Intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters

Abstract

The invention discloses an intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters. After a controller receives a prediction instruction of an upper computer, a three-axis motion sensor is controlled to be started; n times of three-axis motion data of the left and right little fingers of the wearable suite worn by the person to be predicted in the flexion and extension process within the preset time are obtained and uploaded to the upper computer; the upper computer performs real-time oscillogram display on the nth three-axis motion data of any little finger, analyzes whether the three-axis motion data exceeds a corresponding amplitude range or not, if yes, reminds the nth detection again, and if not, the nth detection data is qualified and stored; and for the qualified three-axis motion data of each little finger each time, constructing a motion feature vector, inputting the motion feature vector into the target machine learning model, and predicting a cervical spondylotic myelopathy prediction result of the person to be predicted.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and intelligent diagnosis technology, and in particular to an intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters. Background Art

[0002] Degenerative Cervical Myelopathy (DCM) is the most common cause of spinal cord dysfunction. This degeneration and damage can cause compression of the spinal cord, which in turn affects the transmission of nerve signals and causes functional impairment. Clumsy finger movements, also known as myelopathy hand, are a common symptom of DCM. Characteristics of myelopathy hand include atrophy of the hand muscles on the affected side, poor dexterity, and reduced strength. The clinical significance of myelopathy hand lies in its ability to serve as an indicator of neurological status and assess the severity of DCM. Common finger movement disorders include flexion and extension disorders of the 2 or 3 fingers on the ulnar side. This means that the patient may not be able to flex or extend the ulnar fingers freely. In addition, the patient may not be able to grip and release (Grip & Release) their hands quickly and flexibly, which affects many of their activities in daily life, such as holding a pen, holding chopsticks, unscrewing bottle caps, etc. The above symptoms are all manifestations of myelopathy hand, which cause certain troubles in the patient's life.

[0003] Previous electromyographic studies of patients with degenerative cervical spondylotic myelopathy (DCM) have shown that the anterior horn of the spinal cord is most severely affected at the C7-T1 level, while the anterior horn of the spinal cord at or near the compressed segment is less severely affected. This may be related to the fact that compression-induced vascular impairment in the cervicothoracic spinal cord is most severe. Animal experiments have also confirmed that in monkey models of medullary compression, the anterior horn of the spinal cord is most severely affected in the cervicothoracic segment, reflecting the characteristic of cervical spinal cord compression, which most commonly affects the cervicothoracic segment. A characteristic manifestation of DCM is myelopathy of the hand, characterized by atrophy of the muscles innervated by C7-C8. The flexor and extensor muscles of the little finger (flexor digitorum superficialis, flexor digitorum profundus, and ulnar muscles of the extensor digitorum communis) are primarily innervated by the anterior horn of the spinal cord at C7 and C8. Therefore, we hypothesized that the kinematic parameters of flexion and extension of the little finger would be more suggestive for the diagnosis of DCM than those of the other fingers.

[0004] The 10-second grip test is a classic test used to evaluate the hand in spinal cord disease. Previous studies have shown that patients with DCM have significantly fewer grips in the grip test compared to healthy, asymptomatic individuals. In an article published by the AO Spine Society using the 10-second grip test to predict DCM, receiver operating characteristic (ROC) curve analysis showed an area under the curve (AUC) of 0.744278, which is clearly not ideal, but suggests a significant correlation between the JOA score and the number of 10-second grips (r = 0.5064, P < 0.0001). To improve the testing method, researchers have made the following attempts: Hosono et al. changed the duration of the gripping experiment from 10 seconds to 15 seconds and used a digital camera to record finger movements, but the number of cases was small and not very convincing; Koyama et al. designed a PC platform system and used Leap Motion software to record the positioning data of 35 finger joint positioning points in the gripping experiment. The system had an AUC of 0.8534, which was a certain improvement over previous methods; Professor Shen Hongxing's team studied the waveforms of each finger in the finger gripping experiment, but in this study, the main focus was on the angular velocity indicators of the subjects. In summary, these studies mainly focused on the angular velocity indicators of finger activities (the number of gripping experiments is essentially a reflection of the angular velocity indicators) and the position data of the fingers, without considering the linear acceleration in the finger activities. In addition, Xie et al. used convolutional neural networks to analyze X-rays of patients with cervical spine injuries for auxiliary diagnosis. Zhang et al. used models based on the T2-weighted images of patients' preoperative cervical spine MRI to predict their postoperative recovery, with good internal cross-validation results. Ahmad et al. used smartphones to shoot videos of patients' finger grips for ten seconds and constructed machine learning models to predict whether their finger mobility would decrease after surgery. The above studies require the use of imaging methods to assist in diagnosis and prognosis, and no models were established to verify the results of external data.

