Orthopedic rehabilitation management system based on deep learning

Through deep learning technology, the motor characteristics of orthopedic rehabilitation patients are extracted, mechanical models are established, and personalized rehabilitation suggestions are provided, which solves the problem of lack of targeted rehabilitation treatment in the existing technology, and improves the rehabilitation effect and accuracy.

CN118919018AInactive Publication Date: 2024-11-08SHENZHEN DEYI MEDICAL TECH CO LTD

Patent Information

Application Number
CN202411026744.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-30
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art cannot effectively and targetedly obtain patient exercise data and match rehabilitation suggestions, resulting in a lack of targeted orthopedic rehabilitation treatment and affecting the recovery effect.

Method used

Adopting a deep learning-based orthopedic rehabilitation management system, patient data is collected through the data acquisition module. The data processing module uses convolutional neural network and long-term memory network to extract spatial and temporal characteristics of joint motion. The data evaluation module establishes a mechanical model and a rehabilitation recommendation module. Provide personalized suggestions based on the evaluation results.

Benefits of technology

Accurate analysis and multi-dimensional evaluation of patient exercise data is achieved, targeted and accurate rehabilitation management is improved, complications are reduced, and bone healing is promoted.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118919018A_ABST
    Figure CN118919018A_ABST
Patent Text Reader

Abstract

The invention discloses an orthopedic rehabilitation management system based on deep learning, and relates to the field of medical systems. Comprising a data acquisition module used for acquiring data of a patient; the data processing module is used for analyzing the collected data by using a deep learning model and extracting the characteristics of the data; the data evaluation module is used for evaluating the extracted data features; and the rehabilitation suggestion module is used for judging the recovery condition of the patient according to a data evaluation result and providing subsequent rehabilitation suggestions according to the recovery condition. Various information of the patient is collected, then the motion model of the patient is analyzed, the recovery condition is judged, then the rehabilitation suggestions are generated according to comparison between the actual condition and the database, and therefore rehabilitation management of the patient can be carried out in a more targeted mode.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of medical systems, and in particular to an orthopedic rehabilitation management system based on deep learning. Background Art

[0002] Orthopedic rehabilitation plays a vital role in the recovery process of orthopedic diseases, such as:

[0003] Pain relief: Orthopedic injuries or diseases are often accompanied by severe pain. Rehabilitation therapy can relieve patients' pain and improve their quality of life through physical therapy, drug therapy and other means.

[0004] Restoring function: The core goal of orthopedic rehabilitation is to help patients restore their physical functions. Through rehabilitation training, patients can regain or improve their joint range of motion, muscle strength, balance, etc., so that they can complete daily activities again and return to normal life.

[0005] Prevent complications: After an orthopedic injury, patients may experience complications such as joint stiffness and muscle atrophy. Rehabilitation therapy can effectively prevent the occurrence of these complications and reduce the patient's pain and recovery time.

[0006] Promote healing: Muscle contraction during rehabilitation therapy can promote local blood and lymphatic circulation, stimulate the production of a large amount of callus tissue at the fracture end, thereby promoting bone healing and preventing bone decalcification. In addition, the slight stimulation of the fracture end during rehabilitation training can also help the healing of fractures.

[0007] However, the corresponding rehabilitation treatment plans are different for different recovery stages during the rehabilitation process. In the existing technology, when managing the entire rehabilitation population, it is impossible to effectively and specifically obtain the patient's movement data and match the rehabilitation suggestions.

[0008] To this end, the present invention proposes an orthopedic rehabilitation management system based on deep learning. Summary of the invention

[0009] The purpose of the present invention is to solve the shortcomings of the prior art and propose an orthopedic rehabilitation management system based on deep learning.

[0010] In order to achieve the above object, the present invention adopts the following technical solutions:

[0011] Orthopedic rehabilitation management system based on deep learning, including:

[0012] A data acquisition module, which is used to collect patient data;

[0013] The data processing module uses a deep learning model to analyze the collected data and extract the features of the data;

[0014] A data evaluation module, which evaluates the extracted data features;

[0015] The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

[0016] Preferably, the data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

[0017] Preferably: the data acquisition module also includes a sensor module, which is used to obtain physiological parameters of the patient, and the physiological parameters include a bone structure model and a joint movement trajectory.

[0018] Preferably: the data processing module utilizes a convolutional neural network model and a long short-term memory network model to perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory respectively.

[0019] Preferably: the convolutional neural network model includes an input layer, a convolution layer, a pooling layer and a fully connected layer, and the method for extracting spatial features includes the following steps:

[0020] A1: Define the convolution kernel and define the joint points of the joint motion data as the convolution kernel;

[0021] A2: Sliding convolution kernel, slide the convolution kernel of the joint on the length of the bone. Each slide will perform a dot product operation on the convolution kernel and the input data at the current position to obtain an output value;

[0022] A3: Add bias and activation function. After the dot product operation, add bias term and introduce nonlinear association through activation function.

