Intelligent pressure regulating type lower limb lymphedema treatment method and device based on dynamic grading

Abstract

The application discloses a kind of intelligent pressure regulating type lower limb lymphedema treatment method and device based on dynamic grading, and wherein method includes: real-time acquisition patient lower limb's pressure distribution, tissue impedance, surface temperature and the standardized multimodal data of limb volume;The standardized multimodal data is input to the DCNN edema grade evaluation model that has been trained, and the current edema grade of patient is output;Wherein, the DCNN edema grade evaluation model includes initial DCNN network layer, enhanced DCNN network layer, offset field network layer, RPN network layer, Region proposal network layer and ROI pooling network layer;According to the current edema grade, select initial pressure treatment scheme from pre-set pressure treatment model, control intelligent pneumatic pressurization module to carry out self-adapting dynamic adjustment to the pressure of different parts of lower limb and generate pressure regulating parameter.The application improves the intelligent level and treatment effect of lower limb lymphedema treatment by intelligent pressure regulating control and real-time feedback.

Description

Technical Field

[0001] This invention relates to the field of medical auxiliary treatment technology, specifically to a method and device for treating lower limb lymphedema based on dynamic grading and intelligent pressure regulation, as well as a computing device. Background Technology

[0002] Lower extremity lymphedema is a swelling of the limb caused by abnormal accumulation of lymph fluid in the interstitial spaces due to dysfunction of the lymphatic system. It not only affects the patient's limb function and appearance, but also causes complications such as pain, infection, and skin ulcers, seriously affecting the patient's quality of life.

[0003] Currently, treatment for lower extremity lymphedema mainly includes conservative and surgical treatments. Conservative treatments include manual lymphatic drainage, compression therapy (elastic stockings, bandages), exercise therapy, and skin care, with compression therapy being the most commonly used method. Surgical treatments include lymphovenous anastomosis, lymph node transplantation, and liposuction. Among these treatments, compression therapy typically uses compression stockings or bandages with fixed pressure. However, the pressure is difficult to adjust individually based on the patient's condition; too little pressure may not achieve the desired therapeutic effect, while too much pressure may cause discomfort or even injury. Manual lymphatic drainage requires a professional therapist, is time-consuming and labor-intensive, and its effectiveness is greatly influenced by the therapist's experience. Surgical treatment carries certain risks and complications and is not suitable for all patients. Furthermore, traditional compression devices (such as air pump-driven compression sleeves) usually only provide preset fixed or cyclic pressure, making it difficult to dynamically adjust based on the patient's real-time physiological feedback.

[0004] To address the aforementioned issues, this invention proposes a dynamic grading-based intelligent pressure-regulating treatment method for lower limb lymphedema. Through intelligent pressure regulation control and real-time feedback, it aims to improve the intelligence level and treatment efficacy of lower limb lymphedema treatment. Summary of the Invention

[0005] In view of the above problems, the present invention provides a treatment method and device for lower limb lymphedema based on dynamic grading and intelligent pressure regulation, as well as a computing device.

[0006] According to one aspect of the present invention, a smart pressure-regulating treatment method for lower extremity lymphedema based on dynamic grading is provided, comprising:

[0007] Standardized multimodal data on pressure distribution, tissue impedance, surface temperature, and limb volume of a patient's lower limbs are collected in real time using a multimodal sensor array; the multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor, and a 3D scanner.

[0008] The standardized multimodal data is input into a pre-trained DCNN edema level assessment model, which outputs the patient's current edema level. The DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Region Proposal Network layer, and a ROI Pooling Network layer.

[0009] Based on the current edema level, an initial pressure treatment plan is selected from the preset pressure treatment model. The intelligent pneumatic pressurization module is controlled to adaptively and dynamically adjust the pressure on different parts of the lower limbs to generate pressure adjustment parameters. The state space of the preset pressure treatment model uses the real-time pressure value fed back by the pressure sensor and the tissue impedance change rate fed back by the bioimpedance sensor. The action space uses the pressure gradient, pressure rhythm and duration of the pneumatic pressurization module.