[0005] In recent domestic and international studies, the diagnostic efficiency of only using the number of ten-second grips is not satisfactory, or the diagnostic threshold is raised with the help of imaging equipment. In addition, the mining of patient and kinematic data is not in-depth. In order to further improve the assessment method of spinal cord hand and improve the ability to diagnose and judge the prognosis of spinal cervical spondylotic myelopathy, the present invention constructs a deep learning model for the patient's kinematic data to more comprehensively evaluate the hand function and severity of DCM patients. Summary of the Invention

[0006] In view of the problems and shortcomings of the prior art, the present invention provides an intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters.

[0007] The present invention solves the above technical problems through the following technical solutions:

[0008] The present invention provides an intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters, which is characterized by comprising a wearable kit and a host computer. The wearable kit includes a collar for matching the second joint of the little finger of the person to be predicted, a mounting base fixed to the top of the collar, a housing fixed to the mounting base, and a three-axis motion sensor, a controller, and a wireless transmission module disposed within the housing.

[0009] The controller is used to control the three-axis motion sensor to start after receiving the prediction instruction from the host computer through the wireless transmission module, obtain the three-axis linear acceleration, three-axis angular velocity and three-axis angle of the left and right little fingers of the person to be predicted wearing the wearable kit during the flexion and extension process within a preset time N times, and upload the three-axis linear acceleration, three-axis angular velocity and three-axis angle to the host computer through the wireless transmission module;

[0010] The host computer is used to display a real-time waveform of the three-axis linear acceleration, three-axis angular velocity and three-axis angle of any little finger for the nth time, and analyze whether the three-axis linear acceleration, three-axis angular velocity and three-axis angle exceed the corresponding amplitude range. If so, the nth detection data is discarded and a reminder is given to re-detect the nth time. If not, the nth detection data is qualified and saved, n∈N;

[0011] The host computer is further used to sort the qualified x-axis linear acceleration of each little finger from large to small according to the numerical value, screen out the x-axis linear accelerations in the first K positions, and calculate the average x-axis linear acceleration of each little finger and the maximum x-axis linear acceleration, and then calculate the average of the x-axis linear accelerations corresponding to the 2N times of the left and right little fingers as the final x-axis acceleration average value. And the average value of the maximum x-axis acceleration corresponding to 2N times Similarly, we can get the x-axis y-axis z-axis Constructing motion feature vector And input it into the target machine learning model to predict the prediction result of cervical spondylotic myelopathy for the person to be predicted.

[0012] The positive progress effect of the present invention is:

[0013] By introducing multiple kinematic data as a key indicator, the present invention innovatively designs a finger motion assessment tool that integrates three-axis linear acceleration, three-axis angular velocity and three-axis angle measurement, in order to achieve breakthrough progress in the early diagnosis, disease assessment and postoperative recovery prediction of cervical spondylotic myelopathy. In addition, the present invention can be used as a screening tool for cervical spondylotic myelopathy and can be performed before large-scale examinations such as magnetic resonance imaging, thereby reducing medical costs and avoiding the use of magnetic resonance imaging as a screening tool for cervical spondylotic myelopathy, thereby greatly reducing medical costs.

[0014] This system is a highly sensitive and specific finger motion testing system that accurately measures and records kinematic data such as linear acceleration and angular velocity during specific tasks. The system enables real-time measurement, data analysis, and visualization of finger motion, ultimately providing model diagnostic results. This system is used to quantitatively assess the finger mobility of patients with cervical spondylotic myelopathy and healthy controls, exploring differences between the two groups and their correlation with disease severity. Furthermore, it will analyze differences in finger kinematic parameters across age groups and genders, providing a basis for personalized diagnosis and treatment.

[0015] The present invention uses the patient's little finger kinematic data and the patient's basic information to construct a machine learning model, which can accurately reflect the patient's symptoms related to cervical spondylotic myelopathy, thereby accurately assessing whether the patient has cervical spondylotic myelopathy and the severity of the cervical spondylotic myelopathy.

[0016] The present invention adopts the gold panning optimization algorithm to optimize the hyperparameters of the constructed machine learning model, which can obtain an accurate optimized machine learning model. By using this accurate optimized machine learning model to predict patients, accurate prediction can be achieved and accurate prediction results can be obtained.

