WO2021093745A1 - 一种肌肉疾病评估方法、系统及电子设备 - Google Patents

一种肌肉疾病评估方法、系统及电子设备 Download PDF

Info

Publication number
WO2021093745A1
WO2021093745A1 PCT/CN2020/127903 CN2020127903W WO2021093745A1 WO 2021093745 A1 WO2021093745 A1 WO 2021093745A1 CN 2020127903 W CN2020127903 W CN 2020127903W WO 2021093745 A1 WO2021093745 A1 WO 2021093745A1
Authority
WO
WIPO (PCT)
Prior art keywords
muscle
data
layer
neural network
subject
Prior art date
Application number
PCT/CN2020/127903
Other languages
English (en)
French (fr)
Inventor
肖杨
陈畅明
张湘楠
王丛知
郑海荣
Original Assignee
深圳先进技术研究院
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳先进技术研究院 filed Critical 深圳先进技术研究院
Publication of WO2021093745A1 publication Critical patent/WO2021093745A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
    • A61B8/48—Diagnostic techniques
    • A61B8/485—Diagnostic techniques involving measuring strain or elastic properties
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
    • A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
    • A61B5/1118—Determining activity level
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/45—For evaluating or diagnosing the musculoskeletal system or teeth
    • A61B5/4519—Muscles
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B8/00—Diagnosis using ultrasonic, sonic or infrasonic waves
    • A61B8/52—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves
    • A61B8/5215—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data
    • A61B8/5223—Devices using data or image processing specially adapted for diagnosis using ultrasonic, sonic or infrasonic waves involving processing of medical diagnostic data for extracting a diagnostic or physiological parameter from medical diagnostic data