[0010] In one alternative approach, the DCNN edema level assessment model is trained using a dynamically weighted multi-task loss function, which is:

[0011]

[0012] Among them, L Dice For Dice similarity coefficient loss; L CE Cross-entropy loss; α, β, γ are dynamic weighting coefficients; I(t) is the bioimpedance distribution field at time t; D is the diffusion coefficient matrix; is the divergence operator; T is the duration of a single treatment cycle.

[0013] In one alternative approach, the objective function for optimizing the pressure gradient of the intelligent pneumatic pressurization module is:

[0014]

[0015] Where V(t) is the limb volume at time t; V target Target volume; T is treatment duration; This indicates that the objective function J is minimized by adjusting the pressure gradient ΔP; ΔP(t) is the pressure gradient at time t.

[0016] In one alternative approach, the energy function for surface reconstruction using the 3D scanner limb volume measurement is:

[0017]

[0018] Where f is the implicit surface function; g is the original scan data; κ is the average curvature; H0 is the target curvature; λ and μ are regularization parameters; Ω is the scan space; The boundary of the limb surface.

[0019] In one alternative approach, the multimodal sensor array adopts a six-axis spatial layout scheme;

[0020] Among them, a 4×4 flexible pressure sensor matrix is ​​arranged 5cm above the ankle joint, with a sampling frequency ≥200Hz;

[0021] Two sets of bioimpedance electrode pairs were placed on the anterior and posterior sides of the lower leg, and the four-electrode method was used for measurement, with an excitation frequency range of 1kHz-1MHz.

[0022] Infrared temperature sensors are arranged in a spiral pattern with 1cm spacing, covering the entire calf area;

[0023] The 3D scanner uses binocular structured light, with a baseline length of 150mm, a working distance of 300-500mm, and a single scan accuracy of ≤0.5mm.

[0024] In one alternative embodiment, the intelligent pneumatic pressurization module comprises a three-stage pressure regulation system consisting of a base pressure layer, a dynamic compensation layer, and a safety protection layer;

[0025] The basic pressure layer consists of 12 independent air chambers, each equipped with a proportional pressure valve and a flow sensor;

[0026] The dynamic compensation layer employs a microvalve array driven by shape memory alloy, with a response time ≤20ms;

[0027] The safety protection layer integrates a pressure relief valve and an overpressure alarm circuit. The pressure control algorithm adopts a model predictive control framework with prediction time domain N=15 and control time domain M=5. The optimization objective function includes three dimensions: pressure tracking error, pressure change rate, and air chamber volume change. The constraints include a maximum allowable pressure gradient of 5 kPa / s, a minimum safe pressure of 3 kPa, and a maximum operating frequency of 0.5 Hz.

[0028] When the detected tissue impedance change rate Z>0.2Ω / s, the pressure holding mode is automatically triggered, maintaining the current pressure value until Z<0.05Ω / s. The volume change compensation during the pressure holding phase is achieved through a PID controller with a proportional coefficient Kp=0.8, integral time Ti=30s, and derivative time Td=5s.

[0029] In an alternative approach, the training process of the DCNN edema level assessment model further includes:

[0030] Shallow features of multimodal data are extracted through the initial DCNN network layers. These shallow features include local patterns of pressure distribution, spatial gradients of tissue impedance, texture information of surface temperature, and geometric contours of limb volume.

[0031] By enhancing the DCNN network layers, residual connections and attention mechanisms are used to deepen the network and improve its feature representation capabilities;

[0032] The offset field network layer predicts the offset of each pixel, thereby registering the limb shape and position to solve the measurement error caused by changes in limb posture.

[0033] Candidate regions are generated using the RPN network layer, which generates multiple anchor boxes of different sizes and proportions on the feature map through a sliding window and predicts the foreground / background probability and bounding box regression parameters for each anchor box.

[0034] The Region Proposal Network layer filters out candidate regions with high confidence based on the output of the RPN, which are then used as the input to the ROI Pooling Network layer.

[0035] The ROI pooling network layer maps candidate regions of different sizes to feature vectors of a fixed size.

