CN111540465A

CN111540465A - Method for predicting movement injury risk of college student male football players by neural network model

Info

Publication number: CN111540465A
Application number: CN202010301826.7A
Authority: CN
Inventors: 高晓嶙; 张津沁; 杨慧君; 梁红红; 李闯涛
Original assignee: CHINA INSTITUTE OF SPORT SCIENCE
Current assignee: CHINA INSTITUTE OF SPORT SCIENCE
Priority date: 2020-04-16
Filing date: 2020-04-16
Publication date: 2020-08-14

Abstract

The invention discloses a method for predicting the sports injury risk of college student male football players by a neural network model, which comprises the following steps: s1, collecting basic information data and functional action evaluation index data of the subject; s2, carrying out lower limb non-contact injury condition investigation on the subjects, and dividing the subjects into an injury group and a non-injury group; s3, respectively collecting P values of the data between the damaged group and the non-damaged group, and using the P values as a standard for judging whether various data are screened as damage risk factors; s4, constructing a multilayer perceptron model by using the damage risk factor as an independent variable and using whether non-contact damage exists or not as a dependent variable; and S5, predicting the risk of non-contact injury of the college student male football player by using the multilayer perceptron model. The method has the advantages that the risk of non-contact injury of the lower limbs of the football players of the college students in China is effectively predicted, the diagnosis accuracy of the non-injury crowd is 87%, and the diagnosis accuracy of the injury crowd is 93.3%.

Description

Method for predicting movement injury risk of college student male football players by neural network model

Technical Field

The invention relates to the technical field of damage risk prediction. More particularly, the invention relates to a method for predicting the sports injury risk of a college student male football player by using a neural network model.

Background

Football is the first sports in the world, and in our country there are many people who love the sports, especially enjoyed by many students, and football teams are formed in many colleges and universities. The campus football has important strategic significance on the national level of China, the height and the strength of the campus football are not available, and the improvement of the competitive level of the Chinese football and the improvement of the popularization degree of the football are in a bearing with each other. The football is developed vigorously while hiding the injury risk, and actions such as sprinting, sudden stop and sudden turn, jump, speed change and the like in training and competition easily cause bruise, contusion and strain, even some more serious injuries, wherein the lower limb injury rate of football players is high, and a non-contact injury mechanism is a common injury mechanism (35%). How to scientifically monitor the non-contact injury risk of the male football players of college students in China is a problem which needs to be solved urgently.

Currently, some injury risk factors in soccer have been studied individually, but injury may be multifactorial, and the evaluation of an isolated injury risk factor does not take into account the functional movement pattern required by the athlete to perform the sport, and the interaction of various risk factors.

An Artificial Neural Network (ANN) is a mathematical model which is based on the basic principle of neural networks in biology, simulates the processing mechanism of a neural system of a human brain to complex information by taking network topology knowledge as a theoretical basis after the structure of the human brain and an external stimulation response mechanism are understood and abstracted, has high nonlinearity, and can carry out complex logic operation and nonlinear relationship realization. The ANN has good parallel, processing capacity, distributed storage, high fault tolerance, nonlinear approximation, intellectualization, self-learning and other capabilities, is widely applied to complex tasks of hardware fault detection, medical diagnosis, medical image processing and the like of complex systems at present to classify and predict, and has good effect, but the evaluation of the risk of motion damage is rarely reported.

Therefore, how to screen and determine the lower limb non-contact injury risk factor and provide a scientific method for evaluating the lower limb non-contact injury risk of the college football players according to the artificial neural network is a problem which is urgently needed to be solved at present.

Disclosure of Invention

An object of the present invention is to solve at least the above problems and to provide at least the advantages described later.

The invention also aims to provide a method for predicting the sports injury risk of male football players of college students by using a neural network model, which can screen and determine the non-contact injury risk factors of the lower limbs and evaluate the non-contact injury risk of the lower limbs of the male football players of the college students according to the artificial neural network.