[0017] The present invention utilizes a gold panning optimization algorithm to optimize hyperparameters of a machine learning model. During the optimization process, in order to improve optimization efficiency, the number of gold panning groups and the initialization of gold panning individuals are not randomly set within a parameter setting range as in the prior art. Instead, the number of gold panning groups and the initial positions of each gold panning individual are initialized based on the fitness of the initial model training. When the fitness is high, it indicates that the fitness is close to the ideal fitness. In this case, a smaller number of gold panning groups is set. When the fitness is low, it indicates that the fitness is far from the ideal fitness. In this case, a larger number of gold panning groups is set. A gold panning individual in the group is initialized with the hyperparameters after the initial model training, and other gold panning individuals in the group are initialized with the variations of the hyperparameters after the initial model training. This initialization method can achieve rapid optimization and quickly iterate the optimal hyperparameters. To facilitate convergence, the present invention adds adaptive weights to the migration strategy, mining strategy, and cooperation strategy. To prevent the gold panning individuals from falling into the local optimum, disturbances are added. To prevent the gold panning group from falling into the local optimum, a combination strategy is set and disturbances are added. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 Schematic diagram of the structure of a wearable kit according to a preferred embodiment of the present invention.

[0019] Figure 2 This is a real-time angle line graph of the x-axis in the host computer interface of a preferred embodiment of the present invention.

[0020] Figure 3 This is a three-axis linear acceleration diagram of historical patients with mild and severe cervical spondylotic myelopathy according to a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] This embodiment provides an intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters, comprising a wearable kit and a host computer, which exchange information with each other. The host computer pre-stores a correspondence between gender and age group and a set triaxial acceleration amplitude range, a set triaxial angular velocity amplitude range, and a set triaxial angle amplitude range.

[0023] Among them, such as Figure 1As shown, the wearable kit includes a ring 1 that fits on the second joint of the pinky finger of the person being tested. A mounting base 2 is fixed to the top of the ring 1, and a housing 3 is fixed to the mounting base 2. The housing 3 houses a power supply, a three-axis motion sensor, a controller, and a wireless transmission module. The housing 3 is provided with a power port 31. The three-axis motion sensor uses a three-axis gyroscope or a three-axis accelerometer.

[0024] When a prediction for cervical spondylotic myelopathy is needed for a patient, a host computer issues a prediction command, which is received by a controller via a wireless transmission module. This controller activates the three-axis motion sensor and acquires the linear acceleration, angular velocity, and angle of the left and right pinky fingers of the patient wearing the wearable kit during flexion and extension, respectively, N times (e.g., twice) within a preset time (e.g., 20 seconds). These values ​​are then uploaded to the host computer via the wireless transmission module. The system records changes in linear acceleration and angular velocity of the fingers within the flexion and extension plane, as well as changes in range of motion, in real time. This real-time performance is crucial for capturing the dynamic characteristics of finger motion and helps more accurately assess finger mobility.

[0025] For example, the person to be predicted can first wear the wearable kit on the left little finger to obtain the three-axis linear acceleration, three-axis angular velocity and three-axis angle of the left little finger during flexion and extension twice within 20 seconds, and then wear the wearable kit on the right little finger to obtain the three-axis linear acceleration, three-axis angular velocity and three-axis angle of the right little finger during flexion and extension twice within 20 seconds.

[0026] The host computer is used to obtain the age and gender of the person to be predicted, and to display the three-axis linear acceleration, three-axis angular velocity and three-axis angle of the nth (n∈N) of any little finger (left little finger or right little finger) in real time as a waveform, and to analyze whether the three-axis linear acceleration, three-axis angular velocity and three-axis angle exceed the amplitude range corresponding to the gender and age group of the person to be predicted. If so, the nth test data is discarded and a reminder is given to retest the nth time. If not, the nth test data is qualified and saved.

[0027] For example, if analysis shows that the three-axis linear acceleration, three-axis angular velocity, and three-axis angle of the left little finger during the flexion and extension process within the first 20 seconds exceed the amplitude range corresponding to the gender and age group of the person to be predicted, the first detection data of the left little finger will be discarded, and a reminder will be given to repeat the first detection of the left little finger.

[0028] The displayed waveform diagrams can help researchers and clinicians understand the characteristics and changes of finger movements more intuitively. These visualization tools play an important role in interpreting the severity of cervical spondylosis in patients, formulating treatment plans, and evaluating postoperative rehabilitation effects.