[0023] A4: Generate feature maps by sliding the convolution kernel of the joint point in the length direction of the bone, and combining the bias term and activation function to obtain the feature map of each sliding point;

[0024] A5: The mapped features are processed by average pooling in the pooling layer to reduce the dimension;

[0025] A6: After passing through multiple convolutional layers and pooling layers, one or more fully connected layers are connected; each neuron in the fully connected layer is connected to all neurons in the previous layer, which is used to perform nonlinear combination of the previously extracted features to obtain the final output.

[0026] Preferably: the expression of the bias term in step A4 is f(x)=max(0,x), which means that if the input is greater than 0, the output is equal to the input; if the input is less than or equal to 0, the output is 0, and the expression of the activation function is f(x)=1 / (1+exp(-x)), which maps the input to an output between 0 and 1, where exp(-x) represents the natural exponential function, exp(-x)=e -x .

[0027] Preferably, the long short-term memory network model includes an input layer, an extraction layer and an output layer, and the method for extracting time features from data includes the following steps:

[0028] B1: Data input, input the spatially extracted features into the input layer of the LSTM network model, and establish a data model (batch_size, time_steps, input_dim) at the input layer, where batch_size is the number of samples in the batch, time_steps is the time step, and input_dim is the number of input features at each time step;

[0029] B2: loop iteration, calculate the output of the forget gate, input gate and output gate according to the input data of the current time step, the hidden state of the previous time step and their respective bias terms;

[0030] B3: Update the cell state, use the output of the forget gate to discard unnecessary information in the cell state, and use the output of the input gate to add new information to update the cell state;

[0031] B4: Calculate the hidden state and use the output of the output gate to control which information in the cell state should be used to calculate the hidden state of the current time step;

[0032] B5: Pass the hidden state, passing the hidden state of the current time step to the LSTM unit of the next time step as the input of the next time step;

[0033] B6: After the iteration cycle, the result is output.

[0034] Preferably: the evaluation logic of the data evaluation module includes the following steps:

[0035] C1: Establish a mechanical model based on the extracted spatial and temporal characteristics of joint motion, i.e., a = dv / dt;

[0036] C2: Based on Newton's second law F=ma, determine the force applied by the muscles at the joint, where F is the force applied by the muscle, m is the mass of the bone moving simultaneously with the joint, and a is the acceleration of the joint movement;

[0037] C3: Obtain the characteristic combination of muscle force according to the muscle force magnitude F and time characteristics, and determine the patient's movement characteristics according to the characteristic combination;

[0038] C4: Recovery prediction is made based on the movement characteristics and the severity of the patient’s condition. This process can be done through big data analysis or manual review by doctors.

[0039] Preferably: in step C3, the patient's movement characteristics include:

[0040] Work power P, P = Σ(dF*dv), where F is the force applied by the muscle and v is the speed of muscle movement;

[0041] The amount of work W, W = Σ(dF*dS) or W = Σ(dP*dt), where F is the strength of the muscle, S is the distance the muscle moves, P is the power of the muscle work, and t is the time of work;

[0042] Mean force where dF i is the force applied per unit time;

[0043] Force variance S 2 , where dF i is the force per unit time, is the mean applied force;

[0044] Force mode

[0045] Any combination of .

[0046] Preferably, the establishment logic of the rehabilitation suggestion module is:

[0047] D1: Doctors write recommendations based on various medical conditions in previous diagnosis history;

[0048] D2: Form the suggestions corresponding to each situation into a separate data block;

[0049] D3: Then match the data block according to the evaluation results of the data evaluation module;

[0050] D4: Return the matching suggestions to the patient.

[0051] The beneficial effects of the present invention are:

[0052] 1. The present invention collects various information about the patient and then analyzes the movement model to determine the recovery status, and then generates rehabilitation suggestions based on the comparison between the actual situation and the database, so as to carry out the patient's rehabilitation management in a more targeted manner.

[0053] 2. The present invention adopts a convolutional neural network model and a long short-term memory network model for analyzing the patient's motion information, which can respectively extract the spatial characteristics and temporal characteristics of the motion information, and then establish a kinematic model based on this, so as to perform a more accurate analysis of the patient's actual movement.