[0036] In an alternative approach, the step of solving the objective function for optimizing the pressure gradient further includes:

[0037] The treatment duration T is divided into N time steps, and the length of each time step is Δt = T / N;

[0038] By approximating the volume change rate using the finite difference method, a discrete form of the volume change rate expression is obtained;

[0039] The objective function is transformed into a discrete summation expression, which includes the square of the rate of change of volume at each time step, the square of the pressure gradient, and the square of the deviation between the volume and the target volume at the final time step.

[0040] A quadratic programming solver is used to solve the discretized objective function to obtain the optimal pressure gradient sequence. The solution process includes constraints on the maximum allowable pressure gradient, minimum safe pressure, and maximum operating frequency of the pneumatic pressurization module.

[0041] According to another aspect of the present invention, a smart pressure-regulating lower limb lymphedema treatment device based on dynamic grading is provided, comprising:

[0042] A multimodal data acquisition module is used to acquire standardized multimodal data on pressure distribution, tissue impedance, surface temperature, and limb volume of a patient's lower limb in real time using a multimodal sensor array; the multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor, and a 3D scanner;

[0043] The edema level assessment module is used to input the standardized multimodal data into the pre-trained DCNN edema level assessment model and output the patient's current edema level; wherein, the DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Region proposal network layer, and a ROI pooling network layer;

[0044] The intelligent pressure regulation module is used to select an initial pressure treatment plan from a preset pressure treatment model based on the current edema level, and control the intelligent pneumatic pressurization module to adaptively and dynamically adjust the pressure on different parts of the lower limbs to generate pressure regulation parameters; wherein, the state space of the preset pressure treatment model adopts the real-time pressure value fed back by the pressure sensor and the tissue impedance change rate fed back by the bioimpedance sensor; the action space adopts the pressure gradient, pressure rhythm and duration of the pneumatic pressurization module.

[0045] According to another aspect of the present invention, a computing device is provided, 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 via the communication bus;

[0046] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described intelligent pressure-regulating treatment method for lower extremity lymphedema based on dynamic grading.

[0047] According to the solution provided by this invention, a multimodal sensor array is used to collect standardized multimodal data on pressure distribution, tissue impedance, surface temperature, and limb volume of the patient's lower limbs in real time. The multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor, and a 3D scanner. The standardized multimodal data is input into a pre-trained DCNN edema level assessment model, which outputs the patient's current edema level. The DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Regionproposal network layer, and an ROI pooling network layer. Based on the current edema level, an initial pressure treatment plan is selected from a preset pressure treatment model, and an intelligent pneumatic pressurization module is controlled to adaptively and dynamically adjust the pressure on different parts of the lower limbs to generate pressure regulation parameters. The state space of the preset pressure treatment model uses the real-time pressure value fed back by the pressure sensor and the tissue impedance change rate fed back by the bioimpedance sensor; the action space uses the pressure gradient, pressure application rhythm, and duration of the pneumatic pressurization module. This invention improves the intelligence level and treatment effect of lower limb lymphedema treatment through intelligent pressure regulation control and real-time feedback.

[0048] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0049] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0050] Figure 1 A schematic flowchart of a dynamic grading-based intelligent pressure-regulating treatment method for lower extremity lymphedema is shown in an embodiment of the present invention.

[0051] Figure 2 A schematic diagram of the frame of the intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading according to an embodiment of the present invention is shown;

[0052] Figure 3 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. Detailed Implementation

[0053] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0054] Figure 1 This diagram illustrates a flow chart of a dynamically graded, intelligent pressure-regulating lower extremity lymphedema treatment method according to an embodiment of the present invention. Specifically, as shown... Figure 1 As shown, it includes the following steps:

[0055] Step S101: Real-time acquisition of standardized multimodal data on pressure distribution, tissue impedance, surface temperature, and limb volume of the patient's lower limb using a multimodal sensor array; the multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor, and a 3D scanner.

[0056] In this embodiment, a pressure sensor quantifies the pressure distribution on the limb surface, reflecting changes in tissue tension caused by lymphatic fluid accumulation; a bioimpedance sensor measures the fluid content between tissues, reflecting the degree of edema; an infrared temperature sensor detects inflammatory responses to reflect the metabolic state of the edematous area; and a 3D scanner measures changes in limb volume as an indicator of overall edema. All sensors are non-invasive or minimally invasive, avoiding additional pain or damage to the patient.