[0029] The host computer is also used to sort the qualified x-axis linear acceleration of each little finger from large to small according to the numerical value, screen out the x-axis linear acceleration in the first K digits (such as the first ten digits) and calculate the average x-axis linear acceleration of each little finger each time, and the maximum x-axis linear acceleration, and then calculate the average of the x-axis linear acceleration corresponding to the 2N times of the left and right little fingers as the final x-axis acceleration average value And the average value of the maximum x-axis acceleration corresponding to 2N times

[0030] Similarly, the final average x-axis angular velocity is obtained Maximum x-axis angular velocity average Final x-axis angle average Final average y-axis acceleration Maximum y-axis acceleration average Final average y-axis angular velocity Maximum y-axis angular velocity average Final y-axis angle average Final z-axis acceleration average Maximum z-axis acceleration average Final average z-axis angular velocity Maximum z-axis angular velocity average Final z-axis angle average

[0031] In this embodiment, the average values ​​of the first 10 angular velocities and the average values ​​of linear accelerations can be automatically calculated. These parameters reflect the movement stability of the fingers in the gripping experiment. The maximum angular velocity and maximum linear acceleration can also be calculated. These parameters reveal the maximum explosive force and speed of the fingers during the movement.

[0032] The upper computer is also used to encode the gender and age of the predictor using one-hot encoding or entity embedding to obtain gender encoding and age encoding.

[0033] The host computer is also used to construct the motion feature vector of the person to be predicted (gender code, age code, ) and input it into the target machine learning model to predict the cervical spondylotic myelopathy outcome for the subject. The cervical spondylotic myelopathy diagnosis outcomes include: no cervical spondylotic myelopathy, mild cervical spondylotic myelopathy, plateau cervical spondylotic myelopathy, severe cervical spondylotic myelopathy, and paralysis caused by cervical spondylotic myelopathy.

[0034] In order to obtain the target machine learning model, the host computer constructs a machine learning model, uses historical samples to train the machine learning model, uses the gold panning optimization algorithm to optimize the model hyperparameters, and obtains the optimal hyperparameters through continuous optimization. The machine learning model established with the optimal hyperparameters is the target machine learning model.

[0035] Among them, the gold panning optimization algorithm: improves the group initialization of the traditional gold panning algorithm, adds adaptive weights to the migration strategy, mining strategy and cooperation strategy in the traditional gold panning algorithm to form an adaptive migration strategy, an adaptive mining strategy and an adaptive cooperation strategy, adds disturbance improvements when individuals fall into local optimality, and adds combination strategy improvements when groups fall into local optimality to obtain the gold panning optimization algorithm.

[0036] Among them, the improvement of group initialization includes: first using historical samples to conduct preliminary training on the constructed machine learning model to obtain the fitness f, analyzing the fitness f and the preset fitness f1, when f≥f1, it indicates that the fitness f is relatively close to the ideal fitness, at this time the group size can be set to be relatively small, recorded as M1, when f<f1, it indicates that the fitness f is relatively not close to the ideal fitness, at this time the group size can be set to be relatively large, recorded as M2, M1<M2, and then use the current hyperparameters after the preliminary training of the model and the hyperparameters obtained by mutating the current hyperparameters to initialize the group position in the gold mining optimization algorithm.

[0037] Add adaptive weights to the migration strategy in the traditional gold mining algorithm to form an adaptive migration strategy:

[0038]

[0039]

[0040] Where, and Represent the new position and current position of individual i respectively, A1 represents the variable coefficient, represents the migration vector, W(t) represents the adaptive weight, W min Indicates the minimum weight, W max Indicates the maximum weight.

[0041] Add adaptive weights to the mining strategy in the traditional gold mining algorithm to form an adaptive mining strategy:

[0042]

[0043] Where, represents the current position of the randomly selected individual g, A2 represents the variable coefficient, Represents the mining vector.

[0044] Add adaptive weights to the cooperation strategy in the traditional gold mining algorithm to form an adaptive cooperation strategy:

[0045]

[0046] In the formula, r1 represents a random number between 0 and 1, represents the cooperation vector.

[0047] Add combination strategy improvements when the group falls into local optimality:

[0048] Adaptive migration mining combination strategy:

[0049]

[0050] Adaptive mining cooperative combination strategy:

[0051]

[0052] Where r2 and r3 are both random numbers between 0 and 1.