Definitions

  • This application belongs to the technical field of biological signal processing, and particularly relates to a muscle disease assessment method, system and electronic equipment.
  • Muscular disorders usually refer to skeletal muscle diseases. Skeletal muscle is the main organ for body movement and an important organ for body energy metabolism. There are more than 600 muscles in the human body, which weigh about 40% of adult body weight. A non-inflammatory disease originating from the skeletal muscle or neuromuscular junction, referred to as myopathy. Mainly manifested as the decrease or disappearance of muscle contractility and muscle atrophy. Muscle diseases include myasthenia gravis, progressive muscular dystrophy, and periodic paralysis.
  • DMD muscle disease-Duchenne muscular dystrophy
  • Duchenne muscular dystrophy Duchenne muscular dystrophy
  • Symptoms are a group of primary muscle degeneration diseases, manifested as aggravated symmetrical muscle weakness and muscle atrophy, which are genetic diseases.
  • DMD is one of the most common X-linked recessive lethal genetic diseases, first reported in 1868 by the French medical scientist Duchenne. The population incidence of this disease is as high as 1/3500 live-born male infants. It is a common primary muscle disease with a poor prognosis.
  • the typical clinical features are progressive muscle atrophy, muscle weakness and pseudohypertrophy of the calf gastrocnemius muscle, usually involving adolescent males, who generally lose the ability to stand and walk before the age of 12, and eventually die before the age of due to myocardial and respiratory muscle weakness.
  • adolescent males who generally lose the ability to stand and walk before the age of 12, and eventually die before the age of due to myocardial and respiratory muscle weakness.
  • heart failure or respiratory failure This disease has severely affected the healthy growth of young men, and also brought a heavy mental and economic burden to the family and society.
  • DMD-type Duchenne muscular dystrophy have a minimum age of onset of 1 year, but they are rarely diagnosed due to lack of typical symptoms under 3 years of age. In childhood, he was often taken to a doctor by his parents because of difficulty in running or going upstairs, easy falling, and hypertrophy of the calf muscles. Most children start to show symptoms of muscle weakness at the age of 3-5, and gradually develop gait instability, a special standing posture, and pseudohypertrophy of the gastrocnemius muscle at the age of 6-8.
  • the three typical clinical features are: due to pelvic girdle Muscle weakness, muscle atrophy and hypotonia make the patient walk slowly and fall easily; because of the weakness of the iliopsoas and quadriceps muscles, it is difficult for the child to climb the ladder; the weakness of the back extensor muscles makes the child excessive lordosis of the lumbar spine when standing; Because of the weakness of the gluteus medius, the child’s pelvis swings up and down when walking, showing a typical duck gait; when the child rises from the supine position, due to the weakness of the abdominal and psoas muscles, the child must turn to prone, and then support with both hands Stand up slowly on the ground and lower limbs.
  • This posture is called Gower’s sign; free shoulders and pterygoid shoulders appear due to weakness of the scapular girdle and serratus anterior muscles; the child has pseudohypertrophy of bilateral gastrocnemius muscles, sometimes with deltoid and tongue muscles The hypertrophy and so on.
  • Respiratory muscle weakness begins at the age of 8-9, and lung capacity and total lung capacity decrease. About 40% of patients die of respiratory failure. Myocardial damage is often present in children with DMD. During the course of the disease, 50%-80% of patients have heart enlargement, tachycardia, heart failure, and abnormal electrocardiogram. A small number of children have intellectual disability. Most patients were trapped in a wheelchair before the age of 12 and died around the age of 20. At present, there is no effective treatment method for DMD, so it is very important to efficiently and accurately evaluate and detect the disease early.
  • the traditional assessment process of DMD is generally as follows: clinical suspected-serum creatine kinase increase-electromyographic examination showing myogenic damage-muscle pathological examination or genetic examination confirmed. Among them, electromyography (quadriceps femoris or gastrocnemius muscle and other parts) showed muscle-derived damage and enzyme biochemical examination found that muscle enzyme activity was significantly increased, which is the most reliable clinical evaluation basis.
  • the muscle enzymes commonly used for hoof examination include creatine kinase (CK), lactate dehydrogenase (LDH), pyruvate kinase (PK), alanine aminotransferase (GPT) and aspartate aminotransferase (GOT).
  • Creatine kinase is the same Enzyme (CKMB) also increased significantly. These enzymes not only increase in the serum activity of children, but also increase in heterozygous carriers and even amniotic fluid, so they can be used for heterozygous carrier assessment and newborn screening occasionally.
  • CKMB Enzyme
  • the DMD gene is the largest human gene known today, spanning 2500kb in the genomic DNA group, and the length of the cDNA is about 14kb.
  • the gene is composed of 79 exons and 78 introns. It encodes a protein and a product of quality.
  • the gene has the characteristics of high mutation frequency and diverse mutation forms. Therefore, genetic testing may bring some false positive results, leading to detection errors, and the process is complicated and the detection cycle is long.
  • Minghuan Zhang et al. proposed the use of magnetic resonance images (MRI) and machine learning methods to evaluate and predict DMD in "Using Machine Learning Methods to Categorize and Predict DMD for Rare Neuromuscular Diseases", and use wavelet transform to analyze the DMD patient group Wavelet decomposition was performed on the magnetic resonance image (MRI) of the healthy control group; secondly, several texture feature parameters were extracted from the resulting decomposed image and processed for dimensionality reduction; finally, based on these texture feature parameters, the support vector machine algorithm ( SVM) classifies and predicts the test images.
  • MRI magnetic resonance images
  • SVM support vector machine algorithm
  • the above method uses nuclear magnetic resonance technology and machine learning methods, and uses machine learning methods to evaluate diseases by extracting image features.
  • MRI requires higher patient compliance, because it requires the patient to stay in a fixed position for a certain period of time, the imaging speed is slow, and there is a lot of noise during the detection process.
  • the present application provides a muscle disease assessment method, system and electronic equipment, which aim to solve at least one of the above technical problems in the prior art to a certain extent.
  • a method for evaluating muscle diseases including the following steps:
  • Step a Drive passive movement of the subject through an external drive, and collect the motion angle of the subject;
  • Step b Use ultrasonic shear wave elastography technology to obtain the dynamic changes of the measured skeletal muscle under different motion angles, and extract the dynamic changes of elasticity measurement data;
  • Step c Extract the elastic modulus-joint angle correspondence relationship of the measured skeletal muscle from the elasticity measurement data, and use the elastic modulus-joint angle correspondence relationship as a feature to make a data set;
  • Step d Construct a probabilistic neural network model, and input the data set into the probabilistic neural network for training.
  • the probabilistic neural network uses the elastic modulus of skeletal muscle to analyze the corresponding relationship between the joint angle and the muscle elastic modulus of the subject in passive motion. Measure the trend of changes in the volume, and use this to evaluate the subject’s muscle disease and the type of muscle disease.
  • the technical solution adopted by the embodiment of the present application further includes: in the step a, the passive movement of the testee is driven by an external drive, and the acquisition of the motion angle of the testee specifically includes: placing the testee in a comfortable position and exposing For the joint to be measured, determine the bony landmarks of the joint to be tested, find the axis of motion, fixed arm, and mobile arm; use external force to drive the testee’s passive flexion and extension of the tested joint to complete the movement from -40 degrees to 30 degrees. And record the motion angle of the subject every 2 degrees.
  • the technical solution adopted in the embodiment of the present application further includes: in the step b, the use of ultrasonic shear wave elastography technology to obtain the dynamic changes of the measured skeletal muscle under different motion angles, and extracting the dynamic changes of elasticity measurement data specifically includes : using an ultrasonic evaluator to detect the transverse section of the vertical muscle bundle, determine the thickest position of the muscle belly, and then rotate the probe to check the long axis of the skeletal muscle along the longitudinal section along the direction of the muscle bundle, start the elastic imaging mode, fix the probe position, and continuously measure the area of interest The Young's modulus value of the muscle belly when the joint angle of the muscle tissue changes, and the dynamic ultrasonic elasticity image sequence is obtained; by measuring the Young's modulus value at least twice, the average Young's modulus value of the muscle tissue in the region of interest is calculated as the elasticity of the skeletal muscle Modulus measurement value.
  • the technical solution adopted in the embodiment of the present application further includes: in the step c, extracting the elastic modulus of the measured skeletal muscle from the elasticity measurement data-the corresponding relationship between the joint angle specifically includes: first cleaning the elasticity measurement data And processing to eliminate redundant and abnormal data; then, the elasticity measurement data is made into a 35-dimensional sample set that takes an elastic modulus every 2 degrees as a sample feature, and converts the elasticity measurement data into an elastic modulus-angle curve relationship. Label the data, get the data set, and divide the data set into ten parts, 9 of which are used as the training set data to train the probabilistic neural network in turn, and one is used as the test set data to verify the probabilistic neural network.
  • the technical solution adopted in the embodiment of the present application further includes: in the step d, the probabilistic neural network includes an input layer, a hidden layer, a summation layer, and an output layer;
  • the first layer is the input layer, which is used to receive the values from the training samples and pass the received data to the hidden layer.
  • the number of neurons is equal to the length of the input vector
  • the hidden layer of the second layer is the radial base layer.
  • Each neuron node of the hidden layer has a center, which is used to receive the sample input of the input layer, calculate the distance between the input vector and the center, and finally return a scalar value.
  • the number of elements is the same as the number of input training samples;
  • the third layer is the summation layer.
  • the number of neurons in the summation layer is equal to the number of data classifications.
  • the summation layer calculates the average value of each type of data in the above formula, and then compares the average value of each type of data. The data is classified into the class with the largest average value, and the number of neurons in the summation layer is the same as the total number of classes;
  • the last layer is the output layer, and the output layer takes the class with the largest average value in the summation layer as the output category.
  • a muscle disease assessment system including:
  • Motion angle recording module used to drive the subject's passive movement through an external drive, and collect the subject's motion angle
  • Elasticity measurement module used to obtain the dynamic changes of the measured skeletal muscle under different motion angles by using ultrasonic shear wave elastic imaging technology, and extract the elasticity measurement data of the dynamic changes;
  • Data processing module used to extract the elastic modulus-joint angle correspondence relationship of the measured skeletal muscle from the elasticity measurement data, and use the elastic modulus-joint angle correspondence relationship as a feature to create a data set;
  • Model training module used to construct a probabilistic neural network model, and input the data set into the probabilistic neural network for training.
  • the probabilistic neural network uses the elastic modulus of skeletal muscle to analyze the correspondence between the joint angle and the subject in passive motion. The trend of muscle elastic modulus changes, and use this to evaluate the subject’s muscle disease and the type of muscle disease.
  • the technical solution adopted in the embodiment of the application further includes: the motion angle recording module drives the subject to passively move through an external drive, and collecting the motion angle of the subject specifically includes: placing the subject in a comfortable position and exposing the joint to be measured , Determine the bony marks of the tested joint, find the axis of motion, fixed arm, and mobile arm; use external force to drive the tested joint passive flexion and extension to complete the movement from -40 to 30 degrees, and every 2 Record the angle of motion of the subject once in degrees.
  • the technical solution adopted in the embodiment of the application further includes: the elasticity measurement module uses the ultrasonic shear wave elastography technology to obtain the dynamic changes of the measured skeletal muscle under different motion angles, and extracting the elasticity measurement data of the dynamic changes specifically includes: adopting ultrasound evaluation
  • the instrument detects the transverse section of the vertical muscle bundle, determines the thickest position of the muscle belly, and then rotates the probe to check the long axis of the skeletal muscle along the longitudinal section in the direction of the muscle bundle, activates the elastic imaging mode, fixes the probe position, and continuously measures the joint angle of the muscle tissue in the region of interest