[0054] 3. The present invention adopts multi-dimensional evaluation of work power, work amount, force mean, force variance and force mode for the kinematic model, which can well reflect the total amount of patient movement, movement stability and other data, thereby increasing the accuracy of analysis and laying the foundation for subsequent matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is the architecture diagram of the orthopedic rehabilitation management system based on deep learning proposed by the present invention;

[0056] Figure 2 This is a flow chart of spatial feature extraction of the orthopedic rehabilitation management system based on deep learning proposed by the present invention;

[0057] Figure 3 This is a flow chart of time feature extraction of the orthopedic rehabilitation management system based on deep learning proposed by the present invention;

[0058] Figure 4 This is the evaluation logic flow chart of the deep learning-based orthopedic rehabilitation management system proposed in the present invention. DETAILED DESCRIPTION

[0059] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0060] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", and "set" should be understood in a broad sense, for example, they can be fixedly connected or set, or detachably connected or set, or integrally connected or set. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0061] Embodiment 1:

[0062] Orthopedic rehabilitation management system based on deep learning, which includes:

[0063] A data acquisition module, which is used to collect patient data;

[0064] The data processing module uses a deep learning model to analyze the collected data and extract the features of the data;

[0065] A data evaluation module, which evaluates the extracted data features;

[0066] The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

[0067] The data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

[0068] The data acquisition module also includes a sensor module, which is used to obtain the patient's physiological parameters, including a bone structure model and a joint movement trajectory.

[0069] The data processing module uses a convolutional neural network model and a long short-term memory network model to respectively perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory.

[0070] Embodiment 2:

[0071] Orthopedic rehabilitation management system based on deep learning, which includes:

[0072] A data acquisition module, which is used to collect patient data;

[0073] The data processing module uses a deep learning model to analyze the collected data and extract the features of the data;

[0074] A data evaluation module, which evaluates the extracted data features;

[0075] The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

[0076] The data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

[0077] The data acquisition module also includes a sensor module, which is used to obtain the patient's physiological parameters, including a bone structure model and a joint movement trajectory.

[0078] The data processing module uses a convolutional neural network model and a long short-term memory network model to respectively perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory.

[0079] The convolutional neural network model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer, and the method for extracting spatial features includes the following steps:

[0080] A1: Define the convolution kernel and define the joint points of the joint motion data as the convolution kernel;

[0081] A2: Sliding convolution kernel, slide the convolution kernel of the joint on the length of the bone. Each slide will perform a dot product operation on the convolution kernel and the input data at the current position to obtain an output value;

[0082] A3: Add bias and activation function. After the dot product operation, add bias term and introduce nonlinear association through activation function.

[0083] A4: Generate feature maps by sliding the convolution kernel of the joint point in the length direction of the bone, and combining the bias term and activation function to obtain the feature map of each sliding point;

[0084] A5: The mapped features are processed by average pooling in the pooling layer to reduce the dimension;

[0085] A6: After passing through multiple convolutional layers and pooling layers, one or more fully connected layers are connected; each neuron in the fully connected layer is connected to all neurons in the previous layer, which is used to perform nonlinear combination of the previously extracted features to obtain the final output.

[0086] The expression of the bias term in step A4 is f(x)=max(0,x), which means that if the input is greater than 0, the output is equal to the input; if the input is less than or equal to 0, the output is 0. The expression of the activation function is f(x)=1 / (1+exp(-x)), which maps the input to an output between 0 and 1, where exp(-x) represents the natural exponential function, exp(-x)=e -x .

[0087] The long short-term memory network model includes an input layer, an extraction layer and an output layer, and the method for extracting time features from data includes the following steps:

[0088] B1: Data input, input the spatially extracted features into the input layer of the LSTM network model, and establish a data model (batch_size, time_steps, input_dim) at the input layer, where batch_size is the number of samples in the batch, time_steps is the time step, and input_dim is the number of input features at each time step;

[0089] B2: loop iteration, calculate the output of the forget gate, input gate and output gate according to the input data of the current time step, the hidden state of the previous time step and their respective bias terms;

[0090] B3: Update the cell state, use the output of the forget gate to discard unnecessary information in the cell state, and use the output of the input gate to add new information to update the cell state;

[0091] B4: Calculate the hidden state and use the output of the output gate to control which information in the cell state should be used to calculate the hidden state of the current time step;

[0092] B5: Pass the hidden state, passing the hidden state of the current time step to the LSTM unit of the next time step as the input of the next time step;

[0093] B6: After the iteration cycle, the result is output.

[0094] Embodiment 3:

[0095] Orthopedic rehabilitation management system based on deep learning, which includes:

[0096] A data acquisition module, which is used to collect patient data;

[0097] The data processing module uses a deep learning model to analyze the collected data and extract the features of the data;

[0098] A data evaluation module, which evaluates the extracted data features;

[0099] The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

[0100] The data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

[0101] The data acquisition module also includes a sensor module, which is used to obtain the patient's physiological parameters, including a bone structure model and a joint movement trajectory.

