Skeletal muscle state evaluation method and system based on multi-modal ultrasound

Abstract

The invention relates to the technical field of ultrasonic evaluation, in particular to a skeletal muscle state evaluation method and system based on multi-modal ultrasound, and the evaluation method comprises the following steps: collecting skeletal muscle multi-modal ultrasonic images after treatment twice through an ultrasonic detection instrument, the multi-modal ultrasonic image comprises an ultrasonic B ultrasonic image, a color ultrasound image and an elastic imaging image; acquiring a first target parameter from the multi-modal ultrasonic image acquired for the first time, and acquiring a second target parameter from the multi-modal ultrasonic image acquired for the second time; according to the first target parameter and the second target parameter, generating a repeated detection instruction on the ultrasonic detection instrument, or generating a real-time target parameter; and a hidden Markov model is adopted to obtain a hidden sequence about the real-time target parameters, and the hidden sequence and part of the real-time target parameters are used as input of an evaluation network to generate a rehabilitation effect analysis result.

Description

Technical Field

[0001] The present invention relates to the technical field of ultrasonic evaluation, and in particular to a skeletal muscle status evaluation method and system based on multimodal ultrasound. Background Art

[0002] Cerebral infarction is a type of ischemic hypoxic necrosis of local brain tissue caused by various reasons. It has the characteristics of high morbidity, high disability rate, high recurrence rate and high mortality rate. Cerebral infarction has become a major public health problem that endangers people's health. According to statistics, there are about 2 million new cases of cerebral infarction in my country each year. After cerebral infarction, patients will quickly develop changes corresponding to neurological deficits.

[0003] Among them, spasticity is one of the changes in neurological function deficits. It is a syndrome of abnormally increased muscle tension after damage to the central nervous system. It is a movement dysfunction characterized by speed-dependent tension stretch hyperreflexia caused by increased excitability of stretch reflex. At this time, the lower limb movement patterns are mostly passive, and the body's ability to control movement is weakened or lost. The tension or sexual stretch reflex is overly active, manifested as muscle spasm, contracture, decreased muscle strength and reduced range of joint motion, leading to movement dysfunction.

[0004] Although acupuncture treatment during the spastic period can promote the recovery of the nervous system, some acupoints in traditional acupuncture are not only not conducive to reducing muscle tension, but will aggravate the formation of spasticity patterns when the patient's muscle tension has increased. Studies have confirmed that the feather angle, muscle thickness and cross-sectional area of ​​skeletal muscle can reflect the size of muscle strength, and the length of muscle fibers determines the range of muscle stretch. Muscle thickness, feather angle, and cross-sectional area are positively correlated with muscle strength, and muscle fiber length is negatively correlated with spasticity. Therefore, effective treatment of skeletal muscle during the spastic period of cerebral infarction becomes crucial.

[0005] Clinically, acupuncture is an important means of TCM treatment of patients with cerebral infarction. The antagonistic acupuncture method proposed by Professor Sun Shentian, a master of traditional Chinese medicine, simulates the movement pattern that is difficult for patients to complete actively, strengthens the muscle strength of the antagonistic muscles to correct the abnormal muscle strength of the spastic muscles, and activates the responsible brain area of ​​the brain to inhibit the excitability of the spinal motor neurons of the spastic side muscle group, thereby improving the lower limb muscle spasms. However, how to accurately quantify and objectively evaluate the improvement of various parameters of skeletal muscles in the contracture period of patients with cerebral infarction by antagonistic acupuncture is the key to proving the efficacy of this acupuncture method. At present, the evaluation of antagonistic acupuncture on skeletal muscles of patients with cerebral infarction in the spastic period at home and abroad is usually more dependent on the doctor's experience and the patient's subjective feelings, and lacks objective quantitative indicators to evaluate the treatment effect. Therefore, in the process of using antagonistic acupuncture to treat patients with cerebral infarction in the contracture period, there are often problems such as inaccurate evaluation, which brings certain troubles and inconveniences to patients. Summary of the invention

[0006] In view of the above-mentioned prior art, the present invention provides a skeletal muscle status assessment method and system based on multimodal ultrasound, which mainly solves the technical problems existing in the above-mentioned background technology.