  • the Young's modulus value of the muscle belly is changed to obtain a dynamic ultrasound elasticity image sequence; by measuring the Young's modulus value at least twice, the average Young's modulus value of the muscle tissue in the region of interest is calculated as the elastic modulus measurement value of the skeletal muscle .
  • the technical solution adopted in the embodiment of the application further includes: the data processing module extracts the elastic modulus of the measured skeletal muscle from the elasticity measurement data-the corresponding relationship between joint angles specifically includes: first cleaning and processing the elasticity measurement data to eliminate Redundant and abnormal data; then, the elasticity measurement data is made into a 35-dimensional sample set with one elastic modulus every 2 degrees as the sample feature, and the elasticity measurement data is converted into the elastic modulus-angle curve relationship, and the data is labeled , Obtain the data set, and divide the data set into ten parts, 9 of which are used as the training set data to train the probabilistic neural network in turn, and one is used as the test set data to verify the probabilistic neural network.
  • the technical solution adopted in the embodiment of the present application further includes: the probabilistic neural network includes an input layer, a hidden layer, a summation layer, and an output layer;
  • the first layer is the input layer, which is used to receive the values from the training samples and pass the received data to the hidden layer.
  • the number of neurons is equal to the length of the input vector
  • the hidden layer of the second layer is the radial base layer.
  • Each neuron node of the hidden layer has a center, which is used to receive the sample input of the input layer, calculate the distance between the input vector and the center, and finally return a scalar value.
  • the number of elements is the same as the number of input training samples;
  • the third layer is the summation layer.
  • the number of neurons in the summation layer is equal to the number of data classifications.
  • the summation layer calculates the average value of each type of data in the above formula, and then compares the average value of each type of data. The data is classified into the class with the largest average value, and the number of neurons in the summation layer is the same as the total number of classes;
  • the last layer is the output layer, and the output layer takes the class with the largest average value in the summation layer as the output category.
  • an electronic device including:
  • At least one processor At least one processor
  • a memory communicatively connected with the at least one processor; wherein,
  • the memory stores instructions executable by the one processor, and the instructions are executed by the at least one processor, so that the at least one processor can perform the following operations of the aforementioned muscle disease assessment method:
  • Step a Drive passive movement of the subject through an external drive, and collect the motion angle of the subject;
  • Step b Use ultrasonic shear wave elastography technology to obtain the dynamic changes of the measured skeletal muscle under different motion angles, and extract the dynamic changes of elasticity measurement data;
  • Step c Extract the elastic modulus-joint angle correspondence relationship of the measured skeletal muscle from the elasticity measurement data, and use the elastic modulus-joint angle correspondence relationship as a feature to make a data set;
  • Step d Construct a probabilistic neural network model, and input the data set into the probabilistic neural network for training.
  • the probabilistic neural network uses the elastic modulus of skeletal muscle to analyze the corresponding relationship between the joint angle and the muscle elastic modulus of the subject in passive motion. Measure the trend of changes in the volume, and use this to evaluate the subject’s muscle disease and the type of muscle disease.
  • the beneficial effects produced by the embodiments of the present application are: the muscle disease assessment method, system and electronic equipment of the embodiments of the present application synchronously collect the skeletal muscle elastic imaging sequence of the testee by passively moving the limbs of the testee. Extract the dynamically changing elasticity measurement data to obtain the curve relationship between the movement angle and the elastic modulus, and use the artificial neural network to detect whether the subject has muscle disease and the type of muscle disease.
  • this application has at least the following advantages:
  • Ultrasonic detection is more convenient than CT, MRI, etc., the imaging speed is faster, non-invasive, ultrasonic detection is sensitive, and has good real-time performance.
  • Figure 1 is a flowchart of a muscle disease assessment method according to an embodiment of the present application
  • Figure 2 is a schematic diagram of the change of elastic modulus with joint angle
  • FIG. 3 is a schematic diagram of data processing
  • Figure 4 is a diagram of the PNN network structure
  • Figure 5 is a schematic diagram of model verification and parameter selection
  • Fig. 6 is a schematic structural diagram of a muscle disease assessment system according to an embodiment of the application.
  • FIG. 7 is a schematic diagram of the hardware device structure of the muscle disease assessment method provided by the embodiment of the present application.
  • FIG. 1 is a flowchart of a muscle disease assessment method according to an embodiment of the present application.
  • the muscle disease assessment method of the embodiment of the application includes the following steps:
  • Step 100 Passive movement; Drive the subject to passively move through an external drive, and synchronously and continuously collect the movement angle of the subject;
  • step 100 the specific method of passive exercise is as follows: firstly, record the name, gender, age, weight, height and other information of the testee, and establish the testee's personal information file; secondly, make the testee in a comfortable position and expose the joints to be measured , Determine the bony mark of the tested joint, find the axis of motion, fixed arm, and mobile arm; then, use external force to drive the tested joint passive flexion and extension to complete the movement from -40 degrees to 30 degrees, and The motion angle of the subject is recorded every 2 degrees.
  • Step 200 Elasticity measurement; using ultrasonic shear wave elastography technology to obtain the dynamic changes of the tested skeletal muscle under different motion angles, and extract the dynamically changing elasticity measurement data (ie, the elastic modulus corresponding to different motion angles);
  • the elasticity measurement specifically includes: firstly, use the B-mode imaging mode of the ultrasonic evaluator to detect the vertical cross section of the muscle bundle, determine the thickest position of the muscle belly, and then rotate the probe to check the long axis of the skeletal muscle along the longitudinal section of the muscle bundle. , Start the elastic imaging mode, fix the probe position, and continuously measure the Young's modulus value of the muscle belly when the joint angle of the muscle tissue changes.
  • the area of interest is 10mm ⁇ 10mm to obtain a dynamic ultrasound elasticity image sequence; by measuring the Young's modulus value multiple times , The system automatically calculates the average Young's modulus value of the muscle tissue in the area as the measured value of the skeletal muscle elastic modulus.
  • each subject measures the Young's modulus value 3 times, and then takes the average value.
  • the specific measurement times can be set according to actual applications.
  • the present application uses a non-invasive ultrasonic shear wave elastography system to quickly detect skeletal muscles. With the passive movement of the joints, the ROI area of a set size is synchronously monitored, and the change in elastic modulus caused by muscle contraction is recorded. As shown in Figure 2.
  • Step 300 Data processing, remove redundant data in the elasticity measurement data, convert the processed elasticity measurement data into the elastic modulus-joint angle correspondence relationship, and use the elastic modulus-joint angle correspondence relationship as a feature to create the data Collect and label;
  • step 300 the data processing is shown in Fig. 3, which specifically includes: first cleaning and processing the elasticity measurement data to eliminate redundant and abnormal data generated during the collection process due to operational problems; then, making the elasticity measurement data into Take an elastic modulus every 2 degrees as a 35-dimensional sample set of sample characteristics, convert the elastic measurement data into an elastic modulus-angle curve relationship, label the corresponding data, obtain a data set, and divide the data set into ten parts , 9 of them are used as the training set data to train the probabilistic neural network in turn, and one is used as the test set data to verify the probabilistic neural network.
  • Step 400 Construct a PNN (Probabilistic Neural Network) model, input the training set data into the Probabilistic Neural Network for classification training, and output the muscle disease of the testee and the category of the muscle disease;
  • PNN Probabilistic Neural Network
  • the probabilistic neural network was first proposed by Dr. D.F. Specht in 1989, and it is a neural network commonly used for pattern classification.
  • Probabilistic neural network is a neural network model based on statistical principles. It is equivalent to the optimal Bayes classifier in classification function. Its essence is a parallel algorithm developed based on the Bayesian minimum risk criterion. At the same time, it is not like traditional Multi-layer forward networks need to use the BP algorithm to calculate the backward error propagation, but a completely forward calculation process. Its training time is short, it is not easy to produce a local optimum, and its classification accuracy rate is high.
  • the PNN network structure is shown in Figure 4, which includes an input layer, a hidden layer, a summation layer, and an output layer.
  • the functions of each layer are as follows:
  • the first layer is the input layer, which is used to receive the values from the training samples and pass the received data to the hidden layer.
  • the number of neurons is equal to the length of the input vector.
  • the hidden layer of the second layer is the radial base layer.
  • Each neuron node of the hidden layer has a center (corresponding to a sample data).
  • This layer receives the sample input of the input layer, calculates the distance between the input vector and the center, and finally returns A scalar value, the number of neurons is the same as the number of input training samples.
  • the vector x is input to the hidden layer, and the input/output relationship determined by the jth neuron of the i-th pattern in the hidden layer is defined by the following formula:
  • the x ij in formula (1) is consistent with the RBF (radial basis basis) neural network, which is to find the Euclidean distance between each input and the sample, but this hidden layer divides the data into i categories, and set the i-th category There are j data.
  • the third layer is the summation layer.
  • the number of neurons in the summation layer is equal to the number of data classifications.
  • the summation layer calculates the average value of each type of data in the above formula, and then compares the average value of each type of data. The data is classified into the class with the largest average value.
  • the summation layer takes the weighted average of the output of the hidden neurons of the same category in the hidden layer:
  • Equation (2) v i represents an output category class i, L represents the number of neurons class i.
  • the number of neurons in the summation layer is the same as the total number of classes M.
  • the last layer is the output layer, and the output layer takes the largest one in the summation layer as the output category:
  • this application uses the elastic modulus-joint angle correspondence relationship as the classification feature, and uses the probabilistic neural network to classify and judge muscle diseases, that is, the probabilistic neural network uses the skeletal muscle elastic modulus-joint angle correspondence relationship to quantitatively analyze The trend of changes in the muscle elastic modulus of the subject during passive exercise to assess whether the subject has a muscle disease and the type of muscle disease, which brings a new method for the clinical evaluation of muscle diseases by computer-aided ultrasound. .
  • Step 500 Model verification and parameter selection; input the test set data into the probabilistic neural network, pass the 10-fold cross-validation of the model's evaluation accuracy, and select appropriate model parameters;
  • step 600 model verification and parameter selection are shown in Fig. 5.
  • Ten-fold cross-validation is a commonly used test method to test the accuracy of the algorithm. Each test will get the corresponding correct rate (or error rate). Take the average of the correct rate (or error rate) of 10 trials as an estimate of the accuracy of the algorithm, and then perform 10 times of 10-fold cross-validation, and then find the average value as an estimate of the accuracy of the algorithm.
  • the spread value is the radial basis expansion speed of the neural network.
  • the larger the spread the relatively smooth when the function is fitted, and at the same time the approximation error will increase accordingly, and the calculation will also increase.
  • the spread is smaller, the approximation of the function will be larger and more accurate, but at the same time, the entire approximation process of the function will be relatively unsmooth, and the image will be over-adapted.
  • Reasonable selection of the spread value is very important for the simulation results of the neural network. In this application, by setting the initial spread value to 0.01, the step size to 0.01, the maximum spread value to 2, and the confirmation parameter to 0.6, the accuracy rate reaches a stable level.
  • FIG. 6 is a schematic structural diagram of a muscle disease assessment system according to an embodiment of the present application.
  • the muscle disease assessment system of the embodiment of the present application includes a motion angle recording module, an elasticity measurement module, a data processing module, a model training module, and a model verification module.
