacks real-time monitoring and adjustment of the treatment process, and it is difficult to carry out personalized treatment according to individual differences of patients.

[0004] To solve the above problems, the present application provides a middle ear inflammation intelligent regulation method, which realizes intelligent control and adjustment of electrode needles through an intelligent treatment control interface module and an ARNN treatment control model, and improves the individualization and effectiveness of treatment. SUMMARY

[0005] In view of the above problems, the present application provides a middle ear inflammation intelligent regulation method and device and computer equipment.

[0006] According to one aspect of the present application, a middle ear inflammation intelligent regulation method is provided, comprising:

[0007] connecting a negative electrode needle and a positive electrode needle of a middle ear inflammation regulation device to an intelligent regulation interface module;

[0008] guiding the negative electrode needle into the nasal cavity of a subject by an endoscope;

[0009] real-time analyzing endoscope images by an MSFFN positioning algorithm and a DAN model of the intelligent regulation interface module, wherein preliminary image recognition is performed by the MSFFN positioning algorithm to capture multi-scale features, and a treatment position of the negative electrode needle and the positive electrode needle is learned by the DAN model;

[0010] controlling the negative electrode needle and the positive electrode needle to be inserted into the predetermined treatment position learned by the DAN model after being positioned by an intelligent treatment control interface module;

[0011] The positive electrode needle is connected to the otitis media regulation device to establish a treatment circuit; the received real-time biological feedback signal is adjusted by the ARNN treatment control model of the intelligent treatment control interface module to adjust the position of the electrode needle.

[0012] In an alternative way, the otitis media regulation device comprises an electrode needle assembly, an intelligent regulation interface module, an intelligent treatment control interface module, an endoscope, and a data recording module.

[0013] The electrode needle assembly comprises a negative electrode needle and a positive electrode needle for establishing a treatment circuit; the endoscope is used to guide the negative electrode needle into the nasal cavity and provide real-time images; the data recording module is used to record the position of the electrode needle, the treatment current, the treatment time, and the biological feedback signal during treatment.

[0014] In an alternative way, the otitis media regulation device further comprises an electrode needle pressure monitoring module, which monitors the contact pressure between the electrode needle and the nasal cavity tissue in real time through a pressure sensor and performs pressure feedback control.

[0015] The formula for pressure feedback control is:

[0016]

[0017] where u(t) is the control output, in millinewtons; e(t) is the deviation between the set pressure and the actual pressure, in Pa; K p is the proportional coefficient, in mN / Pa; K i is the integral coefficient, in mN / (Pa×s); K d is the differential coefficient, in mN×s / Pa; t is time, in s.

[0018] In an alternative way, the ARNN treatment control model comprises an adaptive adjustment layer based on Lyapunov stability theory, and the control formula of the adaptive adjustment layer is:

[0019]

[0020] where x is the system state vector; A is the system matrix; B is the input matrix; K e is the diagonal gain matrix; δ(t) is the nonlinear disturbance term considering tissue deformation; D is the disturbance distribution matrix.

[0021] In an alternative way, the diagonal gain matrix is updated by solving the Riccati equation, which is:

[0022] A T P+PA-PBR -1 BT P+Q=0

[0023] Wherein, P is a positive definite solution matrix; Q, R are preset weight matrices.

[0024] In an optional manner, the preliminary image recognition by the MSFFN positioning algorithm to capture multi-scale features further comprises:

[0025] The MSFFN algorithm adopts a hierarchical convolution structure to process endoscopic images in parallel through shallow and deep convolution kernels;

[0026] The MSFFN positioning algorithm constructs a feature pyramid layer through a top-down path and a lateral connection, wherein the 5th layer feature map is up-sampled to the 3rd layer resolution through bilinear interpolation and is added element by element with the 3rd layer feature map; each layer of the feature pyramid independently sets a detection head to output optimal treatment position coordinates through a non-maximum suppression merging redundancy method, each detection head includes a classification branch and a regression branch, the classification branch uses a 3x3 convolution to generate a target confidence map to identify an electrode needle puncturable area, and the regression branch is used to predict a bounding box offset to locate a treatment target point;

[0027] The MSFFN positioning algorithm includes a scale perception module that dynamically adjusts feature fusion weights according to changes in the endoscopic field of view, wherein shallow features are automatically enhanced when the distance between the electrode needle and the target point is less than or equal to 5mm.

[0028] In an optional manner, the learning of the treatment positions of the negative electrode needle and the positive electrode needle by the DAN model further comprises:

[0029] The DAN model receives a contact force distribution map from an electrode needle pressure monitoring module, performs spatio-temporal alignment through a spatial transformation network, and constructs a four-dimensional feature tensor containing anatomical structures, instrument states, and tissue mechanical properties; wherein, in the encoder stage, the DAN model adopts a parallel channel attention module and a spatial attention module; the spatial attention module generates a spatial weight map to focus on a 2mmx2mm area around the electrode needle tip; in the decoder stage, a ConvLSTM unit is used to establish a time series memory of the treatment area to fuse the current frame features with the hidden states of the previous 5 frames, and learn the tissue deformation pattern during the electrode needle puncture process;

[0030] An output layer with a double-branch structure is used to generate a Gaussian heat map, wherein the peak value corresponds to the optimal treatment point coordinates.