[0007] To achieve the above object, the technical solution of the embodiment of the present invention is implemented as follows: The first aspect of the present invention discloses a skeletal muscle status assessment method based on multimodal ultrasound, which is used to assess the therapeutic effect of antagonistic acupuncture, and the assessment method comprises the following steps: The multimodal ultrasound images of the skeletal muscles after treatment are collected twice by an ultrasound detection instrument, wherein the multimodal ultrasound images include ultrasound B-ultrasound images, color ultrasound images, and elastic imaging images; Acquire a first target parameter from the multimodal ultrasound image acquired for the first time, and acquire a second target parameter from the multimodal ultrasound image acquired for the second time; According to the first target parameter and the second target parameter, generating a repeated detection instruction on the ultrasonic detection instrument, or generating a real-time target parameter; A hidden Markov model is used to obtain a hidden sequence of real-time target parameters, and the hidden sequence and part of the real-time target parameters are used as inputs of an evaluation network to generate a rehabilitation effect analysis result.

[0008] Optionally, generating a repeated detection instruction on the ultrasonic detection instrument according to the first target parameter and the second target parameter, or generating a real-time target parameter, specifically includes: Comparing a first target parameter obtained from a multimodal ultrasound image acquired for the first time with a second target parameter obtained from a multimodal ultrasound image acquired for the second time to generate deviation information; Comparing the deviation information with a deviation threshold, and generating a repeated detection instruction if the deviation information is greater than the deviation threshold; If the deviation information is less than or equal to the deviation threshold, an average value of the first target parameter and the second target parameter is calculated, and real-time detection result information is generated according to the average value.

[0009] Optionally, after a repeated detection instruction is formed, the operator collects the multimodal ultrasound image of the skeletal muscle after treatment for the third time, and obtains the third target parameter therefrom, and uses the average value of the first target parameter, the second target parameter, and the third target parameter as the real-time detection result information.

[0010] Optionally, the first target parameter, the second target parameter, and the third target parameter all include muscle thickness, muscle fiber length, feathering angle, muscle cross-sectional area, shear wave velocity, and microblood flow imaging depth.

[0011] Optionally, extracting the distance between the superficial and deep fascia from the multimodal ultrasound image representing the belly of the skeletal muscle as muscle thickness index data; The linear distance between the cross section of the muscle fiber bundle and the aponeurosis at the intersection of the lateral tibial condyle and the lateral malleolus was extracted from the multimodal ultrasound image of the muscle belly representing the midpoint between the proximal tibia and the lateral malleolus as the muscle fiber length index data; Extract the acute angle between the aponeurosis and the measured pennaceous muscle bundle from the multimodal ultrasound image representing the pennaceous muscle bundle as the pennaceous angle index data; The cross-sectional area of ​​the thickest part of the medial gastrocnemius muscle belly is extracted from the multimodal ultrasound image representing the medial gastrocnemius muscle belly as the cross-sectional area index data; Shear wave velocity of skeletal muscle was obtained by ultrasound elastography; Obtain blood flow imaging depth of skeletal muscle through microblood flow imaging.

[0012] Optionally, a hidden Markov model is used to obtain a hidden sequence of real-time target parameters, specifically including: Set the spasm period, acute remission period, early remission period, mid-term recovery period, long-term stable period, relapse warning period, and deterioration warning period as hidden states, and form a hidden state sequence based on the set hidden states: ( ),in The spasm period, In acute remission, In the early remission period, The mid-term recovery period. For long-term stability, The relapse warning period It is the early warning period for deterioration; The muscle thickness, muscle fiber length, feather angle, and muscle cross-sectional area are set as observation parameters, and the observation parameters are discretized to form the observation parameter expression: ( ), multiple observation parameters form an observation sequence: ( ),in represents the observation parameters at time t, represents the muscle thickness at time t, represents the muscle fiber length at time t, represents the muscle thickness at time t, represents the feather angle at time t; Setting model parameters of the hidden Markov model, wherein the model parameters include initial state probability, state transition probability matrix, and observation probability matrix; Based on the model parameters, the Viterbi algorithm is used to infer the The most likely hidden sequence .

[0013] Optionally, apply a discrete wavelet transform to hide the sequence The sequence is decomposed into a periodic component and a non-periodic component, and an inverse discrete wavelet transform is performed on the periodic component and the non-periodic component respectively to obtain a periodic hidden sequence and a non-periodic hidden sequence.

[0014] Optionally, the evaluation network includes a periodic component branch, a non-periodic component branch, and a fully connected layer. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the periodic component branch, and the non-periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the non-periodic component branch. The outputs of the periodic component branch and the non-periodic component branch are spliced ​​and fused and input into the fully connected layer, and finally the fully connected layer outputs the rehabilitation effect analysis results.

[0015] Optionally, the periodic component branch includes a multiple dilated convolution layer, a global pooling layer, and an LSTM layer. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the multiple dilated convolution layer. The receptive field is gradually expanded to capture dependencies on different time scales. The output results of the multiple dilated convolution layer are compressed into a vector of a fixed length through the global pooling layer, and the time dependency in the vector is captured through the LSTM layer. The non-periodic component branch also includes multiple dilated convolutional layers, global pooling layers, and GRU layers. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the multiple dilated convolutional layers. The receptive field is gradually expanded to capture dependencies on different time scales. The output results of the multiple dilated convolutional layers are compressed into vectors of fixed length through the global pooling layer, and the time dependencies in the vectors are captured through the GRU layer.