[0102] The data processing module uses a convolutional neural network model and a long short-term memory network model to respectively perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory.

[0103] The convolutional neural network model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer, and the method for extracting spatial features includes the following steps:

[0104] A1: Define the convolution kernel and define the joint points of the joint motion data as the convolution kernel;

[0105] A2: Sliding convolution kernel, slide the convolution kernel of the joint on the length of the bone. Each slide will perform a dot product operation on the convolution kernel and the input data at the current position to obtain an output value;

[0106] A3: Add bias and activation function. After the dot product operation, add bias term and introduce nonlinear association through activation function.

[0107] A4: Generate feature maps by sliding the convolution kernel of the joint point in the length direction of the bone, and combining the bias term and activation function to obtain the feature map of each sliding point;

[0108] A5: The mapped features are processed by average pooling in the pooling layer to reduce the dimension;

[0109] A6: After passing through multiple convolutional layers and pooling layers, one or more fully connected layers are connected; each neuron in the fully connected layer is connected to all neurons in the previous layer, which is used to perform nonlinear combination of the previously extracted features to obtain the final output.

[0110] The expression of the bias term in step A4 is f(x)=max(0,x), which means that if the input is greater than 0, the output is equal to the input; if the input is less than or equal to 0, the output is 0. The expression of the activation function is f(x)=1 / (1+exp(-x)), which maps the input to an output between 0 and 1, where exp(-x) represents the natural exponential function, exp(-x)=e -x .

[0111] The long short-term memory network model includes an input layer, an extraction layer and an output layer, and the method for extracting time features from data includes the following steps:

[0112] B1: Data input, input the spatially extracted features into the input layer of the LSTM network model, and establish a data model (batch_size, time_steps, input_dim) at the input layer, where batch_size is the number of samples in the batch, time_steps is the time step, and input_dim is the number of input features at each time step;

[0113] B2: loop iteration, calculate the output of the forget gate, input gate and output gate according to the input data of the current time step, the hidden state of the previous time step and their respective bias terms;

[0114] B3: Update the cell state, use the output of the forget gate to discard unnecessary information in the cell state, and use the output of the input gate to add new information to update the cell state;

[0115] B4: Calculate the hidden state and use the output of the output gate to control which information in the cell state should be used to calculate the hidden state of the current time step;

[0116] B5: Pass the hidden state, passing the hidden state of the current time step to the LSTM unit of the next time step as the input of the next time step;

[0117] B6: After the iteration cycle, the result is output.

[0118] The evaluation logic of the data evaluation module includes the following steps:

[0119] C1: Establish a mechanical model based on the extracted spatial and temporal characteristics of joint motion, i.e., a = dv / dt;

[0120] C2: Based on Newton's second law F=ma, determine the force applied by the muscles at the joint, where F is the force applied by the muscle, m is the mass of the bone moving simultaneously with the joint, and a is the acceleration of the joint movement;

[0121] C3: Obtain the characteristic combination of muscle force according to the muscle force magnitude F and time characteristics, and determine the patient's movement characteristics according to the characteristic combination;

[0122] C4: Recovery prediction is made based on the movement characteristics and the severity of the patient’s condition. This process can be done through big data analysis or manual review by doctors.

[0123] The establishment logic of the rehabilitation suggestion module is:

[0124] D1: Doctors write recommendations based on various medical conditions in previous diagnosis history;

[0125] D2: The recommended itinerary corresponding to each situation is a separate data block;

[0126] D3: Then match the data block according to the evaluation results of the data evaluation module;

[0127] D4: Return the matching suggestions to the patient.

[0128] Embodiment 4:

[0129] Orthopedic rehabilitation management system based on deep learning, which includes:

[0130] A data acquisition module, which is used to collect patient data;

[0131] The data processing module uses a deep learning model to analyze the collected data and extract the features of the data;

[0132] A data evaluation module, which evaluates the extracted data features;

[0133] The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

[0134] The data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

[0135] The data acquisition module also includes a sensor module, which is used to obtain the patient's physiological parameters, including a bone structure model and a joint movement trajectory.

[0136] The data processing module uses a convolutional neural network model and a long short-term memory network model to respectively perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory.

[0137] The convolutional neural network model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer, and the method for extracting spatial features includes the following steps:

[0138] A1: Define the convolution kernel and define the joint points of the joint motion data as the convolution kernel;

[0139] A2: Sliding convolution kernel, slide the convolution kernel of the joint on the length of the bone. Each slide will perform a dot product operation on the convolution kernel and the input data at the current position to obtain an output value;

[0140] A3: Add bias and activation function. After the dot product operation, add bias term and introduce nonlinear association through activation function.