[0031] In an optional manner, the control of the negative electrode needle and the positive electrode needle to the predetermined treatment position learned by the DAN model after being in place by the intelligent treatment control interface module further comprises:

[0032] The six-degree-of-freedom pose of the electrode needle is decomposed into a translational subspace and a rotational subspace; wherein the translational subspace is quickly aligned through a proportional-derivative controller, and the rotational subspace compensates for attitude deviation caused by the curved channel of the nasal cavity through an adaptive sliding mode algorithm; the control law of the adaptive sliding mode algorithm is:

[0033]

[0034] Wherein τ is the torque vector; Δx and are position error and velocity error respectively; K pe and K de are diagonal gain matrices; λ is the switching gain of the sliding surface; η is the adaptive compensation coefficient; s is the sliding function;

[0035] When the distance between the electrode needle tip and the treatment area is less than or equal to 2mm, switch to a force-position hybrid control mode and adjust the contact force and position stiffness; wherein the equation for adjusting the contact force and position stiffness is:

[0036]

[0037] Wherein M d , D d , K d are the desired inertia, damping, and stiffness matrices respectively; Λ is a diagonal positive definite adjustment matrix; F ext is the actual contact force; F des is the set force trajectory;

[0038] According to the prior knowledge of the subject, the safety boundary constraint function dynamically limits the movement range of the electrode needle, and according to the deviation between the predicted position of the DAN model and the actual puncture point, the pose error is estimated in real time, and when the pose error exceeds a preset threshold, the B-spline curve interpolation is triggered to locally correct the electrode needle path.

[0039] According to another aspect of the present application, an intelligent control device for otitis media is provided, comprising:

[0040] A connection module for connecting the negative electrode needle and the positive electrode needle of the otitis media control device to the intelligent control interface module;

[0041] A guide module for guiding the negative electrode needle into the nasal cavity of the subject using an endoscope;

[0042] An image analysis module for real-time analysis of endoscopic images through the MSFFN positioning algorithm and the DAN model of the intelligent control interface module, wherein the MSFFN positioning algorithm is used for preliminary image recognition to capture multi-scale features, and the DAN model is used to learn the treatment position of the negative electrode needle and the positive electrode needle;

[0043] a control module, configured to control the negative electrode needle and the positive electrode needle to be inserted into the predetermined treatment position learned by the DAN model after the negative electrode needle and the positive electrode needle are in place through the intelligent treatment control interface module;

[0044] a positioning adjustment module, configured to connect the positive electrode needle to the otitis media regulation device to establish a treatment loop, and to perform positioning adjustment of the electrode needle through the ARNN treatment control model of the intelligent treatment control interface module according to the received real-time biological feedback signal.

[0045] According to another aspect of the present application, a computer device is provided, comprising a processor, a memory, a communication interface and a communication bus, the processor, the memory and the communication interface being capable of communicating with each other through the communication bus;

[0046] The memory is configured to store at least one executable instruction, and the executable instruction is configured to enable the processor to perform the operation corresponding to the above-mentioned intelligent regulation method for otitis media.

[0047] According to the scheme provided in the present application, the negative electrode needle and the positive electrode needle of the otitis media regulation device are connected to the intelligent regulation interface module; the negative electrode needle is guided into the nasal cavity of the subject by using an endoscope; the endoscope image is analyzed in real time by using the MSFFN positioning algorithm and the DAN model of the intelligent regulation interface module, wherein the MSFFN positioning algorithm is used to perform preliminary image recognition to capture multi-scale features, and the DAN model is used to learn the treatment position of the negative electrode needle and the positive electrode needle; the negative electrode needle and the positive electrode needle are controlled to be inserted into the predetermined treatment position learned by the DAN model after the negative electrode needle and the positive electrode needle are in place through the intelligent treatment control interface module; the positive electrode needle is connected to the otitis media regulation device to establish a treatment loop; and the positioning of the electrode needle is adjusted by using the ARNN treatment control model of the intelligent treatment control interface module according to the received real-time biological feedback signal. The intelligent control and adjustment of the electrode needle are realized by using the intelligent treatment control interface module and the ARNN treatment control model, and the individualization and effectiveness of the treatment are improved.

[0048] The above description is only a summary of the technical scheme of the present application, in order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0049] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a better understanding of the preferred embodiments, and are not intended to be a limitation on the application. Moreover, in the drawings, like reference numerals denote same or similar components. In the drawings:

[0050] Figure 1 A flowchart of the middle ear inflammation intelligent regulation method according to an embodiment of the application is shown in FIG. 1.

[0051] Figure 2 A middle ear inflammation intelligent regulation diagram according to an embodiment of the application is shown in FIG. 2.

[0052] Figure 3 A frame diagram of the middle ear inflammation intelligent regulation device according to an embodiment of the application is shown in FIG. 3.

[0053] Figure 4 A structure diagram of the computer device according to an embodiment of the application is shown in FIG. 4. DETAILED DESCRIPTION

[0054] Exemplary embodiments of the application will be described herein below with reference to the accompanying drawings. Although exemplary embodiments of the application are shown in the drawings, it should be understood that the application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thoroughly and completely understood, and so that the scope of the application will be fully conveyed to those skilled in the art.

[0055] Figure 1 A flowchart of the middle ear inflammation intelligent regulation method according to an embodiment of the application is shown in FIG. 1. Figure 1 As shown in FIG. 1, the method comprises the following steps:

[0056] In step S101, the negative electrode needle and the positive electrode needle of the middle ear inflammation regulation device are connected to the intelligent regulation interface module.

[0057] In this embodiment, the intelligent regulation interface module manages the connection and signal transmission of the electrode needles, allowing different specifications of electrode needles (such as adjustable diameter and length) to be connected, adapting to the structural differences of different patients, reducing the complexity of operation, and avoiding operation errors caused by inconsistent device interfaces in middle ear inflammation treatment.