[0053] When individuals fall into local optimality, disturbance improvement is added, and the disturbance formula ξ*r4 is added to the specific strategy:

[0054] For example, when the adaptive migration strategy is used to update the group position:

[0055] r4 represents a random number between 0 and 1.

[0056] When the adaptive mining strategy is used to update the group position:

[0057]

[0058] When the adaptive cooperation strategy is used to update the group position:

[0059]

[0060] When the adaptive migration mining combination strategy is used to update the group position:

[0061]

[0062] When the adaptive mining cooperative combination strategy is used to update the group position:

[0063]

[0064] Initialize the group position in the gold rush optimization algorithm using the current hyperparameters of the model after preliminary training and the hyperparameters obtained by mutating the current hyperparameters. Use the gold rush optimization algorithm to optimize the hyperparameters, obtain the optimal hyperparameters through continuous optimization, and use the optimal hyperparameters to establish the target machine learning model. Specifically, the following steps are involved:

[0065] (1) The number of populations M = M1 or M2 is set according to the fitness f. Each gold prospecting individual in the population represents a set of hyperparameters. After initial training, the model hyperparameter is C. C is used to initialize the position of a gold prospecting individual in the population. C is then mutated within the preset hyperparameter value range to initialize the positions of the remaining M-1 gold prospecting individuals in the population.

[0066] (2) Substitute the hyperparameters corresponding to each individual into the machine learning model and train them using historical samples. Calculate the fitness f of each individual. The fitness f of each individual is taken as the individual's optimal fitness, and the maximum value is taken as the group's optimal fitness. The current position of the individual corresponding to the group's optimal fitness is taken as the optimal gold mine position.

[0067] (3) When 0.5≤r<1, the adaptive migration strategy or adaptive mining strategy is used to update the group position. When the individual best fitness of an individual is not updated for t1 consecutive iterations (such as 3 times), the individual falls into the local optimum, and the position of the individual is updated by the adaptive migration strategy or adaptive mining strategy with added disturbance. When the group best fitness is not updated for t2 consecutive iterations (such as 2 times), the adaptive migration and mining combination strategy is used to update the position. After the adaptive migration and mining combination strategy is used to update the position, if the group best fitness is not updated for t2 consecutive iterations, the adaptive migration and mining combination strategy with added disturbance is used to update the position.

[0068] When r is less than 0.5, the adaptive mining strategy or adaptive cooperation strategy is used to update the group position. When the individual best fitness of an individual is not updated for t1 consecutive iterations, the individual falls into a local optimum, and the position of the individual is updated by the adaptive mining strategy or adaptive cooperation strategy with added disturbance. When the group best fitness is not updated for t2 consecutive iterations, the adaptive mining cooperation combination strategy is used to update the position currently and thereafter. After the adaptive mining cooperation combination strategy is used to update the position, if the group best fitness is not updated for t2 consecutive iterations, the adaptive mining cooperation combination strategy with added disturbance is used to update the position.

[0069] Iterative correlation coefficient r=(t max -t) / t max , t max and t are the maximum and current iteration times respectively;

[0070] Calculate the corresponding fitness f according to the position to be updated of each individual, compare the current fitness f of each individual with its individual best fitness, when the current fitness f of an individual is greater than its individual best fitness, update the individual best fitness of the individual = the current fitness f of the individual, and update the individual position, otherwise do not update; if the maximum value of the current individual best fitness of each individual is greater than the group best fitness, then update the group best fitness, and the current position of the individual corresponding to the current group best fitness is used as the best gold mine position, otherwise do not update;

[0071] (4) Determine whether the optimal fitness of the group reaches the preset fitness (i.e., ideal fitness) f2 or the number of iterations reaches the maximum number of iterations t max If yes, the optimal gold mine location is used as the optimal hyperparameter to construct the target machine learning model, f2>f1, otherwise enter (3) to start the next iteration.

[0072] In this embodiment, a gold panning optimization algorithm is used to optimize the hyperparameters of a machine learning model. During the optimization process, in order to improve the optimization efficiency, the initialization of the number of gold panning groups and the initialization of gold panning individuals are not randomly set as in the prior art, but the number of gold panning groups and the initial position of each gold panning individual are initialized based on the fitness of the initial training of the model; in order to facilitate convergence, the present invention adds adaptive weights to the migration strategy, mining strategy and cooperation strategy; in order to prevent the gold panning individuals from falling into the local optimum, disturbances are added; in order to prevent the gold panning group from falling into the local optimum, a combination strategy is set and disturbances are added.