[0057] In one alternative approach, the energy function for surface reconstruction using the 3D scanner limb volume measurement is:

[0058]

[0059] Where f is the implicit surface function; g is the original scan data; κ is the average curvature; H0 is the target curvature; λ and μ are regularization parameters; Ω is the scan space; The boundary of the limb surface.

[0060] In this embodiment, high-precision limb surface reconstruction is achieved by using implicit surface functions and error terms from the original scan data, reducing noise and errors in the scan data. Constraints on the average curvature and target curvature ensure the smoothness and continuity of the reconstructed surface, avoiding unnatural abrupt changes and sharp angles during reconstruction. Constraining the limb surface boundaries more accurately captures the limb's geometry, improving the accuracy of the reconstructed surface. For example, a 3D scanner is used to scan a patient's lower leg to obtain raw scan data, and an initial implicit surface function (such as an ellipsoid) is selected. The implicit surface function is iteratively updated using gradient descent, calculating the average curvature and target curvature of the current surface in each iteration. Regularization terms are calculated to ensure surface smoothness and data fit. The surface of the lower leg is reconstructed using the optimized implicit surface function.

[0061] In one alternative approach, the multimodal sensor array adopts a six-axis spatial layout scheme;

[0062] Among them, a 4×4 flexible pressure sensor matrix is ​​arranged 5cm above the ankle joint, with a sampling frequency ≥200Hz;

[0063] Two sets of bioimpedance electrode pairs were placed on the anterior and posterior sides of the lower leg, and the four-electrode method was used for measurement, with an excitation frequency range of 1kHz-1MHz.

[0064] Infrared temperature sensors are arranged in a spiral pattern with 1cm spacing, covering the entire calf area;

[0065] The 3D scanner uses binocular structured light, with a baseline length of 150mm, a working distance of 300-500mm, and a single scan accuracy of ≤0.5mm.

[0066] In this embodiment, a multimodal sensor array with a six-axis spatial layout is used, which can simultaneously collect pressure, bioimpedance, temperature and 3D shape information, providing multi-angle data support and improving the accuracy of treatment for lower extremity lymphedema.

[0067] Step S102: Input the standardized multimodal data into the trained DCNN edema level assessment model and output the patient's current edema level; wherein, the DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Region proposal network layer, and an ROI pooling network layer.

[0068] In this embodiment, the Offset Field network layer predicts pixel offsets to correct measurement errors caused by changes in limb posture. An initial DCNN network layer extracts shallow features, and an enhanced DCNN network layer deepens the network through residual connections and attention mechanisms, thereby improving feature representation capabilities. RPN and ROIpooling network layers focus on key areas of the lower limb (such as areas with significant edema), enhancing the ability to identify key pathological features.

[0069] In one alternative approach, the DCNN edema level assessment model is trained using a dynamically weighted multi-task loss function, which is:

[0070]

[0071] Among them, L Dice For Dice similarity coefficient loss; L CE Cross-entropy loss; α, β, γ are dynamic weighting coefficients; I(t) is the bioimpedance distribution field at time t; D is the diffusion coefficient matrix; is the divergence operator; T is the duration of a single treatment cycle.

[0072] In this embodiment, the loss function combines Dice similarity coefficient loss, cross-entropy loss, and a regularization term based on the bioimpedance distribution field. The model simultaneously learns the segmentation of edema regions and the spatiotemporal variation characteristics of bioimpedance. The Dice similarity coefficient loss is specifically designed for segmentation tasks, especially in cases of class imbalance, where it better optimizes the segmentation of small targets. This is crucial for edema region segmentation because the size of edema regions can vary significantly. The cross-entropy loss is a standard pixel-level classification loss function, enabling the model to accurately classify the edema level for each pixel.

[0073] In this embodiment, the training process of the DCNN edema level assessment model further includes:

[0074] Shallow features of multimodal data are extracted through the initial DCNN network layers. These shallow features include local patterns of pressure distribution, spatial gradients of tissue impedance, texture information of surface temperature, and geometric contours of limb volume.