[0058] Specifically, the wire ends of the negative and positive electrode needles are inserted into the corresponding slots of the intelligent control interface module, and the impedance of the electrode needles is automatically detected by the microprocessor (normal range: 50Ω-200Ω). If the impedance is abnormal (short circuit or open circuit), an alarm prompt is triggered by the data recording module. The connection state of the electrode needles (contact pressure, integrity of the current path) is fed back to the intelligent treatment control interface module in real time. During initialization, the initial parameters of the electrode needles (default treatment current: 1 mA, treatment time: 10 minutes) are set. Synchronization with the endoscope image processing module is achieved through CAN bus or USB, ensuring that the coordinate systems of the electrode needle position and the endoscopic field of view are aligned. A preset safety boundary constraint function (the movement range of the electrode needle is limited within 5 mm around the eardrum) is loaded to prevent misoperation. The electrode needle is driven to move in small steps (e.g., ±0.1 mm step), and the contact force change is monitored by the pressure sensor to verify the contact stability of the electrode needle and the nasal tissue.

[0059] In an alternative mode, the otitis media control device comprises an electrode needle assembly, an intelligent control interface module, an intelligent treatment control interface module, an endoscope, and a data recording module.

[0060] The electrode needle assembly comprises a negative electrode needle and a positive electrode needle for establishing a treatment circuit. The endoscope is used to guide the negative electrode needle into the nasal cavity and provide real-time images. The data recording module records the electrode needle position, treatment current, treatment time, and biological feedback signals during treatment.

[0061] In this embodiment, the positive and negative electrode needles can accurately act on the target area, reducing damage to the surrounding tissue. Combined with real-time observation of the electrode needle position in the nasal cavity by the endoscope, especially in complex structures, the safety of the operation is improved, and the risk of injury to important nerves and blood vessels in the nasal cavity is avoided. The combination of the intelligent control interface module and the intelligent treatment control interface module enables the device to have image recognition, path planning, automatic positioning, and control functions, reducing the dependence on the experience of the doctor.

[0062] Specifically, the electrode needle assembly uses micro-electrode needles made of biocompatible materials, the positive electrode needle is a surface electrode attached to the patient's skin, and the endoscope uses a high-definition endoscope that can navigate flexibly in the nasal cavity. The intelligent control interface module connects the electrode needle, endoscope, and intelligent treatment control interface module, integrating an image acquisition card, a signal amplifier, and a data converter. The intelligent treatment control interface module integrates an image processing unit (GPU or FPGA), a control unit (microprocessor), and a motor driver, which are used to run the MSFFN positioning algorithm and the DAN model, enabling automatic positioning and control of the electrode needle. The data recording module records images, electrode needle position, current, time, and biological feedback signals during treatment.

[0063] As Figure 2As shown, a male patient (50 years old) received treatment for chronic suppurative otitis media. The negative electrode needle (diameter 0.8 mm, length 15 mm) and the positive electrode needle (diameter 1.2 mm, length 20 mm) were inserted into the slot of the intelligent control interface module. The interface module automatically detected that the negative electrode needle impedance was 85Ω and the positive electrode needle impedance was 110Ω, both within the normal range, with a bright green light. The treatment current was set to 1.5 mA and the treatment time was set to 15 minutes through the touch screen, and the interface module synchronized the parameters to the ARNN treatment control model. The endoscope displayed the image of the tympanic membrane area, and the MSFFN algorithm initially identified 3 candidate treatment positions (coordinates A(120,80), B(150,90), and C(135,75), respectively). The DAN model combined electrode needle pressure monitoring data (contact force 50 mN) and endoscopic images to predict the optimal treatment point as B(150,90). The intelligent treatment control interface module drove the electrode needle to move to a certain point (such as point B), compensated for the 0.5 mm displacement deviation caused by the nasal curvature through the adaptive sliding mode algorithm, and switched to the force-position hybrid control mode when the electrode needle tip was 2 mm away from point B, with the contact force stabilized at 60 mN to establish the treatment loop. The data recording module saved the electrode needle position (X=150.2 mm, Y=89.8 mm), current curve (1.5 mA±0.1 mA), and biological feedback signal (tissue impedance change ±5%) throughout the treatment process. Through the connection and real-time interaction of the intelligent control interface module, the treatment time was shortened by 20%, the electrode needle positioning accuracy was improved to ±0.2 mm, and the complication rate was reduced to 1.5% (traditional method 5%).

[0064] In an alternative way, the otitis media regulation device further comprises an electrode needle pressure monitoring module, which monitors the contact pressure of the electrode needle and the nasal cavity tissue in real time through a pressure sensor and performs pressure feedback control;

[0065] wherein the formula of pressure feedback control is:

[0066]

[0067] wherein u(t) is the control output, in millinewtons; e(t) is the deviation between the set pressure and the actual pressure, in Pa; K p is the proportional coefficient, in mN / Pa; K i is the integral coefficient, in mN / (Pa×s); K d is the differential coefficient, in mN×s / Pa; t is time, in s.