[0016] The second aspect of the present invention discloses a skeletal muscle state assessment system based on multimodal ultrasound. The assessment system is used to implement any of the assessment methods described above, and the assessment system comprises: A parameter generation module, configured to obtain a first target parameter and a second target parameter from the multimodal ultrasound image; A repetition detection module, configured to generate repetition detection instructions, or, to generate real-time target parameters; The hidden Markov module is configured to obtain a hidden sequence about a real-time target parameter based on a hidden Markov model; The evaluation module is configured to generate a rehabilitation effect analysis result based on the output result of the hidden Markov module and some real-time target parameters.

[0017] The beneficial effects of the present invention are: by measuring multiple key parameters such as muscle thickness, muscle fiber length, feather angle, muscle cross-sectional area, shear wave velocity, micro-blood flow imaging depth, etc., a comprehensive and specific quantitative analysis of skeletal muscle state is achieved, the influence of human factors is reduced, and the accuracy and reliability of the evaluation results are improved; secondly, the present invention adopts a method of collecting multimodal ultrasonic images of skeletal muscles after treatment in two times, and a third collection is performed when necessary to reduce instantaneous changes caused by factors such as muscle instant contraction state or blood circulation fluctuations. By setting a deviation threshold and comparing the data collected twice, abnormal conditions can be discovered in time and repeated detection instructions can be triggered, thereby ensuring that the real-time target parameters finally generated have higher accuracy. This method not only improves the reliability of a single measurement, but also enhances the stability of the overall evaluation; in addition, by analyzing the real-time target parameters using a hidden Markov model, the present invention can identify the potential physiological or pathological states (such as spasm period, acute remission period, early remission period, etc.) hidden behind the observed data, and predict future recovery trends or warn of possible recurrence or deterioration. By using discrete wavelet transform to decompose hidden sequences into periodic components and non-periodic components, and processing dependencies on different time scales respectively, the evaluation network can more accurately capture long-term trends and short-term fluctuations, and provide more sophisticated analysis of rehabilitation effects. This dynamic monitoring and prediction function helps doctors better understand the patient's rehabilitation process, and then adjust the treatment plan to achieve personalized medical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the process of skeletal muscle status assessment method based on multimodal ultrasound in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The technical solution of the present invention is further elaborated in detail below in conjunction with the drawings and specific embodiments of the specification. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In the following description, the expression "some embodiments" is related to a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0020] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present invention. However, it is apparent to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present invention, some technical features well known in the art are not described.

[0021] It should be understood that the present invention can be implemented in different forms and should not be interpreted as being limited to the embodiments proposed herein. On the contrary, providing these embodiments will make the disclosure thorough and complete, and the scope of the present invention will be fully conveyed to those skilled in the art. And the purpose of the terms used herein is only to describe specific embodiments and is not intended to be a limitation of the present invention. When used herein, the singular forms of "one", "one" and "said / the" are also intended to include plural forms, unless the context clearly indicates another way. It should also be understood that the terms "compose" and / or "include" when used in this specification determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0022] It should also be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "inside", "outside", "left", "right" and similar expressions used herein are for illustrative purposes only and are not intended to be the only implementation method.

[0023] In order to fully understand the present invention, a detailed structure will be proposed in the following description to illustrate the technical solution proposed by the present invention. The optional embodiments of the present invention are described in detail as follows, but in addition to these detailed descriptions, the present invention may also have other implementations.

[0024] Please refer to the attached Figure 1 The present application provides a skeletal muscle status assessment method based on multimodal ultrasound, which is used to evaluate the therapeutic effect of antagonistic acupuncture. During the operation of antagonistic acupuncture, the acupoints are selected as follows: refer to "Acupuncture and Moxibustion" Head acupuncture: take the anterior temporal oblique line on the affected side (located on the side of the head, starting from the front shencong and ending at xuanli), motor area (the area 0.5 cm behind the midpoint of the anterior and posterior midline and connected with the brow-occipital line and the zygomatic arch vertical line), sensory area (1.5 cm behind the motor area, the area connecting the midline and the brow-occipital line) Body acupuncture: take the affected side's Jianyu, Jianlao, Jianzhen, Hegu, Shousanli, Quchi, Chize, Neiguan, Zusanli, Yanglingquan, Fenglong, Sanyinjiao, and Taichong.