[0075] By enhancing the DCNN network layers, residual connections and attention mechanisms are used to deepen the network depth and enhance the feature representation capability;

[0076] The offset field network layer is used to predict the offset of each pixel, thereby registering the limb shape and position to solve the measurement error caused by changes in limb posture.

[0077] Candidate regions are generated using the RPN network layer, which generates multiple anchor boxes of different sizes and proportions on the feature map through a sliding window and predicts the foreground / background probability and bounding box regression parameters for each anchor box.

[0078] The Region Proposal Network layer filters out candidate regions with high confidence based on the output of the RPN, which are then used as the input to the ROI Pooling Network layer.

[0079] The ROI pooling network layer maps candidate regions of different sizes to feature vectors of a fixed size.

[0080] Step S103: Select an initial pressure treatment plan from the preset pressure treatment model according to the current edema level, and control the intelligent pneumatic pressurization module to adaptively and dynamically adjust the pressure on different parts of the lower limb to generate pressure adjustment parameters; wherein, the state space of the preset pressure treatment model adopts the real-time pressure value fed back by the pressure sensor and the tissue impedance change rate fed back by the bioimpedance sensor; the action space adopts the pressure gradient, pressure rhythm and duration of the pneumatic pressurization module.

[0081] In this embodiment, pressure sensors and bioimpedance sensors are used to monitor changes in pressure and tissue impedance in the patient's lower limbs in real time. This data is then input into a preset pressure therapy model as a state space, enabling dynamic adjustment of the treatment plan. An intelligent pneumatic pressurization module precisely controls the pressure at different points on the lower limbs, achieving adaptive dynamic adjustment of the pressure gradient to ensure treatment comfort and safety. Dynamically adjusting the pressure gradient, application rhythm, and duration can more effectively promote lymphatic drainage, thereby reducing edema.

[0082] In one alternative approach, the objective function for optimizing the pressure gradient of the intelligent pneumatic pressurization module is:

[0083]

[0084] Where V(t) is the limb volume at time t; V target Target volume; T is treatment duration; This indicates that the objective function J is minimized by adjusting the pressure gradient ΔP; ΔP(t) is the pressure gradient at time t.

[0085] In this embodiment, the objective function This term directly reflects the square of the rate of change in limb volume. Minimizing this term means reducing limb volume as smoothly and efficiently as possible, thereby optimizing lymphatic drainage and avoiding a situation where only the pressure itself is considered while ignoring the actual treatment effect. ||ΔP(t)|| 2 The term is a penalty term for the pressure gradient, which can avoid drastic pressure changes and ensure the comfort and safety of treatment, because excessively rapid pressure changes can cause vasoconstriction or patient discomfort.

[0086] In an alternative approach, the step of solving the objective function for optimizing the pressure gradient further includes:

[0087] The treatment duration T is divided into N time steps, and the length of each time step is Δt = T / N;

[0088] By using the finite difference method to approximate the rate of volume change, a discrete form of the expression for the rate of volume change is obtained;

[0089] The objective function is transformed into a discrete summation expression, which includes the square of the rate of change of volume at each time step, the square of the pressure gradient, and the square of the deviation between the volume and the target volume at the final time step.

[0090] A quadratic programming solver is used to solve the discretized objective function to obtain the optimal pressure gradient sequence. The solution process includes constraints on the maximum allowable pressure gradient, minimum safe pressure, and maximum operating frequency of the pneumatic pressurization module.

[0091] In this embodiment, by discretizing the continuous treatment process into multiple time steps, the pressure gradient of each time step can be optimized, enabling more precise control of the pressurization process, thereby more effectively promoting lymphatic drainage and reducing edema.

[0092] Real-time optimization: The objective function is transformed into a discrete form by approximating the volume change rate using the finite difference method, facilitating rapid solution using a quadratic programming solver. This allows for dynamic adjustment of the pressure gradient based on real-time data, better adapting to individual patient differences and treatment progress. Physical constraints on the pneumatic pressurization module (such as maximum allowable pressure gradient, minimum safe pressure, and maximum operating frequency) are introduced during the optimization process to ensure that the optimized pressure gradient sequence achieves both therapeutic efficacy and safety during treatment.