[0068] In this embodiment, the contact force data of the electrode needle and the nasal tissue is collected in real time by the pressure sensor, and the PID control algorithm is combined to dynamically adjust the electrode needle position or treatment parameters, to ensure that the contact pressure is stable within the preset safety threshold (50-150 mN), to avoid tissue damage (such as mucosal tearing) caused by pressure fluctuations in traditional treatment, and to reduce the risk of postoperative complications. The pressure feedback control compensates for the influence of nasal structure differences (such as nasal septum deviation) on electrode needle positioning, for example, when the electrode needle contact force fluctuates due to changes in tissue hardness, the PID controller automatically adjusts the electrode needle advancement speed (such as from 0.1 mm / s to 0.05 mm / s), to ensure that the target positioning error is ≤0.3 mm. For example, a female patient (35 years old) with secretory otitis media has a mild deviation of the nasal septum. The electrode needle pressure monitoring module sets the target contact pressure to 90 mN (corresponding to a tissue compression amount of about 0.5 mm). p = 0.6 mN / Pa, K i = 0.12 mN / (Pa·s), K d = 0.06 mN·s / Pa. In the first stage, the electrode needle contacts the nasal septum deviation area, and the actual pressure reaches 100 mN (deviation +10 Pa). The PID controller outputs u(t) = 6 + 1.2t + 0. After 2 seconds of integration, the control output increases to 8.4 mN, driving the electrode needle to retreat 0.2 mm, and the pressure drops to 95 mN. In the second stage, the electrode needle enters the relatively flat nasal floor, and the pressure drops to 85 mN (deviation -5 Pa). The PID controller outputs u(t) = -3 - 0.6t + 0. After 1 second of adjustment, the control output decreases to -3.6 mN, the electrode needle advances 0.1 mm, and the pressure returns to 90 mN. Endoscopic examination shows no mucosal damage, and the patient's hearing improves by 20 dB (pre-treatment 40 dB, post-treatment 20 dB). Through the closed-loop control of the pressure monitoring module, the treatment target positioning time is shortened by 30%, the tissue damage rate is reduced to 0.8% (traditional method 5%), and the patient's comfort score is improved to 4.5 / 5 (traditional method 3.2 / 5).

[0069] Step S102, guiding the negative electrode needle into the subject's nasal cavity using an endoscope.

[0070] In this embodiment, the endoscope provides real-time vision, which can clearly display the turbinates, nasal septum, and eustachian tube orifice, thereby guiding the electrode needle to enter along the path of least resistance, avoiding damage to the sphenopalatine artery or branches of the trigeminal nerve.

[0071] Step S103, real-time analysis of the endoscopic image by the MSFFN positioning algorithm and the DAN model of the intelligent control interface module, wherein the MSFFN positioning algorithm is used for preliminary image recognition to capture multi-scale features, and the DAN model is used to learn the treatment positions of the negative electrode needle and the positive electrode needle.

[0072] In this embodiment, the MSFFN algorithm processes features of different scales in endoscopic image capture through a hierarchical convolution structure, adapts to changes in endoscopic field of view and different sizes of lesions. The feature pyramid structure fuses feature information at different levels, especially for small target detection. The high-precision treatment position learning (DAN) model not only uses endoscopic image information, but also fuses the contact force distribution map provided by the electrode needle pressure monitoring module. The spatiotemporal alignment enables the DAN model to focus on the key area and learn the tissue deformation pattern during the electrode needle puncture process, thereby more accurately predicting the optimal treatment point coordinates.

[0073] In an alternative way, the preliminary image recognition by the MSFFN positioning algorithm to capture multi-scale features further comprises:

[0074] The MSFFN algorithm adopts a hierarchical convolution structure to process endoscopic images in parallel through shallow and deep convolution kernels;

[0075] The MSFFN positioning algorithm constructs a feature pyramid layer through a top-down path and lateral connection, wherein the 5th layer feature map is upsampled to the 3rd layer resolution through bilinear interpolation and added element by element with the 3rd layer feature map. Each layer of the feature pyramid independently sets a detection head to output the optimal treatment position coordinates through a non-maximum suppression merging redundant method. Each detection head includes a classification branch and a regression branch. The classification branch uses a 3x3 convolution to generate a target confidence map to identify the electrode needle puncturable area. The regression branch is used to predict the boundary box offset to locate the treatment target point.

[0076] The MSFFN positioning algorithm includes a scale perception module that dynamically adjusts the feature fusion weight according to changes in the endoscopic field of view, wherein the shallow layer feature fusion weight is automatically enhanced when the distance between the electrode needle and the target point is less than or equal to 5mm.

[0077] In this embodiment, shallow convolution kernels (3x3, 5x5) capture mucosal texture and blood vessel branches in endoscopic images, and deep convolution kernels (7x7, 9x9) extract nasal axis and target contour, with a positioning error of ≤0.5mm. When the distance between the electrode needle and the target point is ≤5mm, the scale perception module increases the shallow feature fusion weight from 0.3 to 0.7, suppressing the interference of endoscopic image noise (such as secretions, reflections). The feature pyramid layer is upsampled through bilinear interpolation (the 5th layer feature map is upsampled from 160x120 to 320x240) and added element by element with the 3rd layer feature map, retaining shallow details while fusing deep semantic information, and increasing the detection rate of the tympanic membrane tensor tympani semicanal opening by 30% for diameters <2mm. Each layer of the feature pyramid independently sets a detection head, merges redundant boxes (IoU threshold set to 0.5) through non-maximum suppression (NMS), and outputs the optimal treatment position coordinates, shortening the single-frame inference time to 28ms.