  • Movement angle recording module used to drive the subject's passive movement through an external drive, and synchronously and continuously collect the subject's movement angle; the specific method of passive movement is as follows: first record the subject's name, gender, age, and weight , Height and other information, establish the testee’s personal information file; secondly, put the testee in a comfortable position, expose the joint to be measured, determine the bony landmarks of the tested joint, find the axis of motion, fixed arm, and mobile arm; then, use The external force drives the tested joint to passively flex and extend to complete the movement from -40 to 30 degrees, and record the angle of motion of the tested every 2 degrees.
  • Elasticity measurement module used to obtain the dynamic changes of the measured skeletal muscle under different motion angles by using ultrasonic shear wave elastography technology, and extract the elasticity measurement data of the dynamic changes (ie, the elastic modulus corresponding to different motion angles); among them, the elasticity measurement Specifically: First, use the B-mode imaging mode of the ultrasonic evaluator to detect the transverse section of the vertical muscle bundle, determine the thickest position of the muscle belly, and then rotate the probe to examine the long axis of the skeletal muscle along the longitudinal section of the muscle bundle, and activate the elastic imaging mode. Fix the probe position and continuously measure the Young's modulus value of the muscle belly when the joint angle of the muscle tissue changes.
  • the area of interest is 10mm ⁇ 10mm to obtain a dynamic ultrasound elastic image sequence; by measuring the Young's modulus value multiple times, the system automatically calculates the area
  • the average Young's modulus value of muscle tissue is used as a measurement value of the elastic modulus of skeletal muscle.
  • each subject measures the Young's modulus value 3 times, and then takes the average value.
  • the specific measurement times can be set according to actual applications.
  • Data processing module used to remove redundant data in elasticity measurement data, convert the processed elasticity measurement data into elastic modulus-joint angle correspondence, and use elastic modulus-joint angle correspondence as a feature to make data Collect and label; data processing specifically includes: first cleaning and processing the elastic measurement data to eliminate redundant and abnormal data caused by operational problems during the collection process; then, making the elastic measurement data every 2 degrees A 35-dimensional sample set with elastic modulus as the sample feature, convert the elastic measurement data into elastic modulus-angle curve relationship, label the corresponding data, get the data set, and divide the data set into ten parts, of which 9 parts As the training set data, take turns to train the probabilistic neural network, and one copy is used as the test set data to verify the probabilistic neural network.
  • Model training module used to build a PNN (probabilistic neural network) model, input the training set data into the probabilistic neural network for classification training, and output the muscle disease and the category of the muscle disease of the testee; among them, the probabilistic neural network is based on statistical principles
  • the neural network model is equivalent to the optimal Bayes classifier in classification function. Its essence is a parallel algorithm developed based on the Bayesian minimum risk criterion. At the same time, it does not need to be used like the traditional multi-layer forward network.
  • the BP algorithm performs the calculation of backward error propagation, but is a completely forward calculation process. Its training time is short, it is not easy to produce a local optimum, and its classification accuracy rate is high.
  • the PNN network structure is shown in Figure 4, which includes an input layer, a hidden layer, a summation layer, and an output layer.
  • the functions of each layer are as follows:
  • the first layer is the input layer, which is used to receive the values from the training samples and pass the received data to the hidden layer.
  • the number of neurons is equal to the length of the input vector.
  • the hidden layer of the second layer is the radial base layer.
  • Each neuron node of the hidden layer has a center (corresponding to a sample data).
  • This layer receives the sample input of the input layer, calculates the distance between the input vector and the center, and finally returns A scalar value, the number of neurons is the same as the number of input training samples.
  • the vector x is input to the hidden layer, and the input/output relationship determined by the jth neuron of the i-th pattern in the hidden layer is defined by the following formula:
  • the x ij in formula (1) is consistent with the RBF (radial basis basis) neural network, which is to find the Euclidean distance between each input and the sample, but this hidden layer divides the data into i categories, and set the i-th category There are j data.
  • the third layer is the summation layer.
  • the number of neurons in the summation layer is equal to the number of data classifications.
  • the summation layer calculates the average value of each type of data in the above formula, and then compares the average value of each type of data. The data is classified into the class with the largest average value.
  • the summation layer takes the weighted average of the output of the hidden neurons of the same category in the hidden layer:
  • Equation (2) v i represents an output category class i, L represents the number of neurons class i.
  • the number of neurons in the summation layer is the same as the total number of classes M.
  • the last layer is the output layer, and the output layer takes the largest one in the summation layer as the output category:
  • this application uses the elastic modulus-joint angle correspondence relationship as the classification feature, and uses the probabilistic neural network to classify and judge muscle diseases, that is, the probabilistic neural network uses the skeletal muscle elastic modulus-joint angle correspondence relationship to quantitatively analyze The trend of changes in the muscle elastic modulus of the subject during passive exercise to assess whether the subject has a muscle disease and the type of muscle disease, which brings a new method for the clinical evaluation of muscle diseases by computer-aided ultrasound. .
  • Model verification module used for model verification and parameter selection; input the test set data into the probabilistic neural network, evaluate the correct rate of the model through the ten-fold cross-validation, and select the appropriate model parameters; ten-fold cross-validation is a commonly used test method. To test the accuracy of the algorithm. Each test will get the corresponding correct rate (or error rate). Take the average of the correct rate (or error rate) of 10 trials as an estimate of the accuracy of the algorithm, and then perform 10 times of 10-fold cross-validation, and then find the average value as an estimate of the accuracy of the algorithm.
  • the spread value is the radial basis expansion speed of the neural network.
  • the larger the spread the relatively smooth when the function is fitted, and at the same time the approximation error will increase accordingly, and the calculation will also increase.
  • the spread is smaller, the approximation of the function will be larger and more accurate, but at the same time, the entire approximation process of the function will be relatively unsmooth, and the image will be over-adapted.
  • Reasonable selection of the spread value is very important for the simulation results of the neural network. In this application, by setting the initial spread value to 0.01, the step size to 0.01, the maximum spread value to 2, and the confirmation parameter to 0.6, the accuracy rate reaches a stable level.
  • FIG. 7 is a schematic diagram of the hardware device structure of the muscle disease assessment method provided by the embodiment of the present application.
  • the device includes one or more processors and memory. Taking a processor as an example, the device may also include: an input system and an output system.
  • the processor, the memory, the input system, and the output system may be connected by a bus or in other ways.
  • the connection by a bus is taken as an example.
  • the memory can be used to store non-transitory software programs, non-transitory computer executable programs, and modules.
  • the processor executes various functional applications and data processing of the electronic device by running the non-transitory software programs, instructions, and modules stored in the memory, that is, realizing the processing methods of the foregoing method embodiments.
  • the memory may include a program storage area and a data storage area, where the program storage area can store an operating system and an application program required by at least one function; the data storage area can store data and the like.
  • the memory may include a high-speed random access memory, and may also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices.
  • the memory may optionally include a memory remotely arranged with respect to the processor, and these remote memories may be connected to the processing system through a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.
  • the input system can receive input digital or character information, and generate signal input.
  • the output system may include display devices such as a display screen.
  • the one or more modules are stored in the memory, and when executed by the one or more processors, the following operations of any of the foregoing method embodiments are performed:
  • Step a Drive passive movement of the subject through an external drive, and collect the motion angle of the subject;
  • Step b Use ultrasonic shear wave elastography technology to obtain the dynamic changes of the measured skeletal muscle under different motion angles, and extract the dynamic changes of elasticity measurement data;
  • Step c Extract the elastic modulus-joint angle correspondence relationship of the measured skeletal muscle from the elasticity measurement data, and use the elastic modulus-joint angle correspondence relationship as a feature to make a data set;
  • Step d Construct a probabilistic neural network model, and input the data set into the probabilistic neural network for training.
  • the probabilistic neural network uses the elastic modulus of skeletal muscle to analyze the corresponding relationship between the joint angle and the muscle elastic modulus of the subject in passive motion. Measure the trend of changes in the volume, and use this to evaluate the subject’s muscle disease and the type of muscle disease.
  • the embodiment of the present application provides a non-transitory (non-volatile) computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can perform the following operations:
  • Step a Drive passive movement of the subject through an external drive, and collect the motion angle of the subject;
  • Step b Use ultrasonic shear wave elastography technology to obtain the dynamic changes of the measured skeletal muscle under different motion angles, and extract the dynamic changes of elasticity measurement data;
  • Step c Extract the elastic modulus-joint angle correspondence relationship of the measured skeletal muscle from the elasticity measurement data, and use the elastic modulus-joint angle correspondence relationship as a feature to make a data set;
  • Step d Construct a probabilistic neural network model, and input the data set into the probabilistic neural network for training.
  • the probabilistic neural network uses the elastic modulus of skeletal muscle to analyze the corresponding relationship between the joint angle and the muscle elastic modulus of the subject in passive motion. Measure the trend of changes in the volume, and use this to evaluate the subject’s muscle disease and the type of muscle disease.
  • the embodiment of the present application provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions are executed by a computer To make the computer do the following:
  • Step a Drive passive movement of the subject through an external drive, and collect the motion angle of the subject;
  • Step b Use ultrasonic shear wave elastography technology to obtain the dynamic changes of the measured skeletal muscle under different motion angles, and extract the dynamic changes of elasticity measurement data;
  • Step c Extract the elastic modulus-joint angle correspondence relationship of the measured skeletal muscle from the elasticity measurement data, and use the elastic modulus-joint angle correspondence relationship as a feature to make a data set;
  • Step d Construct a probabilistic neural network model, and input the data set into the probabilistic neural network for training.
  • the probabilistic neural network uses the elastic modulus of skeletal muscle to analyze the corresponding relationship between the joint angle and the muscle elastic modulus of the subject in passive motion. Measure the trend of changes in the volume, and use this to evaluate the subject’s muscle disease and the type of muscle disease.
  • the muscle disease assessment method, system and electronic device of the embodiments of the present application passively move the limbs of the testee, synchronously collect the testee's skeletal muscle elasticity imaging sequence, extract dynamically changing elasticity measurement data, and obtain the movement angle and the elastic modulus.
  • the relationship between the curve and the artificial neural network is used to detect whether the subject has muscle disease and the type of muscle disease.
  • Ultrasonic detection is more convenient than CT, MRI, etc., the imaging speed is faster, non-invasive, ultrasonic detection is sensitive, and has good real-time performance.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Medical Informatics (AREA)
  • Animal Behavior & Ethology (AREA)
  • Physics & Mathematics (AREA)
  • Veterinary Medicine (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Public Health (AREA)
  • Biomedical Technology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Surgery (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Dentistry (AREA)
  • Physiology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Rheumatology (AREA)
  • Orthopedic Medicine & Surgery (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Ultra Sonic Daignosis Equipment (AREA)