To achieve these objects and other advantages in accordance with the present invention, there is provided a method for predicting a college student male soccer player sports injury risk by a neural network model, comprising the steps of:

s1, collecting basic information data and functional action evaluation index data of the subject;

s2, carrying out lower limb non-contact injury condition investigation on the subjects in a preset time period, and classifying the subjects into an injury group and a non-injury group according to whether non-contact injury occurs in the preset time period;

s3, respectively calculating P values of the basic information data and the functional action evaluation index data between the damaged group and the non-damaged group, and taking the P values as a standard for judging whether various basic information and various functional action evaluation indexes are screened as damage risk factors;

s4, constructing a multilayer perceptron model by using the damage risk factor as an independent variable and using whether non-contact damage exists as a dependent variable, training, and stabilizing the training to the multilayer perceptron model;

s5, collecting injury risk factor data of the college student male football players, and predicting the non-contact injury risk of the college student male football players by using the multilayer sensor model.

Preferably, the subject in step S1 is a soccer player for college students in china who has not suffered from injury in the first half year and has not received rehabilitation therapy.

Preferably, the basic information data in step S1 includes at least height data, weight data, whether the data is for the first dealer, and non-invasive history data.

Preferably, the step S1 of collecting the functional action evaluation index data specifically includes:

the subject adopts the American national institute of sports medicine standard to carry out functional action tests at least including lower limb Y balance test and landing error scoring system test, and obtains corresponding evaluation index data, wherein:

when a subject performs a lower limb Y balance test, obtaining evaluation index data at least comprising posterior medial difference data and comprehensive value difference data;

when the subject performs the floor fault scoring system test, the obtained evaluation index data at least comprises evaluation index data of the total LESS score.

Preferably, the predetermined period of time in step S2 is 1 year;

the lower limb non-contact injury in step S2 is defined as an injury in which any part of the lower limb satisfies the following condition: caused by mechanisms other than direct impact, requiring medical intervention and resulting in at least one day's failure to participate in sports-related activities.

Preferably, in step S3, the P values of the basic information data and the functional action evaluation index data between the damaged group and the non-damaged group are calculated, specifically:

when the basic information data or the functional action evaluation index data are numerical data, determining whether the data conform to normal distribution by adopting K-S (K-S) test, and if not, determining the P value of the numerical data between a damaged group and a non-damaged group by adopting non-parameter test;

and when the basic information data or the functional action evaluation index data are non-numerical data, determining the P value of the non-numerical data between the damaged group and the non-damaged group by adopting chi-square test.

Preferably, the criterion for determining whether each type of basic information and each type of functional action evaluation index is screened as the injury risk factor in step S3 with the P value is specifically:

taking P <0.10 as a standard for judging various types of basic information to be screened as damage risk factors;

taking P <0.05 as a standard for judging various functional action evaluation indexes to be screened as damage risk factors;

wherein the injury risk factors comprise height, weight, whether the first team member is present, no injury history, posterior medial difference, comprehensive value difference and LESS total score.

Preferably, the step S4 of constructing the multi-layer perceptron model specifically includes:

randomly grouping the subjects, forming a training group by 70% of the cases, forming a test group by 20% of the cases, and forming a verification group by 10% of the cases;

the training set is used for training a neural network to obtain a multilayer perceptron model, wherein the multilayer perceptron model comprises an input layer, a hidden layer and an output layer, the number of neurons of the input layer is determined according to independent variables, the hidden layer comprises 3 neurons, the output layer comprises 2 neurons with or without non-contact damage, an activation function of the hidden layer is a hyperbolic tangent function, an activation function of the output layer is a Softmax function, and an error function is a cross entropy function;

the test set samples are used to track errors in the training process to prevent over-training;

and the verification group is used for evaluating and testing the constructed multilayer perception model.

Preferably, the determining of the neurons of the input layer according to the independent variables is specifically:

the body height and the body weight are respectively used as covariates to form 2 neurons after being standardized, wherein the standardized processing specifically comprises the following steps: the mean was subtracted from the covariates and divided by the standard deviation;

whether the initial member forms 3 neurons including the initial member, the non-initial member and the waiting member;

2 neurons with damage history and no damage history are formed according to the damage history;

and when the damage risk factors are the rear inner side difference value, the comprehensive value difference value and the LES total time, determining the optimal diagnosis points corresponding to the damage risk factors by adopting an ROC curve method according to the damage risk factor data, and respectively forming 6 neurons by using the data of the corresponding damage risk factors, namely the data of the corresponding damage risk factors, which are not LESS than the optimal diagnosis points, and the data of the corresponding damage risk factors, which are LESS than the optimal diagnosis points.