[0025] The acupuncture operation process is as follows: routine disinfection of the local skin of the acupoints, when acupuncture Yanglingquan, vertically insert the needle 25 mm~40 mm slightly below the standard position of the original acupoint, deep to the peroneus longus muscle; Qiuxu acupoint is vertically punctured 15mm~25mm, and the foot is dorsiflexed when the needle is inserted, and the depth of the needle goes straight through the opposite Zhaohai acupoint; after acupuncture, connect the KWD-808I type electroacupuncture instrument, Shubo (frequency 2Hz), the current is adjusted according to the patient's tolerance, the upper limbs are connected to Quchi-Hegu, the lower limbs are connected to Yanglingquan-Qiuxu, and the foot is everted by electricity; Jiexi-Xuanzhong, the foot is dorsiflexed by electricity, and the remaining acupoints are conventionally acupunctured, the needle is retained for 30min, twice a day, 6 times a week, rest for 1 day, and a total treatment of 4 weeks.

[0026] After each treatment, the rehabilitation effect is evaluated using the evaluation method disclosed in this application, wherein the evaluation method comprises the following steps: S1. After treatment, multimodal ultrasound images of skeletal muscle after treatment are collected twice by ultrasound detection instrument, and the multimodal ultrasound images include ultrasound B-ultrasound images, color ultrasound images, and elastic imaging images, wherein the time interval between the two collections is 30 seconds; S2, acquiring a first target parameter from the multimodal ultrasound image acquired for the first time, and acquiring a second target parameter from the multimodal ultrasound image acquired for the second time; S3, generating a repeated detection instruction on the ultrasonic detection instrument according to the first target parameter and the second target parameter, or generating a real-time target parameter; S4. Using a hidden Markov model, a hidden sequence of real-time target parameters is obtained, and the hidden sequence and part of the real-time target parameters are used as inputs of an evaluation network to generate a rehabilitation effect analysis result.

[0027] Specifically, the state of skeletal muscle may be affected by many factors, such as the instantaneous contraction state of the muscle, fluctuations in blood circulation, etc. These instantaneous changes may cause the image acquired in a single time to fail to truly reflect the overall condition of the muscle. By acquiring twice, the error caused by instantaneous variation can be reduced, ensuring that the acquired data is more stable and reliable. After two acquisitions, the ultrasonic detection instrument can output the corresponding first target parameter and second target parameter according to the acquired multimodal ultrasonic image of skeletal muscle. Based on the output first target parameter and second target parameter, the ultrasonic detection instrument will generate repeated detection instructions, or generate real-time target parameters, specifically including: Comparing a first target parameter obtained from a multimodal ultrasound image acquired for the first time with a second target parameter obtained from a multimodal ultrasound image acquired for the second time to generate deviation information; In an optional implementation, the difference between the first target parameter and the second target parameter is calculated to obtain deviation information.

[0028] Comparing the deviation information with a deviation threshold, and generating a repeated detection instruction if the deviation information is greater than the deviation threshold; After the ultrasonic detection instrument generates deviation information, the ultrasonic detection instrument may compare the deviation information with a preset deviation threshold, wherein the deviation threshold may be a preset value set in advance by the user; if the deviation information is worse than the threshold, the ultrasonic detection instrument may generate a repeated detection instruction. After generating the repeated detection instruction, the operator needs to operate the ultrasonic detection instrument again to collect the patient's skeletal muscle multimodal ultrasonic images in accordance with the requirements of the repeated detection instruction.

[0029] If the deviation information is less than or equal to the deviation threshold, an average value of the first target parameter and the second target parameter is calculated, and real-time detection result information is generated according to the average value.

[0030] Specifically, if the deviation information is less than or equal to the deviation threshold, then it can be considered that the results of the two acquisitions are relatively consistent and have high reliability. At this time, the average value of the first target parameter and the second target parameter is calculated, and the real-time detection result information is generated according to the average value.

[0031] In an optional embodiment, after a repeated detection instruction is formed, the operator collects the multimodal ultrasound image of the skeletal muscle after treatment for the third time, and obtains the third target parameter therefrom, and uses the average value of the first target parameter, the second target parameter, and the third target parameter as the real-time detection result information.

[0032] Specifically, by collecting data three times and calculating the average value, the stability and reliability of the data can be further enhanced. First, adding one more collection means more data points for analysis, which helps reduce the impact of random errors and instantaneous changes on the results. For example, the instantaneous contraction state of a muscle or the fluctuation of blood circulation may change in a short period of time, and three collections can average these instantaneous factors to a certain extent, providing a more stable and representative measurement result.