[0093] In one alternative embodiment, the intelligent pneumatic pressurization module comprises a three-stage pressure regulation system consisting of a base pressure layer, a dynamic compensation layer, and a safety protection layer;

[0094] The basic pressure layer consists of 12 independent air chambers, each equipped with a proportional pressure valve and a flow sensor;

[0095] The dynamic compensation layer employs a microvalve array driven by shape memory alloy, with a response time ≤20ms;

[0096] The safety protection layer integrates a pressure relief valve and an overpressure alarm circuit. The pressure control algorithm adopts a model predictive control framework with prediction time domain N=15 and control time domain M=5. The optimization objective function includes three dimensions: pressure tracking error, pressure change rate, and air chamber volume change. The constraints include a maximum allowable pressure gradient of 5 kPa / s, a minimum safe pressure of 3 kPa, and a maximum operating frequency of 0.5 Hz.

[0097] When the detected tissue impedance change rate Z>0.2Ω / s, the pressure holding mode is automatically triggered, maintaining the current pressure value until Z<0.05Ω / s. The volume change compensation during the pressure holding phase is achieved through a PID controller with a proportional coefficient Kp=0.8, integral time Ti=30s, and derivative time Td=5s.

[0098] In this embodiment, the base pressure layer provides a stable base pressure, the dynamic compensation layer can quickly respond to changes in tissue state, and the safety protection layer ensures the safety of the treatment process. The dynamic compensation layer uses a microvalve array driven by shape memory alloy (SMA), which has a fast response speed (≤20ms) and can quickly adjust the pressure according to changes in tissue impedance, better adapting to individual patient differences and treatment needs. The safety protection layer integrates a pressure relief valve and an overpressure alarm circuit, which can release pressure and issue an alarm in time when the pressure rises abnormally, preventing injury to the patient. The pressure control algorithm adopts a model predictive control (MPC) framework, which can predict pressure changes in the future and optimize control in advance, thereby achieving smoother and more stable pressure control. The rate of change of tissue impedance serves as a condition for triggering the pressure holding mode, which can automatically identify sudden changes in tissue state (possibly caused by excessive pressure) and adjust the treatment plan in time to ensure the safety and effectiveness of the treatment. During the pressure holding phase, a PID controller is used for volume change compensation, ensuring that effective compression of the limb is maintained by adjusting the air chamber volume while keeping the pressure constant, continuously promoting lymphatic return.

[0099] According to the solution provided by this invention, a multimodal sensor array is used to collect standardized multimodal data on pressure distribution, tissue impedance, surface temperature, and limb volume of the patient's lower limbs in real time. The multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor, and a 3D scanner. The standardized multimodal data is input into a pre-trained DCNN edema level assessment model, which outputs the patient's current edema level. The DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Regionproposal network layer, and an ROI pooling network layer. Based on the current edema level, an initial pressure treatment plan is selected from a preset pressure treatment model, and an intelligent pneumatic pressurization module is controlled to adaptively and dynamically adjust the pressure on different parts of the lower limbs to generate pressure regulation parameters. The state space of the preset pressure treatment model uses the real-time pressure value fed back by the pressure sensor and the tissue impedance change rate fed back by the bioimpedance sensor; the action space uses the pressure gradient, pressure application rhythm, and duration of the pneumatic pressurization module. This invention improves the intelligence level and treatment effect of lower limb lymphedema treatment through intelligent pressure regulation control and real-time feedback.

[0100] Figure 2 A schematic diagram of the frame of the intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading according to an embodiment of the present invention is shown. The intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading includes:

[0101] The multimodal data acquisition module 210 is used to acquire standardized multimodal data of pressure distribution, tissue impedance, surface temperature and limb volume of the patient's lower limb in real time using a multimodal sensor array; the multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor and a 3D scanner.

[0102] The edema level assessment module 220 is used to input the standardized multimodal data into the trained DCNN edema level assessment model and output the patient's current edema level; wherein, the DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Region proposal network layer, and a ROI pooling network layer;

[0103] The intelligent pressure regulation module 230 is used to select an initial pressure treatment plan from a preset pressure treatment model based on the current edema level, and control the intelligent pneumatic pressurization module to adaptively and dynamically adjust the pressure on different parts of the lower limbs to generate pressure regulation parameters; wherein, the state space of the preset pressure treatment model adopts the real-time pressure value fed back by the pressure sensor and the tissue impedance change rate fed back by the bioimpedance sensor; the action space adopts the pressure gradient, pressure rhythm and duration of the pneumatic pressurization module.