Preferably, the optimum probability diagnosis point for diagnosing the non-contact injury is determined to be 0.86439 by using an ROC curve method according to the data of the output layer.

The invention at least comprises the following beneficial effects:

the method comprises the steps of carrying out functional action test on football athletes (testees) of college students in China, collecting functional action evaluation index data of the testees, forming a basic information + functional action index data source by combining the collected basic information data, further combining damage investigation, screening out reliable, effective and representative damage risk factors (height, weight, whether the team member is first issued, whether damage history exists, rear inner side difference, comprehensive value difference and LESS total score), and developing a risk prediction model (a multilayer perceptron model) based on an artificial neural network algorithm by using the damage risk factors as independent variables and whether non-contact damage exists as dependent variables. The method has important significance for controlling and reducing the damage risk of the football sports, popularizing the football sports better and guiding the rehabilitation and training of football players, and simultaneously provides a new research idea and template for the damage risk assessment of other sports and national fitness activities. The method is beneficial to the rehabilitation practitioners to better know the limitations and the damage characteristics of the football player body, the multilayer sensor model is applied to daily training to predict the damage risk, the injury prevention training and the rehabilitation training of the player are guided, the damage occurrence is favorably reduced, and the long-term development of the football industry is promoted.

When the functional action test is carried out, the lower limb Y balance test is related to the single-leg supporting action in the football, and the comprehensive capabilities of lower limb strength, flexibility, balance, symmetry and the like during the exercise are embodied; the landing dislocation scoring (LES) test detects the three-dimensional angle of each joint of the lower limb in the jumping and landing process, and is closely related to the non-contact injury mechanism of anterior cruciate ligament, medial collateral ligament and ankle joint in football. The functional movement mode required by the athlete to perform the movement and the interaction of various risk factors are fully considered through purposeful multifunctional movement combination;

when the damage risk factors are screened, P <0.10 is used as a standard for judging that various types of basic information are screened as the damage risk factors; p <0.05 is used as a standard for judging various functional action evaluation indexes to be screened as damage risk factors, so that the screening reliability of the damage risk factors is improved;

functionally determining a functional action evaluation index interception point by adopting an ROC curve method, and constructing a scientific and reasonable functional action evaluation standard; furthermore, the risk probability diagnosis value is adjusted to 0.86439 according to the ROC curve method, the lower limb non-contact injury risk of the Chinese male college student football players is effectively predicted, the accuracy of diagnosis of the non-injury crowd is 87%, and the accuracy of diagnosis of the injury crowd is 93.3%.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.

Drawings

FIG. 1 is a ROC curve for medial differential diagnosis of non-contact injury of a lower limb of a subject following a Y-balance test of the lower limb according to one embodiment of the present invention;

FIG. 2 is a ROC curve of the composite value difference in the Y-balance test of lower limbs of a subject for diagnosing non-contact injury of the lower limbs according to one embodiment of the present invention;

FIG. 3 is a ROC curve for the diagnosis of non-contact injury in a lower limb of a subject according to one embodiment of the present invention using the total LESS score for the Y-balance test of the lower limb;

FIG. 4 is a diagram illustrating the structure of a multi-layered sensor model according to one embodiment of the present invention;

FIG. 5 is a diagram illustrating observation prediction of a multi-layered perceptron model according to one embodiment of the present invention;

FIG. 6 is a ROC curve for a multi-layered sensor model predictive probabilistic diagnosis of non-contact injury of a lower limb according to one embodiment of the present invention;

fig. 7 is a block diagram of a process of predicting the sports injury risk of a college student male football player by using the neural network model according to one embodiment of the present invention.

Detailed Description

The present invention is further described in detail below with reference to examples so that those skilled in the art can practice the invention with reference to the description.

< example 1>

The method for predicting the sports injury risk of the college student male football players by using the neural network model is shown in fig. 7, and is characterized by comprising the following steps:

1. screening test subjects

The initial subjects were 154 college student athletes from football team male in ten colleges and universities in China, and 149 subjects were counted against screening criteria and later follow-up surveys.