[0033] In addition, the average of multiple independent measurements is usually more stable than a single measurement, and its standard error is also reduced. This means that the real-time test result information generated based on the average of three acquisitions has higher accuracy and credibility. For ultrasound images, noise is inevitable, but through averaging processing, the noise level can be reduced to a certain extent, the signal-to-noise ratio can be improved, and the final test results can be clearer and more reliable.

[0034] In addition, three acquisitions can better capture any potential trends or abnormalities. If the results of the first and second acquisitions are very close, but the third acquisition shows significant differences, this may indicate that there are some unforeseen factors, such as changes in the patient's body position, equipment failure, or other external interference. In this case, the doctor can re-evaluate the test conditions or take additional examination measures to ensure the accuracy of the diagnosis.

[0035] In some optional embodiments, the first target parameter, the second target parameter, and the third target parameter all include muscle thickness, muscle fiber length, feathering angle, muscle cross-sectional area, shear wave velocity, and microblood flow imaging depth.

[0036] Furthermore, muscle thickness, muscle fiber length, pennation angle, muscle cross-sectional area, shear wave velocity, and micro-blood flow imaging depth are obtained as follows: The probe is parallel to the long axis of the calf. After the image is stable and clear, the image is fixed. The measuring cursor is placed on the upper and lower parallel strong echo signal bands in ultrasound imaging. The ultrasound probe extracts the distance between the superficial and deep fascia in the multimodal ultrasound image as the muscle thickness index data. The linear distance between the cross section of the muscle fiber bundle and the aponeurosis at the intersection of the lateral tibial condyle and the lateral malleolus was extracted from the multimodal ultrasound image of the muscle belly representing the midpoint between the proximal tibia and the lateral malleolus as the muscle fiber length index data; Extract the acute angle between the aponeurosis and the measured pennaceous muscle bundle from the multimodal ultrasound image representing the pennaceous muscle bundle as the pennaceous angle index data; The cross-sectional area of ​​the thickest part of the medial gastrocnemius muscle belly is extracted from the multimodal ultrasound image representing the medial gastrocnemius muscle belly as the cross-sectional area index data; The shear wave velocity of skeletal muscle is obtained by ultrasonic elastography. The specific method is as follows: the ultrasonic probe vertically presses the surface of the tissue to collect the radio frequency signal, and the strain distribution in the tissue is analyzed to obtain the shear wave velocity; The blood flow imaging depth of skeletal muscle was obtained by micro-blood flow imaging. The specific method was as follows: measurement was performed at 1 / 2 of the length of the gastrocnemius muscle, and the blood flow imaging depth was divided into mild, moderate, and severe.

[0037] In an optional embodiment, a hidden Markov model is used to obtain a hidden sequence of real-time target parameters, wherein the actually measured parameters (such as muscle thickness, muscle fiber length, cross-sectional area, etc.) are obtained. The hidden Markov model is a statistical model suitable for describing a set of observable output sequences generated by a set of unobservable hidden state sequences. The hidden sequence usually contains multiple hidden states, each of which corresponds to a potential physiological or pathological state.

[0038] Obtaining the hidden sequence of real-time target parameters specifically includes the following steps: S401. Set the spasm period, acute remission period, early remission period, mid-term recovery period, long-term stable period, relapse warning period, and deterioration warning period as hidden states. Hidden states represent the potential physiological or pathological state of muscles at different time points. These states cannot be directly observed, but can be indirectly inferred through the measurement values ​​of multiple parameters. Based on the set hidden states, a hidden state sequence is formed: ( ),in The spasm period, In acute remission, In the early remission period, The mid-term recovery period. For long-term stability, The relapse warning period It is the early warning period for deterioration; Among them, the spasm period refers to the state when the patient may show obvious muscle spasms, accompanied by muscle stiffness, pain, limited joint movement and other symptoms. It is generally used to indicate the initial state of the patient before treatment; the acute remission period refers to the immediate reaction period after antagonistic acupuncture treatment. At this time, the acupuncture produces direct mechanical stimulation to the muscles, resulting in local inflammatory reactions and short-term pain aggravation, but then the spasm symptoms begin to ease; the early remission period refers to the state in which the patient shows a significant reduction in muscle spasm symptoms and the muscle tension gradually returns to normal. At this time, the patient's pain is further relieved and the range of joint movement is significantly expanded; the mid-term recovery period refers to the patient's muscle spasm symptoms are further relieved, and muscle strength and motor function are gradually restored. At this time, the patient's muscle tension returns to normal, the range of joint movement is fully restored, and the daily living ability is significantly improved; the long-term stable period refers to the patient's muscle structure and function remain stable, and the treatment effect is maintained for a long time. At this time, the patient's muscle spasm symptoms completely disappear, muscle strength and motor function return to normal, and the quality of life is significantly improved; the recurrence warning period refers to the patient's muscle recovery, but there is still a certain risk of recurrence. At this time, the patient may need to continue rehabilitation training or regular check-ups to prevent recurrence of the disease; the deterioration warning period refers to the state in which the patient fails to perform rehabilitation training on time or other adverse factors occur during the treatment process, the muscle condition may deteriorate, and the spasm symptoms may worsen again.