[0104] Figure 3 The diagram shows a structural schematic of an embodiment of the computing device of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computing device.

[0105] like Figure 3 As shown, the computing device may include: a processor 302, a communications interface 304, a memory 306, and a communications bus 308.

[0106] The processor 302, communication interface 304, and memory 306 communicate with each other via communication bus 308. Communication interface 304 is used to communicate with other network elements, such as clients or other servers. The processor 302 executes program 310, specifically performing the relevant steps in the above-described embodiment of the intelligent pressure-regulating lower limb lymphedema treatment method based on dynamic grading.

[0107] Specifically, program 310 may include program code that includes computer operation instructions.

[0108] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computing device may include one or more processors of the same type, such as one or more CPUs; or it may include processors of different types, such as one or more CPUs and one or more ASICs.

[0109] Memory 306 is used to store program 310. Memory 306 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0110] According to the solution provided by this invention, a multimodal sensor array is used to collect standardized multimodal data on pressure distribution, tissue impedance, surface temperature, and limb volume of the patient's lower limbs in real time. The multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor, and a 3D scanner. The standardized multimodal data is input into a pre-trained DCNN edema level assessment model, which outputs the patient's current edema level. The DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Regionproposal network layer, and an ROI pooling network layer. Based on the current edema level, an initial pressure treatment plan is selected from a preset pressure treatment model, and an intelligent pneumatic pressurization module is controlled to adaptively and dynamically adjust the pressure on different parts of the lower limbs to generate pressure regulation parameters. The state space of the preset pressure treatment model uses the real-time pressure value fed back by the pressure sensor and the tissue impedance change rate fed back by the bioimpedance sensor; the action space uses the pressure gradient, pressure application rhythm, and duration of the pneumatic pressurization module. This invention improves the intelligence level and treatment effect of lower limb lymphedema treatment through intelligent pressure regulation control and real-time feedback.

[0111] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination of all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed can be employed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose. Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims listing several devices, several of these devices may be embodied by the same hardware item. Unless otherwise specified, the steps in the above embodiments should not be construed as limiting the order of execution.

Claims

1. A smart pressure-regulating lower limb lymphedema treatment device based on dynamic grading, characterized in that, include: The multimodal data acquisition module is used to acquire standardized multimodal data on pressure distribution, tissue impedance, surface temperature, and limb volume of the patient's lower limbs in real time using a multimodal sensor array; The multimodal sensor array includes a flexible pressure sensor, a bioimpedance sensor, an infrared temperature sensor, and a 3D scanner; The edema level assessment module is used to input the standardized multimodal data into the pre-trained DCNN edema level assessment model and output the patient's current edema level; wherein, the DCNN edema level assessment model includes an initial DCNN network layer, an enhanced DCNN network layer, an offset field network layer, an RPN network layer, a Region proposal network layer, and a ROI pooling network layer; The intelligent pressure regulation module is used to select an initial pressure treatment plan from a preset pressure treatment model based on the current edema level, and to control the intelligent pneumatic pressurization module to adaptively and dynamically adjust the pressure on different parts of the lower limbs to generate pressure regulation parameters. The state space of the preset pressure treatment model uses real-time pressure values ​​fed back by pressure sensors and the rate of change of tissue impedance fed back by bioimpedance sensors; the action space uses the pressure gradient, pressure application rhythm, and duration of the pneumatic pressurization module. The objective function for optimizing the pressure gradient of the intelligent pneumatic pressurization module is: ， in, Let 't' be the limb volume. For the target volume; Treatment duration; This indicates that by adjusting the pressure gradient Minimize objective function ; The pressure gradient at time t; The steps for solving the objective function of the pressure gradient optimization further include: The treatment duration T is divided into N time steps, each time step having a length of Δt = T / N; By using the finite difference method to approximate the rate of volume change, a discrete form of the expression for the rate of volume change is obtained; The objective function is transformed into a discrete summation expression, which includes the square of the rate of change of volume at each time step, the square of the pressure gradient, and the square of the deviation between the volume and the target volume at the final time step. A quadratic programming solver is used to solve the discretized objective function to obtain the optimal pressure gradient sequence. The solution process includes constraints on the maximum allowable pressure gradient, minimum safe pressure, and maximum operating frequency of the pneumatic pressurization module.