Inclusion and exclusion criteria: subjects were included if they were actively participating in the test content on the day of the test. If (a) injury occurs or rehabilitation therapy is received within the first half of the year of the test; (b) lesion tracking survey of incomplete one year; it is excluded.

2. Acquiring basic information data of a subject, wherein the basic information data comprises: general characteristics (age, height, weight, BMI), whether the player data was first issued, no damage history, on-site location, general characteristics of chinese boy student football players are shown in table 1 below:

TABLE 1 general characteristics of Chinese boy college student football athlete

	Age (y)	Height (m)	Body weight (kg)	BMI
					General of	20.05±1.70	1.77±0.06	69.13±8.62	22.11±2.36

3. Collecting evaluation index data of functional actions of a subject;

the examinee adopts the test program of the standard 'corrective training guideline' of the American national society for sports medicine, and is trained by a unified test method about a lower limb Y balance, single leg squat and landing error scoring system for at least 1 week, so that the test accuracy and reliability are ensured.

3.1 lower limb Y balance test (YBT-LQ)

The testing steps are as follows:

A. the lower limbs are long: the subject was supine (feet and shoulders as wide), in the pelvis neutral position, with the lower limbs straight and the hip joint in neutral position, and the distance from the right lower limb (from the anterior superior iliac spine) to the inferior border of the medial malleolus was measured.

B. The subject stands with the bare foot with the big toe aligned directly forward behind the center point of the lateral line, and the other leg extended as far as possible in the prescribed direction.

C. The right leg stretches forwards 3 times as far as possible, and the maximum value is taken to obtain the enough distance of the front side of the right leg; then, the left leg is changed and repeated for 3 times, and the sufficient distance of the front side of the left leg is obtained.

D. The right leg stretches out 3 times to the inner back side of the left leg, and the maximum value is taken to obtain the enough stretching distance of the inner back side of the right leg; then the left leg is changed and repeated for 3 times to obtain the back inner side extending distance of the left leg.

E. The right leg stretches out 3 times to the rear outer side of the left leg, and the maximum value is taken to obtain the sufficient distance of the rear outer side of the right leg; then the left leg is changed and repeated for 3 times to obtain the back and outer side extending distance of the left leg.

Note that:

before formal test, the subject can practice the legs for 3 times respectively, and the user can not hold the legs by hands without wearing shoes in the practice test process.

All measurements were accurate to 0.5 cm.

The following is a failure case, data is not available: firstly, losing balance when one leg stands; secondly, the standing foot moves obviously (the heel/tiptoe can be lifted up); landing support by extending feet; and fourthly, the feet can not return to the initial position.

If the subject failed to attempt 4 extensions, the direction was scored as 0 cm.

Functional action evaluation index:

the lower limb extension distance measurement device comprises a front side difference value, a rear inner side difference value, a rear outer side difference value and a comprehensive value difference value, wherein the comprehensive value (%) is a single-side (front side distance + rear inner side distance + rear outer side distance)/(3 times lower limb length) multiplied by 100%, and the difference value is an absolute value obtained by subtracting a right side extension distance from a left side extension distance.

3.2 Single leg squat test (SLS)

The testing steps are as follows:

video recorder position: 3m from the front of the subject, at the pelvic height of the subject;

the testee crosses the waist with both hands, looks at the dead ahead, the tiptoe is forward, the foot, knee, lumbar pelvis hip complex are in the natural neutral position;

the single leg squats to the height capable of bearing, then returns to the initial position, repeats the squat action for 5 times, and then changes the leg;

the test person observed the subject's knee, the lumbopelvic hip complex and the shoulder, the knee should be aligned with the second/third toe of the foot, and the lumbopelvic hip complex and the shoulder should remain horizontal and directed straight ahead.

Functional action evaluation index:

SLS Total score: and SLS total score is the total score of the wrong actions of the left leg and the total score of the wrong actions of the right leg, wherein when a certain side leg of the testee supports squatting, the following wrong actions are counted into one score: genu valgus, hip lifting, hip descending, trunk internal rotation, trunk external rotation. And finally, calculating the sum to respectively obtain the total scores of the wrong actions of the left leg and the right leg.