[0141] A4: Generate feature maps by sliding the convolution kernel of the joint point in the length direction of the bone, and combining the bias term and activation function to obtain the feature map of each sliding point;

[0142] A5: The mapped features are processed by average pooling in the pooling layer to reduce the dimension;

[0143] A6: After passing through multiple convolutional layers and pooling layers, one or more fully connected layers are connected; each neuron in the fully connected layer is connected to all neurons in the previous layer, which is used to perform nonlinear combination of the previously extracted features to obtain the final output.

[0144] The expression of the bias term in step A4 is f(x)=max(0,x), which means that if the input is greater than 0, the output is equal to the input; if the input is less than or equal to 0, the output is 0. The expression of the activation function is f(x)=1 / (1+exp(-x)), which maps the input to an output between 0 and 1, where exp(-x) represents the natural exponential function, exp(-x)=e -x .

[0145] The long short-term memory network model includes an input layer, an extraction layer and an output layer, and the method for extracting time features from data includes the following steps:

[0146] B1: Data input, input the spatially extracted features into the input layer of the LSTM network model, and establish a data model (batch_size, time_steps, input_dim) at the input layer, where batch_size is the number of samples in the batch, time_steps is the time step, and input_dim is the number of input features at each time step;

[0147] B2: loop iteration, calculate the output of the forget gate, input gate and output gate according to the input data of the current time step, the hidden state of the previous time step and their respective bias terms;

[0148] B3: Update the cell state, use the output of the forget gate to discard unnecessary information in the cell state, and use the output of the input gate to add new information to update the cell state;

[0149] B4: Calculate the hidden state and use the output of the output gate to control which information in the cell state should be used to calculate the hidden state of the current time step;

[0150] B5: Pass the hidden state, passing the hidden state of the current time step to the LSTM unit of the next time step as the input of the next time step;

[0151] B6: After the iteration cycle, the result is output.

[0152] The evaluation logic of the data evaluation module includes the following steps:

[0153] C1: Establish a mechanical model based on the extracted spatial and temporal characteristics of joint motion, i.e., a = dv / dt;

[0154] C2: Based on Newton's second law F=ma, determine the force applied by the muscles at the joint, where F is the force applied by the muscle, m is the mass of the bone moving simultaneously with the joint, and a is the acceleration of the joint movement;

[0155] C3: Obtain the characteristic combination of muscle force according to the muscle force magnitude F and time characteristics, and determine the patient's movement characteristics according to the characteristic combination;

[0156] C4: Recovery prediction is made based on the movement characteristics and the severity of the patient’s condition. This process can be done through big data analysis or manual review by doctors.

[0157] The establishment logic of the rehabilitation suggestion module is:

[0158] D1: Doctors write recommendations based on various medical conditions in previous diagnosis history;

[0159] D2: Form the suggestions corresponding to each situation into a separate data block;

[0160] D3: Then match the data block according to the evaluation results of the data evaluation module;

[0161] D4: Return the matching suggestions to the patient.

[0162] In the C3 step, the patient's movement characteristics include:

[0163] Work power P, P = ∑ (dF * dv), where F is the force applied by the muscle and v is the speed of muscle movement;

[0164] The amount of work done W, W = ∑ (dF * dS) or W = ∑ (dP * dt), where F is the strength of the muscle, S is the distance the muscle moves, P is the power of the muscle work, and t is the time of work.

[0165] Embodiment 5:

[0166] Orthopedic rehabilitation management system based on deep learning, which includes:

[0167] A data acquisition module, which is used to collect patient data;

[0168] The data processing module uses a deep learning model to analyze the collected data and extract the features of the data;

[0169] A data evaluation module, which evaluates the extracted data features;

[0170] The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

[0171] The data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

[0172] The data acquisition module also includes a sensor module, which is used to obtain the patient's physiological parameters, including a bone structure model and a joint movement trajectory.

[0173] The data processing module uses a convolutional neural network model and a long short-term memory network model to respectively perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory.

[0174] The convolutional neural network model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer, and the method for extracting spatial features includes the following steps:

[0175] A1: Define the convolution kernel and define the joint points of the joint motion data as the convolution kernel;

[0176] A2: Sliding convolution kernel, slide the convolution kernel of the joint on the length of the bone. Each slide will perform a dot product operation on the convolution kernel and the input data at the current position to obtain an output value;

[0177] A3: Add bias and activation function. After the dot product operation, add bias term and introduce nonlinear association through activation function.