[0078] In an alternative way, the learning of the treatment positions of the negative electrode needle and the positive electrode needle by the DAN model further comprises:

[0079] The DAN model receives the contact force distribution map from the electrode needle pressure monitoring module, performs spatio-temporal alignment through a spatial transformation network, and constructs a four-dimensional feature tensor containing anatomical structures, instrument states, and tissue mechanical properties; in the encoder stage, the DAN model adopts a parallel channel attention module and a spatial attention module; the spatial attention module generates a spatial weight map to focus on a 2mm x 2mm area around the electrode needle tip; in the decoder stage, the ConvLSTM unit is used to establish the time series memory of the treatment area to fuse the current frame features with the hidden states of the previous 5 frames, and learn the tissue deformation pattern during the electrode needle puncture process;

[0080] An output layer with a double-branch structure is used to generate a Gaussian heat map, wherein the peak value corresponds to the optimal treatment point coordinates.

[0081] In this embodiment, the DAN model receives the contact force distribution map from the electrode needle pressure monitoring module and performs spatio-temporal alignment with the endoscopic image through a spatial transformation network, and constructs a four-dimensional feature tensor (64 x 64 x 256 x 4) containing the turbinate, the eustachian tube orifice, the electrode needle angle, the curvature, and the mucosal elasticity and bone wall hardness, so that the treatment point prediction error is ≤0.3mm. Through the ConvLSTM unit, the current frame features and the hidden states (hidden state dimension 128) of the previous 5 frames are fused to learn the tissue deformation pattern (such as the change of mucosal compression amount with puncture depth) of the electrode needle, and the repeatability accuracy can reach 98%. The double-branch output layer generates a Gaussian heat map (two-dimensional Gaussian distribution with σ = 2), and the peak value corresponds to the optimal treatment point coordinates (such as [x = 150, y = 90]). The physician can intuitively judge the treatment point confidence (such as the peak value 0.98 corresponding to high confidence).

[0082] Step S104, after the negative electrode needle and the positive electrode needle are positioned by the intelligent treatment control interface module, the electrode needles are inserted into the predetermined treatment positions learned by the DAN model.

[0083] In this embodiment, the prior knowledge safety boundary constraint, pose error estimation and path local correction measures are combined to ensure that the electrode needle moves within the safety range and avoids damaging the surrounding tissue. Specifically, the six-degree-of-freedom pose of the electrode needle is decomposed into translation and rotation parts, and the PD controller is used to control the translation subspace to move the electrode needle to the vicinity of the target position. At the same time, the adaptive sliding mode algorithm controls the rotation subspace to compensate for the attitude deviation caused by the curved nasal cavity, ensuring that the electrode needle is correctly pointed. When the distance between the electrode needle tip and the treatment area is less than 2mm, the force-position hybrid control mode is switched to control the position and contact force of the electrode needle, ensuring that the electrode needle stably contacts the nasal cavity tissue. During the movement of the electrode needle, the safety boundary constraint function monitors the position of the electrode needle in real time to ensure that it is within the safety range. If the predicted position of the DAN model deviates from the actual position by more than the preset threshold, the B-spline curve interpolation is triggered to locally correct the path of the electrode needle, ensuring that the electrode needle finally reaches the target position.

[0084] In an alternative way, the control of the negative electrode needle and the positive electrode needle to the predetermined treatment position learned by the DAN model through the intelligent treatment control interface module further comprises:

[0085] The six-degree-of-freedom pose of the electrode needle is decomposed into a translation subspace and a rotation subspace; wherein the translation subspace is quickly aligned through a proportional-derivative controller, and the rotation subspace compensates for the attitude deviation caused by the curved nasal cavity channel through an adaptive sliding mode algorithm; the control law of the adaptive sliding mode algorithm is:

[0086]

[0087] Where τ is the torque vector; Δx and are position error and velocity error respectively; K pe and K de are diagonal gain matrices; λ is the switching gain of the sliding surface; η is the adaptive compensation coefficient; s is the sliding mode function;

[0088] When the distance between the electrode needle tip and the treatment area is less than or equal to 2mm, the force-position hybrid control mode is switched to adjust the contact force and position stiffness; wherein the equation for adjusting the contact force and position stiffness is:

[0089]

[0090] Where M d , D d , K d are the desired inertia, damping, and stiffness matrices respectively; Λ is a diagonal positive definite adjustment matrix; F ext is the actual contact force; F des is the set force trajectory;

[0091] According to the prior knowledge of the subject, the safety boundary constraint function dynamically limits the electrode needle movement range, the pose error is estimated in real time according to the deviation between the predicted position of the DAN model and the actual puncture point, and when the pose error exceeds a preset threshold, B-spline curve interpolation is triggered to locally correct the electrode needle path.

[0092] In this embodiment, the trajectory tracking error is less than or equal to 0.1 mm by suppressing the disturbance caused by tissue deformation (such as the change of mucosa compression amount with the change of puncture depth) through a sliding mode function (s = λe(t) + η∫e(t)dt). The adaptive adjustment layer of the ARNN model ensures global asymptotic stability through a positive definite solution matrix, establishes a three-dimensional safety boundary of the nasal cavity (distance from the turbinate ≥ 2 mm, avoiding blood vessel dense areas) according to the CT data of the patient, and triggers B-spline curve interpolation for local path correction (correction time ≤ 100 ms) when the electrode needle pose error exceeds a preset threshold (position error > 1 mm, angle error > 5°).