[0039] In addition, in the hidden Markov model, the hidden states can be transformed into each other, for example, they can change from the spasm period to the mid-recovery period.

[0040] S402, setting muscle thickness, muscle fiber length, feather angle, and muscle cross-sectional area as observation parameters, and performing discretization on the observation parameters to form an observation parameter expression: ( ), multiple observation parameters form an observation sequence: ( ),in represents the observation parameters at time t, represents the muscle thickness at time t, represents the muscle fiber length at time t, represents the muscle thickness at time t, represents the feather angle at time t; S403, setting the model parameters of the hidden Markov model, the model parameters include the initial state probability, the state transition probability matrix, the observation probability matrix, wherein the initial state probability represents the state in the hidden state at time t=0. The probability of transitioning from one hidden state to another is represented by the state transition probability matrix, which is an N×N matrix, where N is the number of hidden states. The observation probability matrix represents the probability of observing a specific observation parameter in a given hidden state, and the observation probability matrix is ​​an N×M matrix, where M is the number of observation parameters. S404, based on the model parameters, using the Viterbi algorithm to infer the The most likely hidden sequence , whose hidden sequence The calculation formula is: ,in ,in is the state transition probability matrix, which represents the probability of transitioning from one hidden state to another. is the state transition probability matrix, is the initial state probability.

[0041] The calculation process of the Viterbi algorithm is as follows: For each hidden state , calculate the probability of the initial state , for t=2,3,...,T, calculate each hidden state The maximum probability of: , and finally find the maximum probability of the final state and determine the final state.

[0042] In an alternative embodiment, a discrete wavelet transform is applied to the hidden sequence The sequence is decomposed into a periodic component and a non-periodic component, and an inverse discrete wavelet transform is performed on the periodic component and the non-periodic component respectively to obtain a periodic hidden sequence and a non-periodic hidden sequence.

[0043] Specifically, the discrete wavelet transform is applied to transform the hidden sequence When decomposing into periodic components and non-periodic components, wavelet basis functions including Morlet wavelet, Daubechies wavelet, and Symlets wavelet are selected as the mother wavelet function, and f(x)=x(t) is set as the hidden sequence output by the hidden Markov model. , will hide the sequence Expanded under the wavelet basis function, its expression is:

[0044] Where e is the scale parameter, r is the translation parameter, Represents the hidden sequence output by the hidden Markov model , The function is The complex conjugate function of is a mother wavelet that satisfies the admissibility condition, represents continuous wavelet transform, represents the wavelet basis function, dt represents the integral about t, and dx represents the integral about x.

[0045] Use different filters to perform convolution operations on the expanded function to obtain approximate coefficients and detail factor , for the approximate coefficient and detail factor Downsample to get new approximate coefficients and the new detail factor , for the new approximate coefficient and the new detail factor Continue to repeat the decomposition to obtain higher-level approximate coefficients and detail coefficients until the final approximate coefficients and detail coefficients are obtained after J-layer decomposition. The periodic components are mainly concentrated in the lower-frequency approximate coefficients. These coefficients reflect the long-term trend and periodic changes in the hidden state sequence, and the periodic components are mainly concentrated in the higher-frequency detail coefficients. These coefficients reflect the short-term fluctuations and mutations in the hidden state sequence. The final periodic component and non-periodic component expressions are as follows:

[0046] in and All represent the filter coefficients of the orthogonal wavelet basis, is an orthogonal scaling function, k is a hidden sequence The serial number of , Z represents an integer.

[0047] In an optional embodiment, the evaluation network includes a periodic component branch, a non-periodic component branch, and a fully connected layer. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the periodic component branch, and the non-periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the non-periodic component branch. The outputs of the periodic component branch and the non-periodic component branch are spliced ​​and fused and input into the fully connected layer, and finally the fully connected layer outputs the rehabilitation effect analysis results.

[0048] In an optional embodiment, the periodic component branch includes multiple dilated convolution layers, a global pooling layer, and an LSTM layer. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the multiple dilated convolution layers. The receptive field is gradually expanded to capture dependencies on different time scales. The output results of the multiple dilated convolution layers are compressed into a vector of fixed length through the global pooling layer, and the time dependency in the vector is captured through the LSTM layer.