[0178] A4: Generate feature maps by sliding the convolution kernel of the joint point in the length direction of the bone, and combining the bias term and activation function to obtain the feature map of each sliding point;

[0179] A5: The mapped features are processed by average pooling in the pooling layer to reduce the dimension;

[0180] A6: After passing through multiple convolutional layers and pooling layers, one or more fully connected layers are connected; each neuron in the fully connected layer is connected to all neurons in the previous layer, which is used to perform nonlinear combination of the previously extracted features to obtain the final output.

[0181] The expression of the bias term in step A4 is f(x)=max(0,x), which means that if the input is greater than 0, the output is equal to the input; if the input is less than or equal to 0, the output is 0. The expression of the activation function is f(x)=1 / (1+exp(-x)), which maps the input to an output between 0 and 1, where exp(-x) represents the natural exponential function, exp(-x)=e -x .

[0182] The long short-term memory network model includes an input layer, an extraction layer and an output layer, and the method for extracting time features from data includes the following steps:

[0183] B1: Data input, input the spatially extracted features into the input layer of the LSTM network model, and establish a data model (batch_size, time_steps, input_dim) at the input layer, where batch_size is the number of samples in the batch, time_steps is the time step, and input_dim is the number of input features at each time step;

[0184] B2: loop iteration, calculate the output of the forget gate, input gate and output gate according to the input data of the current time step, the hidden state of the previous time step and their respective bias terms;

[0185] B3: Update the cell state, use the output of the forget gate to discard unnecessary information in the cell state, and use the output of the input gate to add new information to update the cell state;

[0186] B4: Calculate the hidden state and use the output of the output gate to control which information in the cell state should be used to calculate the hidden state of the current time step;

[0187] B5: Pass the hidden state, passing the hidden state of the current time step to the LSTM unit of the next time step as the input of the next time step;

[0188] B6: After the iteration cycle, the result is output.

[0189] The evaluation logic of the data evaluation module includes the following steps:

[0190] C1: Establish a mechanical model based on the extracted spatial and temporal characteristics of joint motion, i.e., a = dv / dt;

[0191] C2: Based on Newton's second law F=ma, determine the force applied by the muscles at the joint, where F is the force applied by the muscle, m is the mass of the bone moving simultaneously with the joint, and a is the acceleration of the joint movement;

[0192] C3: Obtain the characteristic combination of muscle force according to the muscle force magnitude F and time characteristics, and determine the patient's movement characteristics according to the characteristic combination;

[0193] C4: Recovery prediction is made based on the movement characteristics and the severity of the patient’s condition. This process can be done through big data analysis or manual review by doctors.

[0194] The establishment logic of the rehabilitation suggestion module is:

[0195] D1: Doctors write recommendations based on various medical conditions in previous diagnosis history;

[0196] D2: Form the suggestions corresponding to each situation into a separate data block;

[0197] D3: Then match the data block according to the evaluation results of the data evaluation module;

[0198] D4: Return the matching suggestions to the patient.

[0199] In the C3 step, the patient's movement characteristics include:

[0200] Work power P, P = ∑ (dF * dv), where F is the force applied by the muscle and v is the speed of muscle movement;

[0201] The amount of work W, W = ∑ (dF * dS) or W = ∑ (dP * dt), where F is the strength of the muscle, S is the distance the muscle moves, P is the power of the muscle work, and t is the time of work;

[0202] Mean force where dF i is the force applied per unit time;

[0203] Force variance S 2 , where dF i is the force per unit time, is the mean applied force.

[0204] Embodiment 6:

[0205] Orthopedic rehabilitation management system based on deep learning, which includes:

[0206] A data acquisition module, which is used to collect patient data;

[0207] The data processing module uses a deep learning model to analyze the collected data and extract the features of the data;

[0208] A data evaluation module, which evaluates the extracted data features;

[0209] The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

[0210] The data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

[0211] The data acquisition module also includes a sensor module, which is used to obtain the patient's physiological parameters, including a bone structure model and a joint movement trajectory.

[0212] The data processing module uses a convolutional neural network model and a long short-term memory network model to respectively perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory.

[0213] The convolutional neural network model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer, and the method for extracting spatial features includes the following steps:

[0214] A1: Define the convolution kernel and define the joint points of the joint motion data as the convolution kernel;

[0215] A2: Sliding convolution kernel, slide the convolution kernel of the joint on the length of the bone. Each slide will perform a dot product operation on the convolution kernel and the input data at the current position to obtain an output value;

[0216] A3: Add bias and activation function. After the dot product operation, add bias term and introduce nonlinear association through activation function.