[0093] For example, a patient (a 55-year-old female) has chronic otitis media with nasal septum deviation (deviation angle 15°). The PD controller drives the electrode needle to quickly align the treatment point (target coordinates [x = 150, y = 90, z = 5]) along the x / y / z axes, with a response time of 40 ms and an overshoot of 3%. The adaptive sliding mode algorithm compensates for the attitude deviation caused by the nasal septum deviation (pitch angle error from 8° to 2°), and the torque vector τ = 0.15 N·m. When the electrode needle tip is 1.5 mm away from the treatment area, switch to force-position hybrid mode, the contact force increases from 50 mN to 80 mN, the position stiffness increases from 100 N / m to 200 N / m, and the puncture force fluctuates stably at ± 2 mN. The deviation between the predicted position of the DAN model and the actual puncture point (Δx = +0.8 mm, Δy = -0.3 mm) dynamically updates the safety boundary (such as adjusting the turbinate safety distance from 2 mm to 2.5 mm). When the pose error exceeds the threshold (position error 1.2 mm), trigger 5 control point interpolation, correction time 80 ms, ensure that the electrode needle pierces into the treatment point along the safe path (avoiding blood vessel dense areas). Through six degrees of freedom precise control and adaptive compensation, the treatment target positioning time is shortened to 5 minutes (traditional method needs 20 minutes).

[0094] Step S105, connect the positive electrode needle to the otitis media regulation device to establish a treatment circuit; and adjust the positioning of the electrode needle through the ARNN treatment control model of the intelligent treatment control interface module based on the received real-time biological feedback signal.

[0095] In this embodiment, the ARNN treatment control model includes an adaptive adjustment layer based on Lyapunov stability theory, and the control formula of the adaptive adjustment layer is:

[0096]

[0097] Among them, x is the system state vector; A is the system matrix; B is the input matrix; K e is the diagonal gain matrix; δ(t) is the nonlinear perturbation term considering tissue deformation; D is the perturbation distribution matrix.

[0098] In this embodiment, by designing a control law that satisfies the Lyapunov stability condition, it can be ensured that the system can return to a stable state after being disturbed, avoiding divergence or oscillation, thereby improving the safety of treatment. The adaptive adjustment layer can dynamically adjust the control parameters (especially the diagonal gain matrix) according to the biofeedback signals and tissue deformation conditions during the actual treatment process to adapt to individual differences and uncertainties in the treatment process. By considering the nonlinear disturbance term caused by tissue deformation and using the disturbance distribution matrix for modeling, the influence of tissue deformation on the positioning of the electrode needle is better suppressed. The biofeedback signal is received in real time through the ARNN model and the electrode needle position is adjusted in real time based on these signals and the adaptive adjustment layer.

[0099] In one optional manner, the diagonal gain matrix is ​​updated by solving the Riccati equation, which is:

[0100] A T P+PA-PBR -1 B T P+Q=0

[0101] Among them, P is the positive definite solution matrix; Q and R are preset weight matrices.

[0102] In this embodiment, the gain matrix obtained by solving the Riccati equation is used to achieve the optimal regulation effect of the system state in terms of minimizing the cost function, thereby more effectively suppressing disturbances and converging to the target state quickly and smoothly.

[0103] For example, a patient with chronic suppurative otitis media has a sudden change in Eustachian tube pressure due to swallowing during treatment. According to the Riccati equation, K e =[0.8,0.6](Q=diag([1,0.1]),R=0.01), calculating u(t) = 9mN. Control commands Δx = -0.2mm (to compensate for position deviation) and ΔF = -2mN (to suppress sudden pressure changes) were generated. The Lyapunov function V(x) was reduced from its initial value of 2.0 to 0.5, demonstrating system stability. The sudden Eustachian tube pressure change δ(t) = +50Pa was compensated by Dδ(t) = 5mN, resulting in actual contact force fluctuations of ≤2mN. Through dynamic gain adjustment of the Riccati equation, the patient's tolerance score was improved to 9.5 / 10.

[0104] According to the scheme provided by the application, the negative electrode needle and the positive electrode needle of the otitis media regulation device are connected to the intelligent regulation interface module; the negative electrode needle is guided into the nasal cavity of the subject by using an endoscope; endoscope images are analyzed in real time by the MSFFN positioning algorithm and the DAN model of the intelligent regulation interface module, wherein preliminary image recognition is performed by the MSFFN positioning algorithm to capture multi-scale features, and the treatment positions of the negative electrode needle and the positive electrode needle are learned by the DAN model; after the negative electrode needle and the positive electrode needle are positioned, the intelligent treatment control interface module is used to control the negative electrode needle and the positive electrode needle to be inserted into the predetermined treatment positions learned by the DAN model; the positive electrode needle is connected to the otitis media regulation device to establish a treatment circuit; and the received real-time biological feedback signals are subjected to positioning adjustment of the electrode needle by the ARNN treatment control model of the intelligent treatment control interface module. The intelligent control and adjustment of the electrode needle are realized by the intelligent treatment control interface module and the ARNN treatment control model, and the individualization and effectiveness of the treatment are improved.

[0105] Figure 3 A frame schematic diagram of the otitis media intelligent regulation device is shown. The otitis media intelligent regulation device comprises:

[0106] The connection module 310 is configured to connect the negative electrode needle and the positive electrode needle of the otitis media regulation device to the intelligent regulation interface module.

[0107] The guidance module 320 is configured to guide the negative electrode needle into the nasal cavity of the subject by using an endoscope.

[0108] The image analysis module 330 is configured to analyze endoscope images in real time by the MSFFN positioning algorithm and the DAN model of the intelligent regulation interface module, wherein preliminary image recognition is performed by the MSFFN positioning algorithm to capture multi-scale features, and the treatment positions of the negative electrode needle and the positive electrode needle are learned by the DAN model.