Abstract

一种肌肉疾病评估方法、系统及电子设备。包括：步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；步骤c：从弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；步骤d：构建概率神经网络模型，将数据集输入概率神经网络进行训练，概率神经网络输出被测者的肌肉疾病以及肌肉疾病的类别。采用人工智能神经网络，分类结果更加可靠，泛化能力更好，为肌肉疾病的评估提供了新的方法。

Description

一种肌肉疾病评估方法、系统及电子设备 技术领域
本申请属于生物信号处理技术领域，特别涉及一种肌肉疾病评估方法、系统及电子设备。
背景技术
肌肉疾病(muscular disorders)通常是指骨骼肌疾病。骨骼肌是执行机体运动的主要器官，也是机体能量代谢的重要器官，人体共600多块肌肉，其重量约占成人体重的40％。原发于骨骼肌或神经肌肉接头处的非炎症性疾病，简称肌病。主要表现为肌收缩力减退或消失以及肌肉萎缩等。肌肉疾病有重症肌无力、进行性肌营养不良、周期性瘫痪等类型。
目前，临床上针对肌肉疾病的评估基本靠医生的个人经验、电生理指标、生化指标等，以肌肉疾病--假肥大型肌营养不良症(Duchennemusculardystrophy，DMD)为例，假肥大型肌营养不良症是一组原发性肌肉变性病，表现为进行加重的对称性肌无力、肌肉萎缩，为遗传性疾病。DMD是最常见的X-连锁隐性致死性遗传病之一，由法国医学家Duchenne等于1868年首先报道。本病的群体发病率高达1/3500活产男婴，是一种预后不良的常见的原发性肌肉疾病。典型的临床特征是进行性肌萎缩、肌无力伴小腿腓肠肌的假性肥大，通常累及青少年男性，一般在12岁以前丧失站立和行走的能力，最后因心肌以及呼吸肌无力而多于岁前死于心力衰竭或呼吸衰竭。本病严重影响了青少年男性的健康成长，同时也给家庭和社会带来了沉重的精神和经济负担。
DMD型假肥大型肌营养不良症患者起病年龄最小为1岁，但是在3岁以下可因缺乏典型症状而很少被确诊。在儿童期常因跑步或上楼困难、易跌倒和小腿肌肉肥大被父母带去就诊。多数患儿在3-5岁开始表现出肌无力症状，6-8岁时逐渐出现步态不稳、特殊的起 立姿势及腓肠肌假性肥大等三大典型的临床特征，表现为：因骨盆带肌无力、肌萎缩及肌张力低下，患者走路缓慢及易跌倒；因髂腰肌和股四头肌无力而使患儿登梯困难；因背部伸肌无力使患儿站立时腰椎过度前凸；因臀中肌无力导致患儿行走时骨盆两侧上下摇摆，呈典型的鸭行步态；患儿从仰卧位起立时由于腹肌和腰肌无力，患儿必须翻转为俯卧，再以双手支持地面和下肢缓慢地站立起来，这种姿势称Gower氏征；因肩胛带肌和前锯肌无力，出现游离肩与翼状肩；患儿双侧腓肠肌假性肥大，有时还存在三角肌和舌肌的肥大等。8-9岁时开始出现呼吸肌无力，肺活量和肺总量降低，大约40％的患者死于呼吸衰竭。DMD患儿常存在心肌受损，在病程中50％-80％的患者出现心脏扩大、心动过速和心力衰竭以及心电图异常。少数患儿有智力障碍。大多数患者在12岁前囚困于轮椅，20岁左右死亡。目前针对DMD尚无有效的治疗方法，故高效准确地早期评估和检测该疾病显得非常重要。
目前针对DMD评估的方法有以下几种：
一、DMD的传统评估流程一般为：临床疑似--血清肌酸激酶增--肌电图检查呈肌源性损害--肌肉病理检查或基因检查证实。其中肌电图(股四头肌或腓肠肌等部位)显示肌源性损害及酶生化检查发现肌酶活性显著增高是临床上最可靠的评估依据。常被用于蹄查的肌酶包括肌酸激酶(CK)、乳酸脱氢酶(LDH)、丙酮酸激酶(PK)、谷丙转氨酶(GPT)和谷草转氨酶(GOT)等，肌酸激酶同工酶(CKMB)也显著增高。这些酶不但在患儿的血清中活性增高，而且在杂合子携带者甚至羊水中也升高，故而间或又可用于杂合子携带者的评估和新生儿筛查。
上述方法中，肌电图是一项创伤性检查，会带给病人带来极大不适。由于DMD患者多在幼年起病，该项检查必然导致患儿巨大的恐惧与不合作。基因检查，DMD基因是当今已知的人类最大的基因，在基因DNA组上跨越2500kb，cDNA的长度约14kb该基因由79个外显子和78个内含子构成，编码的蛋白，质产物为抗肌萎缩蛋白，该基因具有突变频率高和突变形式多样等特点，因此基因检测可能会带来一些假阳性结果，导致检测错误，而且过程复杂，检测周期长。
二、章鸣嬛等在《利用机器学习方法对神经肌肉罕见病DMD进行分类预测》中提出了用磁共振图像(MRI)和机器学习方法的方法评估和预测DMD，利用小波变换对DMD患者组和健康对照组的磁共振图像(MRI)进行小波分解；其次，从所得的分解图像中提取出若干纹理特征参数并进行降维处理；最后，再基于这些纹理特征参数，利用支持向量机算法(SVM)对试验图像进行分类预测。
上述方法采用了核磁共振技术和机器学习方法，通过提取图像的特征，运用机器学习的方法进行对疾病的评估。核磁共振相较其他检测手段，要求患者的依从性比较高，因为它需要患者，在一个固定的体位，需要维持一定的时间，成像速度较慢，且在检测过程中，有较大的噪音。
发明内容
本申请提供了一种肌肉疾病评估方法、系统及电子设备，旨在至少在一定程度上解决现有技术中的上述技术问题之一。
为了解决上述问题，本申请提供了如下技术方案：
一种肌肉疾病评估方法，包括以下步骤：
步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
步骤c：从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
步骤d：构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
本申请实施例采取的技术方案还包括：在所述步骤a中，所述通过外部驱动带动被测者被动运动，并采集被测者的运动角度具体包括：使被测者处于舒适体位，暴露待测量关节，确定被测关节的骨性标志，找到运动轴、固定臂、移动臂；采用外力驱动带动被测者 的被测关节被动屈伸，使其完成从-40度到30度的运动，并每2度记录一次被测者的运动角度。
本申请实施例采取的技术方案还包括：在所述步骤b中，所述利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据具体包括：采用超声评估仪检测垂直肌束横切面，确定肌腹最厚位置，然后旋转探头，沿肌束方向纵切面检查骨骼肌长轴，启动弹性成像模式，固定探头位置，连续测量感兴趣区域内肌肉组织关节角度变化时肌腹的杨氏模量值，获得动态超声弹性图像序列；通过至少两次测量杨式模量值，计算感兴趣区域内肌肉组织的平均杨氏模量值作为骨骼肌的弹性模量测量值。
本申请实施例采取的技术方案还包括：在所述步骤c中，所述从弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系具体包括：首先对弹性测量数据进行清洗与加工，消除冗余及异常数据；然后，将弹性测量数据制作成每2度取一个弹性模量作为样本特征的一个35维样本集合，将弹性测量数据转化为弹性模量-角度曲线关系，给数据打上标签，得到数据集，并将数据集分成十份，其中9份作为训练集数据，轮流对概率神经网络进行训练，1份作为测试集数据对概率神经网络进行验证。
本申请实施例采取的技术方案还包括：在所述步骤d中，所述概率神经网络包括输入层、隐含层、求和层和输出层；
第一层为输入层，用于接收来自训练样本的值，并将接收数据传递给隐含层，神经元个数与输入向量长度相等；
第二层的隐含层为径向基层，每一个隐含层的神经元节点拥有一个中心，用于接收输入层的样本输入，计算输入向量与中心的距离，最后返回一个标量值，神经元个数与输入训练样本个数相同；
第三层为求和层，求和层的神经元个数与数据分类的个数相等，求和层求得上式中每类数据的平均值，然后比较每类数据平均值的大小，将数据分类到平均值最大的类，求和层的神经元个数与总类数相同；
最后一层为输出层，输出层取求和层中平均值最大的类作为输出的类别。
本申请实施例采取的另一技术方案为：一种肌肉疾病评估系统，包括：
运动角度记录模块：用于通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
弹性测量模块：用于利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
数据处理模块：用于从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
模型训练模块：用于构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
本申请实施例采取的技术方案还包括：所述运动角度记录模块通过外部驱动带动被测者被动运动，并采集被测者的运动角度具体包括：使被测者处于舒适体位，暴露待测量关节，确定被测关节的骨性标志，找到运动轴、固定臂、移动臂；采用外力驱动带动被测者的被测关节被动屈伸，使其完成从-40度到30度的运动，并每2度记录一次被测者的运动角度。
本申请实施例采取的技术方案还包括：所述弹性测量模块利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据具体包括：采用超声评估仪检测垂直肌束横切面，确定肌腹最厚位置，然后旋转探头，沿肌束方向纵切面检查骨骼肌长轴，启动弹性成像模式，固定探头位置，连续测量感兴趣区域内肌肉组织关节角度变化时肌腹的杨氏模量值，获得动态超声弹性图像序列；通过至少两次测量杨式模量值，计算感兴趣区域内肌肉组织的平均杨氏模量值作为骨骼肌的弹性模量测量值。
本申请实施例采取的技术方案还包括：所述数据处理模块从弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系具体包括：首先对弹性测量数据进行清洗与加工，消除冗余及异常数据；然后，将弹性测量数据制作成每2度取一个弹性模量作为样本特征的一个35维样本集合，将弹性测量数据转化为弹性模量-角度曲线关系，给数据打上 标签，得到数据集，并将数据集分成十份，其中9份作为训练集数据，轮流对概率神经网络进行训练，1份作为测试集数据对概率神经网络进行验证。
本申请实施例采取的技术方案还包括：所述概率神经网络包括输入层、隐含层、求和层和输出层；
第一层为输入层，用于接收来自训练样本的值，并将接收数据传递给隐含层，神经元个数与输入向量长度相等；
第二层的隐含层为径向基层，每一个隐含层的神经元节点拥有一个中心，用于接收输入层的样本输入，计算输入向量与中心的距离，最后返回一个标量值，神经元个数与输入训练样本个数相同；
第三层为求和层，求和层的神经元个数与数据分类的个数相等，求和层求得上式中每类数据的平均值，然后比较每类数据平均值的大小，将数据分类到平均值最大的类，求和层的神经元个数与总类数相同；
最后一层为输出层，输出层取求和层中平均值最大的类作为输出的类别。
本申请实施例采取的又一技术方案为：一种电子设备，包括：
至少一个处理器；以及
与所述至少一个处理器通信连接的存储器；其中，
所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述的肌肉疾病评估方法的以下操作：
步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
步骤c：从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
步骤d：构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
相对于现有技术，本申请实施例产生的有益效果在于：本申请实施例的肌肉疾病评估方法、系统及电子设备通过被动活动被测者肢体，同步采集被测者的骨骼肌弹性成像序列，提取动态变化的弹性测量数据，得到运动角度与弹性模量之间的曲线关系，并通 过人工神经网络检测被测者是否患有肌肉疾病，以及肌肉疾病的类别。与现有技术相比，本申请至少具有如下优点：
1、采用超声检测，相比CT、MRI等，更加方便操作，成像速度更快，无创，超声检测敏感，具有很好的实时性。
2、采集数据方便，满足临床需要，只需检测单块骨骼肌运动即可得到评估结果。
3、采用人工智能神经网络，分类结果更加可靠，泛化能力更好，为肌肉疾病的评估提供了新的方法，扩展了计算机辅助超声评估在肌肉疾病早期评估的应用领域。
附图说明
图1是本申请实施例的肌肉疾病评估方法的流程图；
图2为弹性模量随关节角度变化示意图；
图3为数据处理示意图；
图4是PNN网络结构图；
图5为模型验证与参数选择示意图；
图6为本申请实施例的肌肉疾病评估系统的结构示意图；
图7是本申请实施例提供的肌肉疾病评估方法的硬件设备结构示意图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
请参阅图1，是本申请实施例的肌肉疾病评估方法的流程图。本申请实施例的肌肉疾病评估方法包括以下步骤：
步骤100：被动运动；通过外部驱动带动被测者被动运动，并同步、连续采集被测者的运动角度；
步骤100中，被动运动具体方法如下：首先记录被测者的姓名、性别、年龄、体重、身高等信息，建立被测者个人信息档案；其次，使被测者处于舒适体位，暴露待测量关节，确定被测关节的骨性标志，找到运动轴、固定臂、移动臂；然后，采用外力驱动带动被测者的被测关节被动屈伸，使其完成从-40度到30度的运动，并每2度记录一次被测者的运动角度。
步骤200：弹性测量；利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据(即不同运动角度对应的弹性模量)；
步骤200中，弹性测量具体包括：首先，采用超声评估仪的B-mode成像模式检测垂直肌束横切面，确定肌腹最厚位置，然后旋转探头，沿肌束方向纵切面检查骨骼肌长轴，启动弹性成像模式，固定探头位置，连续测量肌肉组织关节角度变化时肌腹的杨氏模量值，感兴趣区域大小为10mm×10mm，获得动态超声弹性图像序列；通过多次测量杨式模量值，系统自动计算区域内肌肉组织的平均杨氏模量值作为骨骼肌的弹性模量测量值。本申请实施例中，每个被测者测量杨式模量值3次，然后取平均值，具体测量次数可根据实际应用进行设定。
上述中，本申请通过采用无创的超声剪切波弹性成像系统快速检测骨骼肌，随着关节被动运动，同步监测设定大小的ROI区域，记录因肌肉收缩导致的弹性模量的改变，具体如图2所示。