[0217] A4: Generate feature maps by sliding the convolution kernel of the joint point in the length direction of the bone, and combining the bias term and activation function to obtain the feature map of each sliding point;

[0218] A5: The mapped features are processed by average pooling in the pooling layer to reduce the dimension;

[0219] A6: After passing through multiple convolutional layers and pooling layers, one or more fully connected layers are connected; each neuron in the fully connected layer is connected to all neurons in the previous layer, which is used to perform nonlinear combination of the previously extracted features to obtain the final output.

[0220] The expression of the bias term in step A4 is f(x)=max(0,x), which means that if the input is greater than 0, the output is equal to the input; if the input is less than or equal to 0, the output is 0. The expression of the activation function is f(x)=1 / (1+exp(-x)), which maps the input to an output between 0 and 1, where exp(-x) represents the natural exponential function, exp(-x)=e -x .

[0221] The long short-term memory network model includes an input layer, an extraction layer and an output layer, and the method for extracting time features from data includes the following steps:

[0222] B1: Data input, input the spatially extracted features into the input layer of the LSTM network model, and establish a data model (batch_size, time_steps, input_dim) at the input layer, where batch_size is the number of samples in the batch, time_steps is the time step, and input_dim is the number of input features at each time step;

[0223] B2: loop iteration, calculate the output of the forget gate, input gate and output gate according to the input data of the current time step, the hidden state of the previous time step and their respective bias terms;

[0224] B3: Update the cell state, use the output of the forget gate to discard unnecessary information in the cell state, and use the output of the input gate to add new information to update the cell state;

[0225] B4: Calculate the hidden state and use the output of the output gate to control which information in the cell state should be used to calculate the hidden state of the current time step;

[0226] B5: Pass the hidden state, passing the hidden state of the current time step to the LSTM unit of the next time step as the input of the next time step;

[0227] B6: After the iteration cycle, the result is output.

[0228] The evaluation logic of the data evaluation module includes the following steps:

[0229] C1: Establish a mechanical model based on the extracted spatial and temporal characteristics of joint motion, i.e., a = dv / dt;

[0230] C2: Based on Newton's second law F=ma, determine the force applied by the muscles at the joint, where F is the force applied by the muscle, m is the mass of the bone moving simultaneously with the joint, and a is the acceleration of the joint movement;

[0231] C3: Obtain the characteristic combination of muscle force according to the muscle force magnitude F and time characteristics, and determine the patient's movement characteristics according to the characteristic combination;

[0232] C4: Recovery prediction is made based on the movement characteristics and the severity of the patient’s condition. This process can be done through big data analysis or manual review by doctors.

[0233] The establishment logic of the rehabilitation suggestion module is:

[0234] D1: Doctors write recommendations based on various medical conditions in previous diagnosis history;

[0235] D2: Form the suggestions corresponding to each situation into a separate data block;

[0236] D3: Then match the data block according to the evaluation results of the data evaluation module;

[0237] D4: Return the matching suggestions to the patient.

[0238] In the C3 step, the patient's movement characteristics include:

[0239] Work power P, P = ∑ (dF * dv), where F is the force applied by the muscle and v is the speed of muscle movement;

[0240] The amount of work W, W = ∑ (dF * dS) or W = ∑ (dP * dt), where F is the strength of the muscle, S is the distance the muscle moves, P is the power of the muscle work, and t is the time of work;

[0241] Mean force where dF i is the force applied per unit time;

[0242] Force variance S 2 , where dF i is the force per unit time, is the mean applied force;

[0243] Force mode

[0244] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. Orthopedic rehabilitation management system based on deep learning, characterized by: include: A data acquisition module, which is used to collect patient data; The data processing module uses a deep learning model to analyze the collected data and extract the features of the data; A data evaluation module, which evaluates the extracted data features; The rehabilitation advice module determines the patient's recovery status based on the results of data evaluation and provides subsequent rehabilitation advice based on the recovery status.

2. The orthopedic rehabilitation management system based on deep learning according to claim 1, characterized in that: The data collection module collects patient data including the patient's basic information, medical history, operation records, and rehabilitation training records.

3. The orthopedic rehabilitation management system based on deep learning according to claim 2 is characterized in that: The data acquisition module also includes a sensor module, which is used to obtain the patient's physiological parameters, including a bone structure model and a joint movement trajectory.

4. The orthopedic rehabilitation management system based on deep learning according to claim 3 is characterized in that: The data processing module uses a convolutional neural network model and a long short-term memory network model to respectively perform spatial feature extraction and temporal feature extraction on the collected joint motion trajectory.