[0049] In an optional embodiment, the non-periodic component branch also includes multiple dilated convolutional layers, global pooling layers, and GRU layers. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the multiple dilated convolutional layers. The receptive field is gradually expanded to capture dependencies on different time scales. The output results of the multiple dilated convolutional layers are compressed into vectors of fixed length through the global pooling layer, and the time dependencies in the vectors are captured through the GRU layer.

[0050] Specifically, in the periodic component branch, multiple dilated convolutional layers are used to process the periodic hidden sequence, shear wave velocity, and blood flow imaging depth. The dilation factor of each layer is gradually increased to gradually expand the receptive field and capture the dependencies on different time scales. The dilation factor is set to , then Output of the layer ,in represents the dilated causal convolution operation using the dilation factor, represents the normalization operation, Represents the activation function. In order to compress the feature map after multi-layer dilated convolution into a vector of fixed length, the global pooling layer can be used for average pooling. The final output of the global pooling layer is .in is the global average pooling operation, Compressed into a vector of fixed length, the LSTM layer is then introduced to capture the temporal dependencies in the periodic components. LSTM is a special recurrent neural network that can effectively process long sequence data and avoid the gradient vanishing problem. LSTM controls the flow of information through three gating mechanisms (input gate, forget gate, and output gate) to better capture long-term dependencies. The final expression of its output gate is:

[0051] in, is the weight matrix of the output gate, is the bias term of the output gate, represents the Sigmoid activation function, Represents the output result of LSTM.

[0052] Similarly, the final output of multiple dilated convolutional layers and the global pooling layer in the non-periodic component branch is Then the GRU layer is introduced to capture the time dependency in the non-periodic components. GRU controls the flow of information through two gating mechanisms (reset gate and update gate). It can handle shorter time dependencies while maintaining high efficiency. Under the action of the GRU layer, the feature extraction results are output. .

[0053] The feature extraction results of LSTM Feature extraction results with GRU layer The splicing result is expressed as: + , the concatenated results are input into the fully connected layer, and the rehabilitation effect analysis results are finally output through the fully connected layer.

[0054] The second aspect of the present invention discloses a skeletal muscle state assessment system based on multimodal ultrasound. The assessment system is used to implement any of the assessment methods described above, and the assessment system comprises: A parameter generation module, configured to obtain a first target parameter and a second target parameter from the multimodal ultrasound image; A repetition detection module, configured to generate repetition detection instructions, or, to generate real-time target parameters; The hidden Markov module is configured to obtain a hidden sequence about a real-time target parameter based on a hidden Markov model; The evaluation module is configured to generate a rehabilitation effect analysis result based on the output result of the hidden Markov module and some real-time target parameters.

[0055] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. The protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A skeletal muscle status assessment method based on multimodal ultrasound, which is used to evaluate the therapeutic effect of antagonistic acupuncture, characterized in that: The evaluation method comprises the following steps: The multimodal ultrasound images of the skeletal muscles after treatment are collected twice by an ultrasound detection instrument, wherein the multimodal ultrasound images include ultrasound B-ultrasound images, color ultrasound images, and elastic imaging images; Acquire a first target parameter from the multimodal ultrasound image acquired for the first time, and acquire a second target parameter from the multimodal ultrasound image acquired for the second time; According to the first target parameter and the second target parameter, generating a repeated detection instruction on the ultrasonic detection instrument, or generating a real-time target parameter; A hidden Markov model is used to obtain a hidden sequence of real-time target parameters, and the hidden sequence and part of the real-time target parameters are used as inputs of an evaluation network to generate a rehabilitation effect analysis result.

2. A skeletal muscle state assessment method based on multimodal ultrasound according to claim 1, characterized in that: Generating a repeated detection instruction on the ultrasonic detection instrument according to the first target parameter and the second target parameter, or generating a real-time target parameter, specifically includes: Comparing a first target parameter obtained from a multimodal ultrasound image acquired for the first time with a second target parameter obtained from a multimodal ultrasound image acquired for the second time to generate deviation information; Comparing the deviation information with a deviation threshold, and generating a repeated detection instruction if the deviation information is greater than the deviation threshold; If the deviation information is less than or equal to the deviation threshold, an average value of the first target parameter and the second target parameter is calculated, and real-time detection result information is generated according to the average value.

3. A skeletal muscle state assessment method based on multimodal ultrasound according to claim 2, characterized in that: After the repeated detection instruction is formed, the operator collects the multimodal ultrasound image of the skeletal muscle after treatment for the third time, and obtains the third target parameter therefrom, and uses the average value of the first target parameter, the second target parameter, and the third target parameter as the real-time detection result information.