[0109] The control module 340 is configured to control the negative electrode needle and the positive electrode needle to be inserted into the predetermined treatment positions learned by the DAN model after the negative electrode needle and the positive electrode needle are positioned.

[0110] The positioning adjustment module 350 is configured to connect the positive electrode needle to the otitis media regulation device to establish a treatment circuit, and to perform positioning adjustment of the electrode needle by the ARNN treatment control model of the intelligent treatment control interface module on the basis of received real-time biological feedback signals.

[0111] Figure 4 A structure schematic diagram of the computer device embodiment of the application is shown, and the specific embodiments of the application do not limit the specific implementation of the computer device.

[0112] AsFigure 4 As shown, the computer device can include a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0113] The processor 402, the communications interface 404, and the memory 406 communicate with each other through the communications bus 408. The communications interface 404 is configured to communicate with network elements such as clients or other servers. The processor 402 is configured to execute the program 410, and in particular, can execute the related steps in the middle ear inflammation intelligent regulation method embodiments described above.

[0114] Specifically, the program 410 can include program code including computer operation instructions.

[0115] The processor 402 can be a central processing unit CPU, or an application specific integrated circuit ASIC, or one or more integrated circuits configured to implement embodiments of the present application. The one or more processors included in the computer device can be the same type of processor, such as one or more CPUs; or can be different types of processors, such as one or more CPUs and one or more ASICs.

[0116] The memory 406 is configured to store the program 410. The memory 406 can include a high-speed RAM memory, and can also include a non-volatile memory such as at least one disk memory.

[0117] According to the scheme provided by the present application, the negative electrode needle and the positive electrode needle of the middle ear inflammation regulation device are connected to the intelligent regulation interface module; the negative electrode needle is guided into the nasal cavity of the subject by the endoscope; the endoscope image is analyzed in real time by the MSFFN positioning algorithm and the DAN model of the intelligent regulation interface module, wherein the MSFFN positioning algorithm is used for preliminary image recognition to capture multi-scale features, and the DAN model is used to learn the treatment position of the negative electrode needle and the positive electrode needle; after the negative electrode needle and the positive electrode needle are positioned by the intelligent treatment control interface module, they are inserted into the predetermined treatment position learned by the DAN model; the positive electrode needle is connected to the middle ear inflammation regulation device to establish a treatment circuit; and the received real-time biological feedback signal is subjected to positioning adjustment of the electrode needle by the ARNN treatment control model of the intelligent treatment control interface module. The intelligent treatment control interface module and the ARNN treatment control model are used to realize intelligent control and adjustment of the electrode needle, thereby improving the individualization and effectiveness of the treatment.

[0118] Those skilled in the art will appreciate that the modules in the apparatuses in the embodiments can be adapted and placed in one or more apparatuses other than that of the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and furthermore can be divided into multiple sub-modules or sub-units or sub-components. Any combination of all the features disclosed in the present specification (including the accompanying claims, abstract and drawings), and any method or process of any combination of the features disclosed in the present specification may be adopted, except that at least some of such features and / or processes are mutually exclusive. Unless explicitly stated, each feature disclosed in the present specification (including the accompanying claims, abstract and drawings) can be replaced by an alternative feature providing the same, equivalent or similar function. Moreover, those skilled in the art will appreciate that the combination of features of different embodiments means that the scope of the application includes different embodiments and that the features of the different embodiments can be combined with each other in other embodiments. For example, in the claims, any of the embodiments can be used in any combination. The application can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unitary claim, several of the devices claimed in conjunction can be embodied by one and the same item of hardware. The sequence of steps in the embodiments described above, other than when specifically stated otherwise, should not be understood as limiting the order of execution.

Claims

1. A method for intelligent regulation of otitis media, characterized in that, The method comprises the following steps: connecting the negative electrode needle and the positive electrode needle of the middle ear inflammation regulation device to the intelligent regulation interface module; guiding the negative electrode needle into the nasal cavity of the subject by the endoscope; real-time analyzing the endoscope image by the MSFFN positioning algorithm and the DAN model of the intelligent regulation interface module, wherein the MSFFN positioning algorithm is used for preliminary image recognition to capture multi-scale features, and the DAN model is used for learning the treatment position of the negative electrode needle and the positive electrode needle; after the negative electrode needle and the positive electrode needle are positioned by the intelligent treatment control interface module, the negative electrode needle and the positive electrode needle are inserted into the predetermined treatment position learned by the DAN model; connecting the positive electrode needle to the middle ear inflammation regulation device to establish a treatment circuit, and adjusting the position of the electrode needle by the ARNN treatment control model of the intelligent treatment control interface module according to the received real-time biological feedback signal.

2. The method of claim 1, wherein, The middle ear inflammation regulation device comprises an electrode needle assembly, an intelligent regulation interface module, an intelligent treatment control interface module, an endoscope and a data recording module. The electrode needle assembly comprises a negative electrode needle and a positive electrode needle, which are used to establish a treatment circuit; the endoscope is used to guide the negative electrode needle into the nasal cavity and provide real-time images; and the data recording module is used to record the electrode needle position, treatment current, treatment time and biological feedback signal during treatment.

3. The method of claim 1, wherein, The middle ear inflammation regulation device further comprises an electrode needle pressure monitoring module, which monitors the contact pressure between the electrode needle and the nasal cavity tissue in real time by a pressure sensor and performs pressure feedback control. The formula of the pressure feedback control is: wherein u(t) is the control output in milli-Newton; e(t) is the deviation between the set pressure and the actual pressure in Pa; K p is the proportional coefficient in mN / Pa; K i is the integral coefficient in mN / (Pa x s); K d is the derivative coefficient in mN x s / Pa; t is the time in s.