步骤300：数据处理，去除弹性测量数据中的冗余数据，将处理后的弹性测量数据转化为弹性模量--关节角度对应关系，将弹性模量--关节角度对应关系作为特征，制作数据集，并打好标签；
步骤300中，数据处理如图3所示，其具体包括：首先对弹性测量数据进行清洗与加工，消除采集过程中因操作等问题产生的冗余及异常数据；然后，将弹性测量数据制作成每2度取一个弹性模量作为样本特征的一个35维样本集合，将弹性测量数据转化为弹性模量-角度曲线关系，给相应的数据打上标签，得到数据集，并将数据集分成十份，其中9份作为训练集数据，轮流对概率神经网络进行训练，1份作为测试集数据对概率神经网络进行验证。
步骤400：构建PNN(概率神经网络)模型，将训练集数据输入概率神经网络进行分类训练，输出被测者的肌肉疾病以及肌肉疾病的类别；
步骤400中，概率神经网络于1989年由D.F.Specht博士首先提出，是一种常用于模式分类的神经网络。概率神经网络是基于统计原理的神经网络模型，在分类功能上与最优Bayes分类器等价，其实质是基于贝叶斯最小风险准则发展而来的一种并行算法，同时它不像传统的多层前向网络那样需要用BP算法进行反向误差传播的计算，而是完全前向的计算过程。它训练时间短、不易产生局部最优，而且它的分类正确率较高。
具体的，PNN网络结构如图4所示，其包括输入层、隐含层、求和层和输出层。各层功能如下：
第一层为输入层，用于接收来自训练样本的值，并将接收数据传递给隐含层，神经元个数与输入向量长度相等。
第二层的隐含层为径向基层，每一个隐含层的神经元节点拥有一个中心(对应一个样本数据)，该层接收输入层的样本输入，计算输入向量与中心的距离，最后返回一个标量值，神经元个数与输入训练样本个数相同。向量x输入到隐含层，隐含层中第i类模式的第j个神经元所确定的输入/输出关系由下式定义：
Figure PCTCN2020127903-appb-000001
公式(1)中的x ij与RBF(径向基)神经网络一致，就是求每个输入与样本的欧式距离，只不过此隐含层将数据分为了i个类，并且设第i个类有j个数据。
第三层为求和层，求和层的神经元个数与数据分类的个数相等，求和层求得上式中每类数据的平均值，然后比较每类数据平均值的大小，将数据分类到平均值最大的类。
假设i＝1,2,…，M,M为训练样本中的总类数，d为样本空间数据的维数，x ij为第i类样本的第j个屮心。求和层将隐含层中属中同一类的隐含神经元的输出做加权平均：
Figure PCTCN2020127903-appb-000002
公式(2)中，v i表示第i类类别的输出，L表示第i类的神经元个数。求和层的神经元个数与总类数M相同。
最后一层为输出层，输出层取求和层中最大的一个作为输出的类别：
y＝argmax(v i)  (3)
上述中，本申请通过将弹性模量--关节角度对应关系作为分类特征，通过概率神经网络进行肌肉疾病的分类判断，即概率神经网络利用骨骼肌的弹性模量--关节角度对应关系定量分析被检测者在被动运动中，肌肉弹性模量变化的趋势，以此评估被测者是否患有肌肉疾病，以及所患肌肉疾病的类型，为计算机辅助超声评估肌肉疾病临床评估带来新的方法。
步骤500：模型验证与参数选择；将测试集数据输入概率神经网络，通过十折交叉验证模型的评估正确率，并选择合适的模型参数；
步骤600中，模型验证与参数选择如图5所示，十折交叉验证是常用的测试方法，用于测试算法准确性。每次试验都会得出相应的正确率(或差错率)。取10次试验的正确率(或差错率)的平均值作为对算法精度的估计，再进行10次10折交叉验证，再求其均值，作为对算法准确性的估计。
模型参数选择具体为：隐含层神经元各参数阈值中，spread值是神经网络的径向基扩展速度。spread越大，在函数进行拟合时就相对平滑，同时逼近误差就会随之变大，计算也会变大。相反，如果spread越小，函数的逼近程度就会越大越精准，但是，与此同时，函数的整个逼近过程则会相对的不平滑，进而会出现图像过适应的糟糕现象。合理地选择spread值对神经网络的仿真结果非常重要，本申请通过设定初始spread值为0.01,步长为0.01，spread最大值为2，确认参数为0.6时，正确率达到稳定水平。
请参阅图6，是本申请实施例的肌肉疾病评估系统的结构示意图。本申请实施例的肌肉疾病评估系统包括运动角度记录模块、弹性测量模块、数据处理模块、模型训练模块和模型验证模块。
运动角度记录模块：用于通过外部驱动带动被测者被动运动，并同步、连续采集被测者的运动角度；其中，被动运动具体方法如下：首先记录被测者的姓名、性别、年龄、体重、身高等信息，建立被测者个人信息档案；其次，使被测者处于舒适体位，暴露待测量关节，确定被测关节的骨性标志，找到运动轴、固定臂、移动臂；然后，采用 外力驱动带动被测者的被测关节被动屈伸，使其完成从-40度到30度的运动，并每2度记录一次被测者的运动角度。
弹性测量模块：用于利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据(即不同运动角度对应的弹性模量)；其中，弹性测量具体包括：首先，采用超声评估仪的B-mode成像模式检测垂直肌束横切面，确定肌腹最厚位置，然后旋转探头，沿肌束方向纵切面检查骨骼肌长轴，启动弹性成像模式，固定探头位置，连续测量肌肉组织关节角度变化时肌腹的杨氏模量值，感兴趣区域大小为10mm×10mm，获得动态超声弹性图像序列；通过多次测量杨式模量值，系统自动计算区域内肌肉组织的平均杨氏模量值作为骨骼肌的弹性模量测量值。本申请实施例中，每个被测者测量杨式模量值3次，然后取平均值，具体测量次数可根据实际应用进行设定。
数据处理模块：用于去除弹性测量数据中的冗余数据，将处理后的弹性测量数据转化为弹性模量--关节角度对应关系，将弹性模量--关节角度对应关系作为特征，制作数据集，并打好标签；数据处理具体包括：首先对弹性测量数据进行清洗与加工，消除采集过程中因操作等问题产生的冗余及异常数据；然后，将弹性测量数据制作成每2度取一个弹性模量作为样本特征的一个35维样本集合，将弹性测量数据转化为弹性模量-角度曲线关系，给相应的数据打上标签，得到数据集，并将数据集分成十份，其中9份作为训练集数据，轮流对概率神经网络进行训练，1份作为测试集数据对概率神经网络进行验证。
模型训练模块：用于构建PNN(概率神经网络)模型，将训练集数据输入概率神经网络进行分类训练，输出被测者的肌肉疾病以及肌肉疾病的类别；其中，概率神经网络是基于统计原理的神经网络模型，在分类功能上与最优Bayes分类器等价，其实质是基于贝叶斯最小风险准则发展而来的一种并行算法，同时它不像传统的多层前向网络那样需要用BP算法进行反向误差传播的计算，而是完全前向的计算过程。它训练时间短、不易产生局部最优，而且它的分类正确率较高。
具体的，PNN网络结构如图4所示，其包括输入层、隐含层、求和层和输出层。各层功能如下：
第一层为输入层，用于接收来自训练样本的值，并将接收数据传递给隐含层，神经元个数与输入向量长度相等。
第二层的隐含层为径向基层，每一个隐含层的神经元节点拥有一个中心(对应一个样本数据)，该层接收输入层的样本输入，计算输入向量与中心的距离，最后返回一个标量值，神经元个数与输入训练样本个数相同。向量x输入到隐含层，隐含层中第i类模式的第j个神经元所确定的输入/输出关系由下式定义：
Figure PCTCN2020127903-appb-000003
公式(1)中的x ij与RBF(径向基)神经网络一致，就是求每个输入与样本的欧式距离，只不过此隐含层将数据分为了i个类，并且设第i个类有j个数据。
第三层为求和层，求和层的神经元个数与数据分类的个数相等，求和层求得上式中 每类数据的平均值，然后比较每类数据平均值的大小，将数据分类到平均值最大的类。
假设i＝1,2,…，M,M为训练样本中的总类数，d为样本空间数据的维数，x ij为第i类样本的第j个屮心。求和层将隐含层中属中同一类的隐含神经元的输出做加权平均：
Figure PCTCN2020127903-appb-000004
公式(2)中，v i表示第i类类别的输出，L表示第i类的神经元个数。求和层的神经元个数与总类数M相同。
最后一层为输出层，输出层取求和层中最大的一个作为输出的类别：
y＝argmax(v i)  (3)
上述中，本申请通过将弹性模量--关节角度对应关系作为分类特征，通过概率神经网络进行肌肉疾病的分类判断，即概率神经网络利用骨骼肌的弹性模量--关节角度对应关系定量分析被检测者在被动运动中，肌肉弹性模量变化的趋势，以此评估被测者是否患有肌肉疾病，以及所患肌肉疾病的类型，为计算机辅助超声评估肌肉疾病临床评估带来新的方法。
模型验证模块：用于模型验证与参数选择；将测试集数据输入概率神经网络，通过十折交叉验证模型的评估正确率，并选择合适的模型参数；十折交叉验证是常用的测试方法，用于测试算法准确性。每次试验都会得出相应的正确率(或差错率)。取10次试验的正确率(或差错率)的平均值作为对算法精度的估计，再进行10次10折交叉验证，再求其均值，作为对算法准确性的估计。
模型参数选择具体为：隐含层神经元各参数阈值中，spread值是神经网络的径向基扩展速度。spread越大，在函数进行拟合时就相对平滑，同时逼近误差就会随之变大，计算也会变大。相反，如果spread越小，函数的逼近程度就会越大越精准，但是，与此同时，函数的整个逼近过程则会相对的不平滑，进而会出现图像过适应的糟糕现象。合理地选择spread值对神经网络的仿真结果非常重要，本申请通过设定初始spread值为0.01,步长为0.01，spread最大值为2，确认参数为0.6时，正确率达到稳定水平。
为了验证本申请的可行性和有效性，采集了DMD患者86例、正常患者31例，通过PNN进行训练分类，当spread选择0.6时，基本达到最佳，平均正确率为93.18％，相对于现有技术本申请的分类结果更加可靠。本申请可广泛应用于肌无力、肌疲劳、肌痛、肌萎缩症等多种肌肉疾病的评估，并可应用GRNN、BP等不同的神经网络。
图7是本申请实施例提供的肌肉疾病评估方法的硬件设备结构示意图。如图7所示，该设备包括一个或多个处理器以及存储器。以一个处理器为例，该设备还可以包括：输入系统和输出系统。
处理器、存储器、输入系统和输出系统可以通过总线或者其他方式连接，图7中以通过总线连接为例。
存储器作为一种非暂态计算机可读存储介质，可用于存储非暂态软件程序、非暂态计算机可执行程序以及模块。处理器通过运行存储在存储器中的非暂态软件程序、指令以及模块，从而执行电子设备的各种功能应用以及数据处理，即实现上述方法实施例的处理方法。
存储器可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储数据等。此外，存储器可以包括高速随机存取存储器，还可以包括非暂态存储器，例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施例中，存储器可选包括相对于处理器远程设置的存储器，这些远程存储器可以通过网络连接至处理系统。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入系统可接收输入的数字或字符信息，以及产生信号输入。输出系统可包括显示屏等显示设备。
所述一个或者多个模块存储在所述存储器中，当被所述一个或者多个处理器执行时，执行上述任一方法实施例的以下操作：
步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
步骤c：从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
步骤d：构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
上述产品可执行本申请实施例所提供的方法，具备执行方法相应的功能模块和有益效果。未在本实施例中详尽描述的技术细节，可参见本申请实施例提供的方法。
本申请实施例提供了一种非暂态(非易失性)计算机存储介质，所述计算机存储介质存储有计算机可执行指令，该计算机可执行指令可执行以下操作：
步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
步骤c：从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
步骤d：构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
本申请实施例提供了一种计算机程序产品，所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算机程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，使所述计算机执行以下操作：
步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
步骤c：从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
步骤d：构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
本申请实施例的肌肉疾病评估方法、系统及电子设备通过被动活动被测者肢体，同步采集被测者的骨骼肌弹性成像序列，提取动态变化的弹性测量数据，得到运动角度与弹性模量之间的曲线关系，并通过人工神经网络检测被测者是否患有肌肉疾病，以及肌肉疾病的类别。与现有技术相比，本申请至少具有如下优点：
1、采用超声检测，相比CT、MRI等，更加方便操作，成像速度更快，无创，超声检测敏感，具有很好的实时性。
2、采集数据方便，满足临床需要，只需检测单块骨骼肌运动即可得到评估结果。
3、采用人工智能神经网络，分类结果更加可靠，泛化能力更好，为肌肉疾病的评估提供了新的方法，扩展了计算机辅助超声评估在肌肉疾病早期评估的应用领域。
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本申请。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本申请中所定义的一般原理可以在不脱离本申请的精神或范围的情况下，在其它实施例中实现。因此，本申请将不会被限制于本申请所示的这些实施例，而是要符合与本申请所公开的原理和新颖特点相一致的最宽的范围。