4. A skeletal muscle status assessment method based on multimodal ultrasound according to claim 3, characterized in that: The first target parameter, the second target parameter and the third target parameter all include muscle thickness, muscle fiber length, feather angle, muscle cross-sectional area, shear wave velocity and microblood flow imaging depth.

5. A skeletal muscle status assessment method based on multimodal ultrasound according to claim 4, characterized in that: The distance between the superficial and deep fascia was extracted from the multimodal ultrasound images representing the skeletal muscle belly as the muscle thickness index data; The linear distance between the cross section of the muscle fiber bundle and the aponeurosis at the intersection of the lateral tibial condyle and the lateral malleolus was extracted from the multimodal ultrasound image of the muscle belly representing the midpoint between the proximal tibia and the lateral malleolus as the muscle fiber length index data; Extract the acute angle between the aponeurosis and the measured pennaceous muscle bundle from the multimodal ultrasound image representing the pennaceous muscle bundle as the pennaceous angle index data; The cross-sectional area of ​​the thickest part of the medial gastrocnemius muscle belly is extracted from the multimodal ultrasound image representing the medial gastrocnemius muscle belly as the cross-sectional area index data; Shear wave velocity of skeletal muscle was obtained by ultrasound elastography; Obtain blood flow imaging depth of skeletal muscle through microblood flow imaging.

6. A skeletal muscle status assessment method based on multimodal ultrasound according to claim 5, characterized in that: The hidden Markov model is used to obtain the hidden sequence of real-time target parameters, including: Set the spasm period, acute remission period, early remission period, mid-term recovery period, long-term stable period, relapse warning period, and deterioration warning period as hidden states, and form a hidden state sequence based on the set hidden states: ( ),in The spasm period, In acute remission, In the early remission period, The mid-term recovery period. For long-term stability, It is the relapse warning period. It is the early warning period for deterioration; Muscle thickness, muscle fiber length, feather angle, and muscle cross-sectional area are set as observation parameters, and the observation parameters are discretized to form the observation parameter expression: ( ), multiple observation parameters form an observation sequence: ( ),in represents the observation parameters at time t, represents the muscle thickness at time t, represents the muscle fiber length at time t, represents the muscle thickness at time t, represents the feather angle at time t; Setting model parameters of the hidden Markov model, wherein the model parameters include initial state probability, state transition probability matrix, and observation probability matrix; Based on the model parameters, the Viterbi algorithm is used to infer the The most likely hidden sequence .

7. A skeletal muscle status assessment method based on multimodal ultrasound according to claim 6, characterized in that: Applying discrete wavelet transform transforms the hidden sequence The sequence is decomposed into a periodic component and a non-periodic component, and an inverse discrete wavelet transform is performed on the periodic component and the non-periodic component respectively to obtain a periodic hidden sequence and a non-periodic hidden sequence.

8. The method for evaluating skeletal muscle status based on multimodal ultrasound according to claim 7, characterized in that: The evaluation network includes a periodic component branch, a non-periodic component branch, and a fully connected layer. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the periodic component branch, and the non-periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the non-periodic component branch. The outputs of the periodic component branch and the non-periodic component branch are spliced ​​and fused and then input into the fully connected layer, and finally the fully connected layer outputs the rehabilitation effect analysis results.

9. The method for evaluating skeletal muscle status based on multimodal ultrasound according to claim 8, characterized in that: The periodic component branch includes a multiple dilated convolution layer, a global pooling layer, and an LSTM layer. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the multiple dilated convolution layer. The receptive field is gradually expanded to capture dependencies on different time scales. The output results of the multiple dilated convolution layer are compressed into a vector of fixed length through the global pooling layer, and the time dependency in the vector is captured through the LSTM layer. The non-periodic component branch also includes multiple dilated convolutional layers, global pooling layers, and GRU layers. The periodic hidden sequence, shear wave velocity, and blood flow imaging depth are used as inputs of the multiple dilated convolutional layers. The receptive field is gradually expanded to capture dependencies on different time scales. The output results of the multiple dilated convolutional layers are compressed into vectors of fixed length through the global pooling layer, and the time dependencies in the vectors are captured through the GRU layer.

10. A skeletal muscle status assessment system based on multimodal ultrasound, characterized in that: The evaluation system is used to implement the evaluation method according to any one of claims 1 to 9, and the evaluation system includes: A parameter generation module, configured to obtain a first target parameter and a second target parameter from the multimodal ultrasound image; A repetition detection module, configured to generate repetition detection instructions, or, to generate real-time target parameters; A hidden Markov module is configured to obtain a hidden sequence about a real-time target parameter based on a hidden Markov model; The evaluation module is configured to generate a rehabilitation effect analysis result based on the output result of the hidden Markov module and some real-time target parameters.

Images