4. The middle ear inflammation intelligent regulation method according to claim 3, characterized in that, The ARNN treatment control model comprises an adaptive adjustment layer based on Lyapunov stability theory, and the control formula of the adaptive adjustment layer is: where x is the system state vector; A is the system matrix; B is the input matrix; K e is a diagonal gain matrix; δ(t) is a nonlinear disturbance term that accounts for tissue deformation; and D is a disturbance distribution matrix.

5. The method of claim 4, wherein, The diagonal gain matrix is updated by solving the Riccati equation, and the Riccati equation is: A T P+PA-PBR -1 B T P+Q=0 wherein P is a positive definite solution matrix; Q and R are preset weight matrices.

6. The middle ear inflammation intelligent regulation method according to claim 1, characterized in that, The preliminary image recognition by the MSFFN positioning algorithm to capture multi-scale features further comprises: The MSFFN algorithm adopts a hierarchical convolution structure, and the endoscope image is processed by shallow convolution kernels and deep convolution kernels in parallel; The MSFFN positioning algorithm constructs a feature pyramid layer through a top-down path and a lateral connection, wherein the fifth layer feature map is up-sampled to the third layer resolution by bilinear interpolation and is added element by element with the third layer feature map; each layer of the feature pyramid independently sets a detection head to output the optimal treatment position coordinates by a non-maximum suppression merging redundancy method, each detection head includes a classification branch and a regression branch, the classification branch uses a 3x3 convolution to generate a target confidence map to identify the electrode needle puncture area, and the regression branch is used to predict the boundary box offset to locate the treatment target; The MSFFN positioning algorithm comprises a scale perception module, which dynamically adjusts the feature fusion weight according to the change of the endoscope field of view, wherein the shallow layer features are automatically enhanced when the distance between the electrode needle and the target is less than or equal to 5mm.

7. The method of claim 1, wherein, The learning of the treatment position of the negative electrode needle and the positive electrode needle by the DAN model further comprises: The DAN model receives a contact force distribution map from an electrode needle pressure monitoring module, performs spatio-temporal alignment through a spatial transformation network, and constructs a four-dimensional feature tensor containing anatomical structures, instrument states, and tissue mechanical properties; wherein, in the encoder stage, the DAN model adopts a parallel channel attention module and a spatial attention module; the spatial attention module generates a spatial weight map to focus on a 2mm×2mm area around the electrode needle tip; in the decoder stage, a ConvLSTM unit is used to establish a time series memory of the treatment area to fuse the current frame features with the hidden states of the previous 5 frames, and learn the tissue deformation pattern during the electrode needle puncture process; An output layer with a double-branch structure is used to generate a Gaussian heat map, where the peak corresponds to the optimal treatment point coordinates.

8. The middle ear inflammation intelligent regulation method according to claim 1, characterized in that, The control of the negative electrode needle and the positive electrode needle into the predetermined treatment position learned by the DAN model after positioning through the intelligent treatment control interface module further comprises: The six-degree-of-freedom pose of the electrode needle is decomposed into a translational subspace and a rotational subspace; wherein, the translational subspace is quickly aligned through a proportional-differential controller, and the rotational subspace compensates for the attitude deviation caused by the curved channel of the nasal cavity through an adaptive sliding mode algorithm; the control law of the adaptive sliding mode algorithm is: where τ is the torque vector; Δx and are the position error and velocity error, respectively; K pe and K de are diagonal gain matrices; λ is the sliding surface switching gain; and η is the adaptive compensation coefficient; and s is the sliding function. When the distance between the electrode needle tip and the treatment area is less than or equal to 2mm, switch to a force-position hybrid control mode and adjust the contact force and position stiffness; wherein, the equation for adjusting the contact force and position stiffness is: where M d , D d , K d are the desired inertia, damping, stiffness matrices, respectively; Λ is a diagonal positive definite conditioning matrix; F ext is the actual contact force; F des is the set force trajectory; The prior knowledge safety boundary constraint function of the subject dynamically limits the electrode needle movement range, and the pose error is estimated in real time according to the deviation between the predicted position of the DAN model and the actual puncture point, and when the pose error exceeds a preset threshold, a B-spline curve interpolation is triggered to locally correct the electrode needle path.

9. A middle ear inflammation intelligent regulation device, characterized in that, It comprises: A connecting module for connecting the negative electrode needle and the positive electrode needle of the otitis media regulation device to the intelligent regulation interface module; A guiding module for guiding the negative electrode needle into the subject's nasal cavity using an endoscope; An image analysis module for real-time analysis of endoscope images through the MSFFN positioning algorithm and the DAN model of the intelligent regulation interface module, wherein the MSFFN positioning algorithm is used for preliminary image recognition to capture multi-scale features, and the DAN model is used to learn the treatment position of the negative electrode needle and the positive electrode needle; A control module for controlling the negative electrode needle and the positive electrode needle to penetrate into the predetermined treatment position learned by the DAN model after positioning through the intelligent treatment control interface module; A positioning adjustment module for connecting the positive electrode needle to the otitis media regulation device to establish a treatment circuit, and adjusting the positioning of the electrode needle through the ARNN treatment control model of the intelligent treatment control interface module according to the received real-time biological feedback signal.

10. A computer device comprising: A processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; The memory is used to store at least one executable instruction, and the executable instruction makes the processor execute the corresponding operation of the above-mentioned otitis media intelligent regulation method.

Images