3.3 floor error scoring System (LESS)

The testing steps are as follows:

determining that the two video recorders are respectively positioned at the positions 4.5m away from the front and the side of the testee, and the height of the two video recorders is 1.22m away from the ground, and testing the step height: 30 cm.

Measuring the height of a subject, and marking the distance of half height of the subject before the step;

standing on a step, jumping forwards by feet to a mark with more than half height, and jumping upwards as high as possible instantly when the testee lands on the ground, wherein the distance of jumping forwards by the testee is less than half height, or jumping to the highest height after landing, and then retesting;

the testers scored according to the video and scoring criteria, each scoring a point for one error action, wherein the scoring criteria are shown in table 2 below:

TABLE 2 LES Scoring criteria

Functional action evaluation index:

total sagittal plane division: numbers 1, 2, 3, 4, 12, 13, 15, 16 in table 2 correspond to the sum of the item scores;

total score of coronal plane: the other item scores in Table 2 except 17 for the total score for the sagittal plane;

total LESS score: left total score sagittal + coronal total score +17 score (impression score)

4. Investigation of damage status

The system tracking investigation is carried out on the non-contact injury of the lower limb within 1 year after the test of the subject, the detailed injury information such as injury occurrence property (contact, non-contact), injury part, diagnosis and training stopping days and the like is recorded in detail, and the non-contact injury of the lower limb is defined as the injury of any part (including overuse or chronic disease) of the lower limb meeting the following conditions: (a) caused by other mechanisms than direct impact; (b) a need for medical intervention; (c) causing at least one day to be unavailable for participation in athletic related activities; wherein, impact damage: injuries caused by direct impact of other players or external force, such as the injuries caused by shoveling, elbow-beating and impact, etc.; spontaneous injury: the injury caused by direct impact of exogenic forces, such as sprain of knee ankle joint, sprain of muscle and ligament, chronic recurrent injury and pain, etc.

The injury survey data was compiled and the subjects were divided into an injured group (with non-contact injury of the lower limbs) and a non-injured group (without non-contact injury of the lower limbs) based on the non-contact injury of the lower limbs.

5. Results of the study

5.1 general and non-contact injury

Determining that the age, height, weight and BMI of the subject do not conform to normal distribution by adopting a K-S test;

determining the overall average age of the subjects, the average age of the injured group, the average age of the non-injured group and the P value of the injured group (P value of the age between the injured group and the non-injured group) by adopting a non-parameter test;

determining the overall average height of the testee, the average height of the injured group, the average height of the non-injured group and the P value of the injured group (the P value of the height between the injured group and the non-injured group) by adopting non-parameter test;

determining the overall average body weight of the subject, the average body weight of the injured group, the average body weight of the non-injured group and the P value of the injured group and the non-injured group (P value of the body weight between the injured group and the non-injured group) by adopting a non-parameter test;

determining the mean BMI of the subject population, the mean BMI of the lesion group, the mean BMI of the non-lesion group, and,

The P-values (P-values of BMI between injured and non-injured groups) in the non-injured group are shown in table 3 below:

TABLE 3 general characteristics of Chinese boy college student football athlete

	Age (y)	Height (m)	Body weight (kg)	BMI
					Non-injury group	20.01±1.71	1.77±0.06	69.19±8.79	22.14±2.40
Injury group	20.50±1.59	1.77±0.05	68.66±7.30	21.83±2.05
					P value	0.453	0.096	0.098	0.481

5.2, damage to non-contact location on field

The field position data belongs to non-numerical data, and the P value of the field position data between a damage group and a non-damage group is determined to be 0.823 by adopting chi-square test, wherein the field position comprises: frontier, backgate, frontier, goalkeeper, midcourt, undetermined; the on-site location of each type is shown in table 4 below with respect to the injury group and non-injury group personnel:

TABLE 4 location of male college student football players on the spot in our country

Position on field	Non-injury group	Injury group	General of
				Edge	17	3	20
Rear guard	44	5	49
				Front edge	18	2	20
Goalkeeper	8	1	9
				Medium field	21	3	24
To be determined	25	2	27
				General of	133	16	149