2. The intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading according to claim 1, characterized in that, The DCNN edema level assessment model is trained using a dynamically weighted multi-task loss function, which is: ， in, For Dice similarity coefficient loss; Cross-entropy loss; These are dynamic weighting coefficients; The bioimpedance distribution field at time t; Here is the diffusion coefficient matrix; It is a divergence operator; This refers to the duration of a single treatment cycle.

3. The intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading according to claim 1, characterized in that, The energy function for surface reconstruction using the 3D scanner for limb volume measurement is: ， in, For implicit surface functions; This is the original scan data; The mean curvature; For target curvature; For regularization parameters; For the scan space; The boundary of the limb surface.

4. The intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading according to claim 1, characterized in that, The multimodal sensor array adopts a six-axis spatial layout scheme; Among them, a 4×4 flexible pressure sensor matrix is ​​arranged 5cm above the ankle joint, with a sampling frequency ≥200Hz; Two sets of bioimpedance electrode pairs were placed on the anterior and posterior sides of the lower leg, and the four-electrode method was used for measurement, with an excitation frequency range of 1kHz-1MHz. Infrared temperature sensors are arranged in a spiral pattern with 1cm spacing, covering the entire calf area; The 3D scanner uses binocular structured light, with a baseline length of 150mm, a working distance of 300-500mm, and a single scan accuracy of ≤0.5mm.

5. The intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading according to claim 1, characterized in that, The intelligent pneumatic pressurization module comprises a three-level pressure regulation system consisting of a base pressure layer, a dynamic compensation layer, and a safety protection layer. The basic pressure layer consists of 12 independent air chambers, each equipped with a proportional pressure valve and a flow sensor; The dynamic compensation layer employs a microvalve array driven by shape memory alloy, with a response time ≤20ms; The safety protection layer integrates a pressure relief valve and an overpressure alarm circuit. The pressure control algorithm adopts a model predictive control framework with prediction time domain N=15 and control time domain M=5. The optimization objective function includes three dimensions: pressure tracking error, pressure change rate, and air chamber volume change. The constraints include a maximum allowable pressure gradient of 5 kPa / s, a minimum safe pressure of 3 kPa, and a maximum operating frequency of 0.5 Hz. When the detected tissue impedance change rate Z > 0.2Ω / s, the pressure holding mode is automatically triggered, maintaining the current pressure value until Z < 0.05Ω / s. The volume change compensation during the pressure holding phase is achieved through a PID controller with a proportional coefficient Kp = 0.8, integral time Ti = 30s, and derivative time Td = 5s.

6. The intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading according to claim 1, characterized in that, The training process of the DCNN edema level assessment model further includes: Shallow features of multimodal data are extracted through the initial DCNN network layers. These shallow features include local patterns of pressure distribution, spatial gradients of tissue impedance, texture information of surface temperature, and geometric contours of limb volume. By enhancing the DCNN network layers, residual connections and attention mechanisms are used to deepen the network depth and enhance the feature representation capability; The offset field network layer is used to predict the offset of each pixel, thereby registering the limb shape and position to solve the measurement error caused by changes in limb posture. Candidate regions are generated using the RPN network layer, which generates multiple anchor boxes of different sizes and proportions on the feature map through a sliding window and predicts the foreground / background probability and bounding box regression parameters for each anchor box. The Region Proposal Network layer filters out candidate regions with high confidence based on the output of the RPN, which are then used as the input to the ROI Pooling Network layer. The ROI pooling network layer maps candidate regions of different sizes to feature vectors of a fixed size.

7. A computing device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the intelligent pressure-regulating lower limb lymphedema treatment device based on dynamic grading as described in any one of claims 1-6.

Images