5. The orthopedic rehabilitation management system based on deep learning according to any one of claims 1 to 4, characterized in that: The convolutional neural network model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer, and the method for extracting spatial features includes the following steps: A1: Define the convolution kernel and define the joint points of the joint motion data as the convolution kernel; A2: Sliding convolution kernel, slide the convolution kernel of the joint on the length of the bone. Each slide will perform a dot product operation on the convolution kernel and the input data at the current position to obtain an output value; A3: Add bias and activation function. After the dot product operation, add bias term and introduce nonlinear association through activation function. A4: Generate feature maps by sliding the convolution kernel of the joint point in the length direction of the bone, and combining the bias term and activation function to obtain the feature map of each sliding point; A5: The mapped features are processed by average pooling in the pooling layer to reduce the dimension; A6: After passing through multiple convolutional layers and pooling layers, one or more fully connected layers are connected; each neuron in the fully connected layer is connected to all neurons in the previous layer, which is used to perform nonlinear combination of the previously extracted features to obtain the final output.

6. The orthopedic rehabilitation management system based on deep learning according to claim 5, characterized in that: The expression of the bias term in step A4 is f(x)=max(0,x), which means that if the input is greater than 0, the output is equal to the input; If the input is less than or equal to 0, the output is 0. The expression of the activation function is f(x) = 1 / (1+exp(-x)), which maps the input to an output between 0 and 1, where exp(-x) represents the natural exponential function, exp(-x) = e -x .

7. The orthopedic rehabilitation management system based on deep learning according to any one of claims 1 to 4, characterized in that: The long short-term memory network model includes an input layer, an extraction layer and an output layer, and the method for extracting time features from data includes the following steps: B1: Data input, input the spatially extracted features into the input layer of the LSTM network model, and establish a data model (batch_size, time_steps, input_dim) at the input layer. Batch_size is the number of samples in the batch, time_steps is the time step, and input_dim is the number of input features in each time step. B2: loop iteration, calculate the output of the forget gate, input gate and output gate according to the input data of the current time step, the hidden state of the previous time step and their respective bias terms; B3: Update the cell state, use the output of the forget gate to discard unnecessary information in the cell state, and use the output of the input gate to add new information to update the cell state; B4: Calculate the hidden state and use the output of the output gate to control which information in the cell state should be used to calculate the hidden state of the current time step; B5: Pass the hidden state, passing the hidden state of the current time step to the LSTM unit of the next time step as the input of the next time step; B6: After the iteration cycle, the result is output.

8. The orthopedic rehabilitation management system based on deep learning according to claim 1, characterized in that: The evaluation logic of the data evaluation module includes the following steps: C1: Establish a mechanical model based on the extracted spatial and temporal characteristics of joint motion, i.e., a = dv / dt; C2: Based on Newton's second law F=ma, determine the force applied by the muscles at the joint, where F is the force applied by the muscle, m is the mass of the bone moving simultaneously with the joint, and a is the acceleration of the joint movement; C3: Obtain the characteristic combination of muscle force according to the muscle force magnitude F and time characteristics, and determine the patient's movement characteristics according to the characteristic combination; C4: Recovery prediction is made based on the movement characteristics and the severity of the patient’s condition. This process can be done through big data analysis or manual review by doctors.

9. The orthopedic rehabilitation management system based on deep learning according to claim 8, characterized in that: In the C3 step, the patient's movement characteristics include: Work power P, P = ∑ (dF * dv), where F is the force applied by the muscle and v is the speed of muscle movement; The amount of work W, W = Σ(dF*dS) or W = Σ(dP*dt), where F is the strength of the muscle, S is the distance the muscle moves, P is the power of the muscle work, and t is the time of work; Mean force where dF i is the force applied per unit time; Force variance S 2 , where dF i is the force per unit time, is the mean applied force; Force mode Any combination of .

10. The orthopedic rehabilitation management system based on deep learning according to claim 1, characterized in that: The establishment logic of the rehabilitation suggestion module is: D1: Doctors write recommendations based on various medical conditions in previous diagnosis history; D2: Form the suggestions corresponding to each situation into a separate data block; D3: Then match the data block according to the evaluation results of the data evaluation module; D4: Return the matching suggestions to the patient.

Citation Information

Patent Citations

  • Stroke patient hand rehabilitation training action recognition method based on array myoelectricity

    CN111938660A

  • Deep neural network plug-in based on attention mechanism and image recognition method

    CN112580777A

  • Rehabilitation method and system based on brain-computer interface technology of motion observation

    CN112998725A

  • Rehabilitation evaluation method, system, device and storage medium

    CN113436736A

  • Cerebral palsy hand function rating method based on multi-mode signal analysis and myoelectricity mode recognition

    CN116898460A

Cited By

  • Exercise rehabilitation data prediction system based on deep learning

    CN121483489A

  • Deep learning based motion rehabilitation data prediction system

    CN121483489B