5.3, first-time team member and non-contact injury

The data of the initial member belongs to non-numerical data, and the P value of the data of the initial member between the damage group and the non-damage group is determined to be 0.068 by adopting chi-square test, wherein the conditions of the initial member on the personnel of the damage group and the non-damage group are shown in the following table 5:

TABLE 5 first-serve situation of male college student football athlete in China

5.4 history of injury and non-contact injury

The damage history data belongs to non-numerical data, and the P value of the damage history data between a damage group and a non-damage group is determined to be 0.000 by adopting a chi-square test, wherein the damage history is shown in the following table 6 about the personnel conditions of the damage group and the non-damage group:

TABLE 6 Damage-free history of male college student football athletes in China

	Non-injury group	Injury group	General of
				History of injury	82	0	82
Has a history of injury	51	16	67
				General of	133	16	149

5.5 lower limb Y balance test and non-contact injury

Determining that the front side difference value, the back inner side difference value, the back outer side difference value and the comprehensive value difference value of the subject do not conform to normal distribution by adopting a K-S test;

the P-values of the subjects' antero-medial, postero-lateral and composite differences between the injured and uninjured groups were determined using a non-parametric test (independent sample Mann-Whitney U test) as shown in Table 7 below:

TABLE 7 symmetry indexes of lower limb Y balance test for male college student athletes in China

Index (I)	General of	Injury group	Non-injury group	P value
					Front side difference (cm)	3.81±3.14	5.69±4.15	3.68±3.04	0.140
Inside difference (cm)	3.82±3.51	7.37±5.14*	3.59±3.28	0.014
					Difference of rear and outside (cm)	4.36±3.79	3.56±4.21	4.40±3.76	0.319
Integrated value difference (%)	3.29±2.63	5.31±2.99*	2.85±2.69	0.034

Note: p < 0.05.

5.6 Single leg squat test and non-contact injury

Determining that the SLS total score of the subject does not conform to a normal distribution by using a K-S test;

the P-value of the total SLS fraction of the subjects between the lesion and non-lesion groups was determined using a non-parametric test (independent sample Mann-Whitney U test), as shown in table 8 below:

TABLE 8 Chinese Male college student football athlete single leg squat score

Index (I)	General of	Injury group	Non-injury group	P value
					Total score	1.59±0.743	1.84±0.746	1.56±0.739	0.120

Subjects the score ratios for each item for the one-leg squat test are shown in table 9 below, for example:

TABLE 9 ratio of scores of the subjects

5.7 floor error scoring System test and non-contact Damage

Determining that the LESS total score, the sagittal plane total score and the coronal plane total score of the subject do not conform to normal distribution by adopting a K-S test;

the P values of the subjects LESS total score, sagittal total score, coronal total score between the injured and non-injured groups were determined using a non-parametric test, as shown in table 10 below:

TABLE 10 floor fault scoring System test scores for Subjects

Index (I)	General of	Injury group	Non-injury group	P value
					Total LESS score	8.28±1.736	9.63±1.088**	8.12±1.732	0.001
Sagittal plane general division	4.55±1.159	5.06±1.237	4.49±1.139	0.080
					General division of coronal plane	2.60±1.207	2.56±0.964	2.61±1.236	0.874

Note: p < 0.01.

Wherein, the sagittal plane index score of the subject is specifically shown in the following table 11:

TABLE 11 sagittal plane index score of Chinese male college student football athlete landing error scoring system

6. Injury risk factor determination

and determining the injury risk factors including height, weight, whether the first team member has no injury history, the difference value of the inner side of the back, the difference value of the comprehensive value and the total LESS score.