Claims (11)

  1. 一种肌肉疾病评估方法，其特征在于，包括以下步骤：
    步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
    步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
    步骤c：从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
    步骤d：构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
  2. 根据权利要求1所述的肌肉疾病评估方法，其特征在于，在所述步骤a中，所述通过外部驱动带动被测者被动运动，并采集被测者的运动角度具体包括：使被测者处于舒适体位，暴露待测量关节，确定被测关节的骨性标志，找到运动轴、固定臂、移动臂；采用外力驱动带动被测者的被测关节被动屈伸，使其完成从-40度到30度的运动，并每2度记录一次被测者的运动角度。
  3. 根据权利要求1或2所述的肌肉疾病评估方法，其特征在于，在所述步骤b中，所述利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据具体包括：采用超声评估仪检测垂直肌束横切面，确定肌腹最厚位置，然后旋转探头，沿肌束方向纵切面检查骨骼肌长轴，启动弹性成像模式，固定探头位置，连续测量感兴趣区域内肌肉组织关节角度变化时肌腹的杨氏模量值，获得动态超声弹性图像序列；通过至少两次测量杨式模量值，计算感兴趣区域内肌肉组织的平均杨氏模量值作为骨骼肌的弹性模量测量值。
  4. 根据权利要求3所述的肌肉疾病评估方法，其特征在于，在所述步骤c中，所述从弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系具体包括：首先对弹性测量数据进行清洗与加工，消除冗余及异常数据；然后，将弹性测量数据制作成每2度取一个弹性模量作为样本特征的一个35维样本集合，将弹性测量数据转化为弹性模量-角度曲 线关系，给数据打上标签，得到数据集，并将数据集分成十份，其中9份作为训练集数据，轮流对概率神经网络进行训练，1份作为测试集数据对概率神经网络进行验证。
  5. 根据权利要求4所述的肌肉疾病评估方法，其特征在于，在所述步骤d中，所述概率神经网络包括输入层、隐含层、求和层和输出层；
    第一层为输入层，用于接收来自训练样本的值，并将接收数据传递给隐含层，神经元个数与输入向量长度相等；
    第二层的隐含层为径向基层，每一个隐含层的神经元节点拥有一个中心，用于接收输入层的样本输入，计算输入向量与中心的距离，最后返回一个标量值，神经元个数与输入训练样本个数相同；
    第三层为求和层，求和层的神经元个数与数据分类的个数相等，求和层求得上式中每类数据的平均值，然后比较每类数据平均值的大小，将数据分类到平均值最大的类，求和层的神经元个数与总类数相同；
    最后一层为输出层，输出层取求和层中平均值最大的类作为输出的类别。
  6. 一种肌肉疾病评估系统，其特征在于，包括：
    运动角度记录模块：用于通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
    弹性测量模块：用于利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
    数据处理模块：用于从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
    模型训练模块：用于构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
  7. 根据权利要求6所述的肌肉疾病评估系统，其特征在于，所述运动角度记录模块通过外部驱动带动被测者被动运动，并采集被测者的运动角度具体包括：使被测者处于舒适体位，暴露待测量关节，确定被测关节的骨性标志，找到运动轴、固定臂、移动臂；采用外力驱动带动被测者的被测关节被动屈伸，使其完成从-40度到30度的运动，并每2度记录一次被测者的运动角度。
  8. 根据权利要求6或7所述的肌肉疾病评估系统，其特征在于，所述弹性测量模块利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据具体包括：采用超声评估仪检测垂直肌束横切面，确定肌腹最厚位置，然后旋转探头，沿肌束方向纵切面检查骨骼肌长轴，启动弹性成像模式，固定探头位置，连续测量感兴趣区域内肌肉组织关节角度变化时肌腹的杨氏模量值，获得动态超声弹性图像序列；通过至少两次测量杨式模量值，计算感兴趣区域内肌肉组织的平均杨氏模量值作为骨骼肌的弹性模量测量值。
  9. 根据权利要求8所述的肌肉疾病评估系统，其特征在于，所述数据处理模块从弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系具体包括：首先对弹性测量数据进行清洗与加工，消除冗余及异常数据；然后，将弹性测量数据制作成每2度取一个弹性模量作为样本特征的一个35维样本集合，将弹性测量数据转化为弹性模量-角度曲线关系，给数据打上标签，得到数据集，并将数据集分成十份，其中9份作为训练集数据，轮流对概率神经网络进行训练，1份作为测试集数据对概率神经网络进行验证。
  10. 根据权利要求9所述的肌肉疾病评估系统，其特征在于，所述概率神经网络包括输入层、隐含层、求和层和输出层；
    第一层为输入层，用于接收来自训练样本的值，并将接收数据传递给隐含层，神经元个数与输入向量长度相等；
    第二层的隐含层为径向基层，每一个隐含层的神经元节点拥有一个中心，用于接收输入层的样本输入，计算输入向量与中心的距离，最后返回一个标量值，神经元个数与输入训练样本个数相同；
    第三层为求和层，求和层的神经元个数与数据分类的个数相等，求和层求得上式中每类数据的平均值，然后比较每类数据平均值的大小，将数据分类到平均值最大的类，求和层的神经元个数与总类数相同；
    最后一层为输出层，输出层取求和层中平均值最大的类作为输出的类别。
  11. 一种电子设备，包括：
    至少一个处理器；以及
    与所述至少一个处理器通信连接的存储器；其中，
    所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器能够执行上述1至5任一项所述的肌肉疾病评估方法的以下操作：
    步骤a：通过外部驱动带动被测者被动运动，并采集被测者的运动角度；
    步骤b：利用超声剪切波弹性成像技术获取不用运动角度下被测骨骼肌的动态变化，提取动态变化的弹性测量数据；
    步骤c：从所述弹性测量数据中提取被测骨骼肌的弹性模量--关节角度对应关系，并将弹性模量--关节角度对应关系作为特征，制作数据集；
    步骤d：构建概率神经网络模型，将所述数据集输入概率神经网络进行训练，所述概率神经网络利用骨骼肌的弹性模量--关节角度对应关系分析被测者在被动运动中肌肉弹性模量变化的趋势，并以此评估被测者的肌肉疾病以及肌肉疾病的类别。
PCT/CN2020/127903 2019-11-11 2020-11-10 一种肌肉疾病评估方法、系统及电子设备 WO2021093745A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911094835.7A CN110693526A (zh) 2019-11-11 2019-11-11 一种肌肉疾病评估方法、系统及电子设备
CN201911094835.7 2019-11-11