[0073] See Figure 2 and Figure 3 , construction of historical samples:

[0074] The host computer is also used to receive the three-axis linear acceleration, three-axis angular velocity and three-axis angle of each patient's left and right little fingers during flexion and extension for N times within a preset time from the controller.

[0075] The host computer is also used to display the three-axis linear acceleration, three-axis angular velocity and three-axis angle of any little finger of any historical patient in real time as a waveform, and analyze whether the three-axis linear acceleration, three-axis angular velocity and three-axis angle exceed the corresponding amplitude range. If so, the n-th test data is discarded and a reminder is given to re-test the nth time. If not, the n-th test data is qualified and saved.

[0076] The host computer is also used to sort the qualified x-axis linear acceleration of each little finger of any historical patient from large to small according to the numerical value, screen out the x-axis linear acceleration in the first K positions and calculate the average x-axis linear acceleration of each little finger each time and the maximum x-axis linear acceleration, and then calculate the average of the x-axis linear acceleration corresponding to the 2N times of the left and right little fingers as the final x-axis acceleration average value And the average value of the maximum x-axis acceleration corresponding to 2N times Similarly, we can get the x-axis y-axis z-axis The motion feature vectors of each historical patient are constructed and the corresponding cervical spondylotic myelopathy diagnosis results are annotated to form the historical sample. The cervical spondylotic myelopathy diagnosis results include: no cervical spondylotic myelopathy, mild cervical spondylotic myelopathy, plateau cervical spondylotic myelopathy, severe cervical spondylotic myelopathy, and paralysis caused by cervical spondylotic myelopathy.

[0077] Numerical results from finger motion sensors for cervical spondylotic myelopathy can serve as a quantitative diagnostic tool for cervical spondylotic myelopathy. Previous studies have used counting the number of finger flexions and extensions within 10 seconds, but this approach is subject to significant randomness and very low reliability, resulting in a low diagnostic value for 10-second grip counts for cervical spondylotic myelopathy. Finger motion sensors accurately record three-axis linear acceleration, three-axis angular velocity, and three-axis angle during finger flexion and extension. By recording the extreme values ​​of acceleration and angular velocity at the mid-range and end of the grip, the sensor reliably reflects the maximum value of finger activity. Our previous research has confirmed the diagnostic cutoff values ​​for acceleration, area under the curve, sensitivity, and specificity for cervical spondylotic myelopathy in patients of different age groups. We found that the mean of the top 10 accelerations has the highest diagnostic value for cervical spondylotic myelopathy, while the mean of the top 10 angular velocities and the mean of the top 10 angles also have high diagnostic value for cervical spondylotic myelopathy. The team also developed a diagnostic model for cervical spondylotic myelopathy using artificial intelligence machine learning models such as random forests and decision trees. This model not only distinguishes patients with cervical spondylotic myelopathy from healthy individuals, but also differentiates between mild and severe cervical spondylotic myelopathy, assisting in the grading of cervical spondylosis. The average of the top 10 triaxial accelerations, the top 10 triaxial angular velocities, and the top 10 triaxial angles can serve as numerical diagnostic criteria for cervical spondylotic myelopathy.

[0078] Widely used in clinical practice to improve diagnostic efficiency: By comprehensively analyzing kinematic parameters such as linear acceleration and angular velocity during finger movements, combined with the patient's demographic data (gender and age), a more accurate and efficient diagnostic model is constructed. This model can not only predict the risk of degenerative cervical spondylotic myelopathy at an early stage, but also accurately assess the severity of the disease, providing strong support for clinical decision-making. Its broad application prospects are not limited to spinal surgery, but can also be expanded to multiple fields such as neurology and rehabilitation medicine, providing accurate diagnosis and treatment services for patients in different disciplines.

[0079] In summary, this system is an innovative and practical assessment tool. Its high precision, real-time performance, and comprehensive data analysis capabilities provide strong support for the assessment and research of hand function.

[0080] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these are merely illustrative and that the scope of the present invention is defined by the appended claims. Those skilled in the art may make various changes or modifications to these embodiments without departing from the principles and essence of the present invention, and such changes and modifications are intended to fall within the scope of the present invention.