7. Determination of diagnostic criteria for injury risk factors for functional action evaluation indices

7.1 diagnostic criteria for inside Difference after Y balance test

Determining the upper limit and the lower limit of the rear inner side difference value, the group distance and the truncation points according to the rear inner side difference value of a subject, listing a cumulative frequency distribution table according to the group distance interval, respectively calculating the sensitivity and the false positive rate (1-specificity) of all the truncation points, representing the true positive rate by taking the sensitivity as an ordinate, representing the false positive rate by taking the sensitivity as an abscissa, and drawing a rear inner side difference value ROC curve (figure 1, wherein the diagonal line of the ROC curve has binding value generation);

as can be seen from fig. 1, the area AUC under the ROC curve is 0.759(P ═ 0.014), the jotan index is 0.444 (sensitivity 1, specificity 0.444) corresponding to the optimal diagnosis point of 2.25cm, i.e., 2.25cm is taken as the dividing line, 2.25cm is positive, <2.25cm is negative, the number of damaged persons in the positive group is 16, the number of undamaged persons is 74, the number of damaged persons in the negative group is 0, the number of undamaged persons is 59, the chi-square test calculates the OR value to be 1.216(P ═ 0.001, 95% CI: 1.105-1.339), as shown in table 12:

TABLE 12 evaluation results of inner side difference ROC curve after Y-balance test of lower limbs of test subject

a. By nonparametric assumptions

b. The original assumption is that: true area is 0.5

7.2, Y balance test comprehensive difference value diagnostic standard

Determining the upper limit, the lower limit, the group distance and the truncation point of the comprehensive difference value according to the comprehensive difference value of a subject, listing a cumulative frequency distribution table according to the group distance interval, respectively calculating the sensitivity and the false positive rate (1-specificity) of all the truncation points, representing the true positive rate by taking the sensitivity as an ordinate, representing the false positive rate by taking the sensitivity (1-specificity) as an abscissa, and drawing a comprehensive difference value ROC curve (figure 2);

as can be seen from fig. 2, the area AUC under the ROC curve is 0.725(P ═ 0.034), the john index is 0.410 (sensitivity 0.500, specificity 0.910), and corresponds to the optimal diagnostic point of 6.50%. That is, the results are shown in Table 13, in which 6.50% of the results were cut, 6.50% of the results were positive, 6.50% of the results were negative, 12 damages were observed in the positive group, 44 damages were observed in the non-damaged group, 4 damages were observed in the negative group, 89 was observed in the non-damaged group, and the OR value was 10.083 (P0.000, 95% CI: 3.207-31.703) by Chi-Square test:

TABLE 13 evaluation results of comprehensive value difference ROC curve of test subjects

a. By nonparametric assumptions

b. The original assumption is that: true area is 0.5

7.3 Total LES diagnostic Standard of LES after floor error Scoring System test

Determining upper and lower limits, group distances and truncation points of the total LESS score according to the total LESS score of a subject, listing a cumulative frequency distribution table according to group distance intervals, respectively calculating the sensitivity and the false positive rate (1-specificity) of all the truncation points, representing the true positive rate by taking the sensitivity as an ordinate, representing the false positive rate by taking the sensitivity as an abscissa, and plotting and drawing a ROC curve (figure 3) of the total LESS score;

as can be seen from fig. 3, the area AUC under the ROC curve is 0.768(P ═ 0.000), and the john index is 0.492 (sensitivity 0.875, specificity 0.617) corresponding to the optimal diagnosis point 8.5. That is, 8.5 is taken as a dividing line, 8.5 OR more is positive, 8.5 OR less is negative, the number of damaged persons in the positive group is 14, the number of undamaged persons is 51, the number of damaged persons in the negative group is 2, the number of undamaged persons is 82, and the chi-square test calculates an OR value of 11.255(P is 0.000, 95% CI: 2.456-51.577), which is specifically shown in Table 14:

TABLE 14 Total score ROC Curve evaluation results of the LESS test of the test subjects

a. By nonparametric assumptions

b. The original assumption is that: true area is 0.5

8. Multi-layer perceptron model construction

A prediction model (multilayer perceptron model) of the lower limb non-contact sports injury risk of Chinese college student male football players is constructed by adopting an SPSS19.0 multilayer perceptron neural network (MLP), and the prediction model specifically comprises the following steps:

(1) variable selection

The independent variables comprise factors and covariates, wherein the factors comprise whether the first generation exists, whether the non-injury history exists, the difference value of the inner side of the back, the comprehensive difference value and the total score of the LESS; covariates included height (cm), weight (kg), and covariates (x) were normalized: the mean (mean) is subtracted and divided by the standard deviation(s), (x-mean)/s.