Publications (1)

Publication Number Publication Date
WO2021093745A1 true WO2021093745A1 (zh) 2021-05-20

Family

ID=69205734

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/127903 WO2021093745A1 (zh) 2019-11-11 2020-11-10 一种肌肉疾病评估方法、系统及电子设备

Country Status (2)

Country Link
CN (1) CN110693526A (zh)
WO (1) WO2021093745A1 (zh)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110693526A (zh) * 2019-11-11 2020-01-17 深圳先进技术研究院 一种肌肉疾病评估方法、系统及电子设备
CN112233750B (zh) * 2020-10-20 2024-02-02 吾征智能技术(北京)有限公司 一种基于咯血性状及疾病的信息匹配系统
CN112288733A (zh) * 2020-11-06 2021-01-29 深圳先进技术研究院 一种肌肉超声图像检测方法、系统、终端以及存储介质
CN113517077A (zh) * 2021-06-18 2021-10-19 东莞市人民医院 用于预测臀位外倒转术疗效的控制方法、系统和存储介质
CN113693633B (zh) * 2021-10-28 2022-02-08 深圳高性能医疗器械国家研究院有限公司 骨骼肌被动弹性系数获取装置、设备及存储介质
CN115177291B (zh) * 2022-08-01 2023-11-10 首都医科大学附属北京朝阳医院 用于识别重症监护病房获得性肌无力的方法和装置
CN115486869A (zh) * 2022-09-08 2022-12-20 深圳大学 一种用于模型小动物的肌肉超声图像分析方法

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5810731A (en) * 1995-11-13 1998-09-22 Artann Laboratories Method and apparatus for elasticity imaging using remotely induced shear wave
CN103584884A (zh) * 2013-11-07 2014-02-19 中国科学院深圳先进技术研究院 肌力评估方法及装置、肌肉康复训练跟踪评估方法及系统
WO2015104582A1 (en) * 2014-01-08 2015-07-16 Amid S.R.L. Method and device for estimation of the elastic properties of tissues, particularly muscle tissues
CN106251059A (zh) * 2016-07-27 2016-12-21 中国电力科学研究院 一种基于概率神经网络算法的电缆状态评估方法
CN108113709A (zh) * 2017-12-01 2018-06-05 深圳先进技术研究院 一种肌肉萎缩程度的检测方法、装置、设备及存储介质
CN109620270A (zh) * 2018-12-14 2019-04-16 深圳先进技术研究院 一种肌张力评估方法及装置
CN110292396A (zh) * 2018-03-22 2019-10-01 深圳迈瑞生物医疗电子股份有限公司 定量成像的预测使用
CN110693526A (zh) * 2019-11-11 2020-01-17 深圳先进技术研究院 一种肌肉疾病评估方法、系统及电子设备

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102068264B (zh) * 2009-11-23 2012-10-10 财团法人资讯工业策进会 挥击动作的肌能状态分析系统及方法
US8562529B2 (en) * 2010-06-25 2013-10-22 John C. Hill Method and system for non-invasive determination of glycogen stores
CN103027713A (zh) * 2012-12-22 2013-04-10 中国科学院深圳先进技术研究院 基于超声图像的肌肉厚度测量方法和系统
CN103417218A (zh) * 2013-08-27 2013-12-04 中山大学附属第一医院 一种上肢运动的参数采集评估系统及方法
KR20150098119A (ko) * 2014-02-19 2015-08-27 삼성전자주식회사 의료 영상 내 거짓양성 병변후보 제거 시스템 및 방법
CN106264573B (zh) * 2016-07-26 2019-03-22 吉林大学 便携式肌肉力学参数及肌肉力在体超声检测装置及方法
CN106890009A (zh) * 2017-03-17 2017-06-27 深圳大学 一种骨骼肌体积变化的自动测量方法和装置
CN108814597B (zh) * 2018-04-16 2021-09-07 深圳睿瀚医疗科技有限公司 一种基于运动信息与肌电交互的康复训练评测方法及系统
CN109820482A (zh) * 2019-03-05 2019-05-31 浙江强脑科技有限公司 肌肉状态检测方法、装置及计算机可读存储介质

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5810731A (en) * 1995-11-13 1998-09-22 Artann Laboratories Method and apparatus for elasticity imaging using remotely induced shear wave
CN103584884A (zh) * 2013-11-07 2014-02-19 中国科学院深圳先进技术研究院 肌力评估方法及装置、肌肉康复训练跟踪评估方法及系统
WO2015104582A1 (en) * 2014-01-08 2015-07-16 Amid S.R.L. Method and device for estimation of the elastic properties of tissues, particularly muscle tissues
CN106251059A (zh) * 2016-07-27 2016-12-21 中国电力科学研究院 一种基于概率神经网络算法的电缆状态评估方法
CN108113709A (zh) * 2017-12-01 2018-06-05 深圳先进技术研究院 一种肌肉萎缩程度的检测方法、装置、设备及存储介质
CN110292396A (zh) * 2018-03-22 2019-10-01 深圳迈瑞生物医疗电子股份有限公司 定量成像的预测使用
CN109620270A (zh) * 2018-12-14 2019-04-16 深圳先进技术研究院 一种肌张力评估方法及装置
CN110693526A (zh) * 2019-11-11 2020-01-17 深圳先进技术研究院 一种肌肉疾病评估方法、系统及电子设备

Also Published As

Publication number Publication date
CN110693526A (zh) 2020-01-17

Similar Documents

Publication Publication Date Title
WO2021093745A1 (zh) 一种肌肉疾病评估方法、系统及电子设备
Gandhi et al. Automation, machine learning, and artificial intelligence in echocardiography: a brave new world
US8308646B2 (en) Trainable diagnostic system and method of use
EP1832233A1 (en) Cardiac valve data measuring method and device
Mao et al. Neck sensor-supported hyoid bone movement tracking during swallowing
Chen et al. Quantitative analysis and automated lung ultrasound scoring for evaluating COVID-19 pneumonia with neural networks
CN113284126B (zh) 人工神经网络图像分析预测脑积水分流手术疗效的方法
US20210358126A1 (en) Trained model, learning method, learning program, medical information acquisition device, medical information acquisition method, and medical information acquisition program
Hossain et al. Deepbbwae-net: A cnn-rnn based deep superlearner for estimating lower extremity sagittal plane joint kinematics using shoe-mounted imu sensors in daily living
US20220370031A1 (en) Automated system and method of monitoring anatomical structures
Hanif et al. Estimation of apnea-hypopnea index using deep learning on 3-D craniofacial scans
Chia et al. A decision support system to facilitate identification of musculoskeletal impairments and propose recommendations using gait analysis in children with cerebral palsy
CN115458175A (zh) 基于数字孪生的给药剂量评估系统及其应用
Sadeghi et al. Quantifying dysfunction of the lumbar multifidus muscle after radiofrequency neurotomy and fusion surgery: a preliminary study
Ryu et al. Diagnosis of obstructive sleep apnea with prediction of flow characteristics according to airway morphology automatically extracted from medical images: computational fluid dynamics and artificial intelligence approach
Buzoianu et al. Adjusting for verification bias in diagnostic test evaluation: a Bayesian approach
Sun et al. Classification of young healthy individuals with different exercise levels based on multiple musculoskeletal ultrasound images
Bortone et al. A supervised approach to classify the status of bone mineral density in post-menopausal women through static and dynamic baropodometry
Abdolmaleki et al. Neural network analysis of breast cancer from MRI findings
Yeung et al. Rapid muscle volume prediction using anthropometric measurements and population-derived statistical models
Frigo et al. A procedure for the automatic analysis of high-resolution manometry data to support the clinical diagnosis of esophageal motility disorders
Servi et al. Outpatient monitoring of Pectus Excavatum: a Neural Network-based approach
Zhu et al. A Probabilistic Disease Progression Model for Predicting Future Clinical Outcome
Obied Different Deep Learning Techniques in Heart Disease Classification: Survey
Jaremko et al. Use of neural networks to correlate spine and rib deformity in scoliosis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20887689

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20887689

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 16.01.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20887689

Country of ref document: EP

Kind code of ref document: A1