Claims

1. An intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters, characterized by: The wearable kit includes a wearable kit and a host computer. The wearable kit includes a ring for matching the second knuckle of the little finger of the person to be predicted. The top of the ring is fixed with a mounting base, and a housing is fixed on the mounting base. The housing contains a three-axis motion sensor, a controller, and a wireless transmission module. The controller is used to control the three-axis motion sensor to start after receiving the prediction instruction from the host computer through the wireless transmission module, obtain the three-axis linear acceleration, three-axis angular velocity and three-axis angle of the left and right little fingers of the person to be predicted wearing the wearable kit during the flexion and extension process within a preset time N times, and upload the three-axis linear acceleration, three-axis angular velocity and three-axis angle to the host computer through the wireless transmission module; The host computer is used to display a real-time waveform of the three-axis linear acceleration, three-axis angular velocity and three-axis angle of any little finger for the nth time, and analyze whether the three-axis linear acceleration, three-axis angular velocity and three-axis angle exceed the corresponding amplitude range. If so, the nth detection data is discarded and a reminder is given to re-detect the nth time. If not, the nth detection data is qualified and saved, n∈N; The host computer is further used to sort the qualified x-axis linear acceleration of each little finger from large to small according to the numerical value, screen out the x-axis linear accelerations in the first K positions, and calculate the average x-axis linear acceleration of each little finger and the maximum x-axis linear acceleration, and then calculate the average of the x-axis linear accelerations corresponding to the 2N times of the left and right little fingers as the final x-axis acceleration average value. And the average value of the maximum x-axis acceleration corresponding to 2N times Similarly, we can get the x-axis y-axis z-axis Construct motion feature vector ( ) and input it into the target machine learning model to predict the prediction result of cervical spondylotic myelopathy for the person to be predicted.

2. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 1, characterized in that: The host computer pre-stores the correspondence between gender and age group and the set three-axis acceleration amplitude range, the set three-axis angular velocity amplitude range and the set three-axis angle amplitude range; The host computer is also used to obtain the age and gender of the person to be predicted, and to display a real-time waveform of the three-axis linear acceleration, three-axis angular velocity and three-axis angle of any little finger for the nth time, and to analyze whether the three-axis linear acceleration, three-axis angular velocity and three-axis angle exceed the amplitude range corresponding to the gender and age group of the person to be predicted. If so, the nth test data is discarded and a reminder is given to re-test the nth time. If not, the nth test data is qualified and saved.

3. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 2, characterized in that: The host computer is also used to encode the gender and age of the person to be predicted using one-hot encoding or entity embedding to obtain the gender code and age code, and to construct the motion feature vector of the person to be predicted (gender code, age code, ) and input it into the target machine learning model to predict the prediction result of cervical spondylotic myelopathy for the person to be predicted.

4. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 1, characterized in that: The host computer is also used to improve the group initialization of the traditional gold panning algorithm, add adaptive weights to the migration strategy, mining strategy and cooperation strategy in the traditional gold panning algorithm to form an adaptive migration strategy, an adaptive mining strategy and an adaptive cooperation strategy, add disturbance improvement when an individual falls into a local optimum, and add combination strategy improvement when a group falls into a local optimum, so as to obtain a gold panning optimization algorithm; The host computer is also used to build a machine learning model, train the machine learning model using historical samples, optimize the model hyperparameters using the gold panning optimization algorithm, obtain the optimal hyperparameters through continuous optimization, and the machine learning model established using the optimal hyperparameters is the target machine learning model.

5. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters as claimed in claim 4 is characterized in that: Initialization improvements include: The host computer is also used to first use historical samples to perform preliminary training on the machine learning model to obtain fitness f, analyze the fitness f and the preset fitness f1, set the population size to M1 when f≥f1, set the population size to M2 when f<f1, M1<M2, and then use the current hyperparameters of the model and the hyperparameters obtained by randomly mutating the current hyperparameters to initialize the population position in the gold panning optimization algorithm, use the gold panning optimization algorithm to optimize the hyperparameters, obtain the optimal hyperparameters through continuous optimization, and use the optimal hyperparameters to establish the target machine learning model.

6. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 5, characterized in that: The host computer is also used to initialize the group position in the gold panning optimization algorithm using the current hyperparameters of the model and the hyperparameters obtained by randomly mutating the current hyperparameters, optimize the hyperparameters using the gold panning optimization algorithm, obtain the optimal hyperparameters through continuous optimization, and establish the target machine learning model using the optimal hyperparameters: (1) The number of populations M = M1 or M2 is set according to the fitness f. Each gold prospecting individual in the population represents a set of hyperparameters. After initial training, the model hyperparameter is C. C is used to initialize the position of a gold prospecting individual in the population. C is then mutated within the preset hyperparameter value range to initialize the positions of the remaining M-1 gold prospecting individuals in the population. (2) Substitute the hyperparameters corresponding to each individual into the machine learning model and train them using historical samples. Calculate the fitness f of each individual. The fitness f of each individual is taken as the individual's optimal fitness, and the maximum value is taken as the group's optimal fitness. The current position of the individual corresponding to the group's optimal fitness is taken as the optimal gold mine position. (3) When 0.5≤r<1, the adaptive migration strategy or adaptive mining strategy is used to update the group position. When r<0.5, the adaptive mining strategy or adaptive cooperation strategy is used to update the group position. The iterative correlation coefficient r=(t max -t) / t max , t max and t are the maximum and current iteration times respectively; Calculate the corresponding fitness f according to the position to be updated of each individual, compare the current fitness f of each individual with its individual best fitness, when the current fitness f of an individual is greater than its individual best fitness, update the individual best fitness of the individual = the current fitness f of the individual, and update the individual position, otherwise do not update; if the maximum value of the current individual best fitness of each individual is greater than the group best fitness, then update the group best fitness, and the current position of the individual corresponding to the current group best fitness is used as the best gold mine position, otherwise do not update; (4) Determine whether the optimal fitness of the group reaches the preset fitness f2 or the number of iterations reaches the maximum number of iterations t max If yes, the optimal gold mine location is used as the optimal hyperparameter to construct the target machine learning model, f2>f1, otherwise enter (3) to start the next iteration.

7. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 6, characterized in that: In (3), when 0.5≤r<1, the adaptive migration strategy or adaptive mining strategy is used to update the group position. When the individual best fitness of an individual is not updated for t1 consecutive iterations, the individual falls into the local optimum, and the position of the individual is updated by the adaptive migration strategy or adaptive mining strategy with added disturbance. When the group best fitness is not updated for t2 consecutive iterations, the adaptive migration and mining combination strategy is used to update the position. After the adaptive migration and mining combination strategy is used to update the position, if the group best fitness is not updated for t2 consecutive iterations, the adaptive migration and mining combination strategy with added disturbance is used to update the position. When r is less than 0.5, the adaptive mining strategy or the adaptive cooperation strategy is used to update the group position. When the individual best fitness of an individual is not updated for t1 consecutive iterations, the individual falls into the local optimum, and the adaptive mining strategy or the adaptive cooperation strategy with added disturbance is used to update the position of the individual. When the best fitness of the group is not updated for t2 consecutive iterations, the adaptive mining cooperation combination strategy is used to update the position currently and thereafter. After the adaptive mining cooperation combination strategy is used to update the position, if the best fitness of the group is not updated for t2 consecutive iterations, the adaptive mining cooperation combination strategy with added disturbance is used to update the position.

8. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 4, characterized in that: Construction of historical samples: The host computer is also used to receive the three-axis linear acceleration, three-axis angular velocity and three-axis angle of each patient's left and right little fingers during flexion and extension for N times within a preset time from the controller; The host computer is also used to display a real-time waveform of the three-axis linear acceleration, three-axis angular velocity and three-axis angle of any little finger of any historical patient for the nth time, and analyze whether the three-axis linear acceleration, three-axis angular velocity and three-axis angle exceed the corresponding amplitude range. If so, the nth test data is discarded and a reminder is given to retest the nth time. If not, the nth test data is qualified and saved; The host computer is further used to sort the qualified x-axis linear acceleration of each little finger of any historical patient from large to small by numerical value, screen out the x-axis linear accelerations in the top K positions, calculate the average x-axis linear acceleration of each little finger each time, and the maximum x-axis linear acceleration, and then calculate the average of the x-axis linear accelerations corresponding to the 2N times of the left and right little fingers as the final x-axis acceleration average value. And the average value of the maximum x-axis acceleration corresponding to 2N times Similarly, we can get the x-axis y-axis z-axis The motion feature vectors of each historical patient are constructed and the corresponding cervical spondylotic myelopathy diagnosis results are annotated to form a historical sample.

9. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 1, characterized in that: The three-axis motion sensor adopts a three-axis gyroscope or a three-axis accelerometer.

10. The intelligent diagnosis and prediction system for cervical spondylotic myelopathy based on finger motion parameters according to claim 1, characterized in that: The diagnosis or prediction results of cervical spondylotic myelopathy include: no cervical spondylotic myelopathy, mild cervical spondylotic myelopathy, plateau stage cervical spondylotic myelopathy, severe cervical spondylotic myelopathy, and paralysis caused by cervical spondylotic myelopathy.

Images