Dependent variable: with or without non-impact injury of lower limbs

(2) Partitioning: the subjects were randomly grouped by percentage, 70% of the cases formed a training set, 20% of the cases formed a test set, and 10% of the cases formed a verification set;

wherein, the training group samples are used for training the neural network to obtain a multilayer perceptron model;

the validation set of samples is another independent set of data records used to evaluate the final neural network, and the validation set of samples error gives a "true" estimate of the predictive power of the model.

(3) The parameters for training the multi-layered perceptron model are specifically shown in table 15 below:

TABLE 15 model training parameters

Options for	Value of
		Initial Lambda	0.0000005
Initial Sigma	0.00005
		Center point of interval	0
Offset of interval	±0.5

(4) As shown in fig. 4, the structure of the multi-layered sensor model is specifically shown in table 16 below:

TABLE 16 Structure of multilayer perceptron model

Wherein, the 13 neurons of the input layer are specifically:

9. Performance of multilayer perceptron model

9.1 abstract of multilayer perceptron model

As shown in table 17 below, the model training time was 0.05 seconds and the training was terminated without a decrease in error over 1 consecutive step. The training set percentage of incorrect predictions was 3.4%, the test set percentage of incorrect predictions was 12%, and the validation set percentage of incorrect predictions was 11.8%.

TABLE 17 abstract of multi-layer perceptron model

9.2 classification results

As shown in table 18 below, the accuracy of the prediction of the training group, the test group, and the verification group was 100% in the case of no lower limb non-contact injury (no), and 0% in the case of lower limb non-contact injury (yes).

TABLE 18 results of classification

9.3 Multi-layer perceptron Observation prediction map

As shown in fig. 5, when the model is divided by the 0.5 prediction probability, the non-damage risk recognition effect is good, but the damage risk is difficult to recognize. Moving the cut point up to around 0.85 will cause most of the damaged samples in the 3 rd bin map to be reclassified correctly without significantly reducing the accuracy of the classification of the undamaged samples. Therefore, the boundary value of the damage risk prediction probability needs to be determined anew.

9.4 adjustment of probability diagnosis points of multilayer perceptron model

As shown in fig. 6, the ROC curve method is used to find the optimal probability diagnosis point of the multi-layer sensor model, where the area under the curve (AUC) is 0.934, the P value is 0.000, and there is a significant difference from the assumption that AUC is 0.5, which indicates that the probability prediction of the lower limb non-contact injury has higher accuracy, as shown in table 19 below:

TABLE 19 area under the curve (AUC)

And determining the predicted risk probability value corresponding to the maximum value of the Johnson index as the optimal diagnosis boundary value. The john index is the sum of sensitivity and specificity minus 1, as shown in table 20 below, where the maximum john index is 0.828 and the corresponding risk prediction probability value is 86.439%.

TABLE 20 Risk probability vs. Johnson index

9.5, accuracy of model after adjustment of damage risk probability

And (3) taking the risk prediction probability value of 0.86439 as an optimal probability diagnosis point of the lower limb non-contact injury risk, diagnosing the injury with the probability value less than 0.86439, and judging the injury with the probability value more than or equal to 0.86439 as no injury. The model has 87% accuracy in diagnosis of non-injured people and 93.3% accuracy in diagnosis of injured people, as shown in table 21 below:

accuracy of adjusted table 21 model

While embodiments of the invention have been described above, it is not limited to the applications set forth in the description and the embodiments, which are fully applicable in various fields of endeavor to which the invention pertains, and further modifications may readily be made by those skilled in the art, it being understood that the invention is not limited to the details shown and described herein without departing from the general concept defined by the appended claims and their equivalents.

Claims

1. The method for predicting the sports injury risk of the college student male football players by the neural network model is characterized by comprising the following steps of:

2. The method for predicting sports injury risk of college student male football players according to the neural network model of claim 1, wherein the subject in step S1 is a chinese male college student football player who has not suffered from injury in the first half year and has not received rehabilitation therapy.

3. The method of claim 1, wherein the basic information data in step S1 at least includes height data, weight data, whether the player is first, and non-invasive history data.

4. The method for predicting the sports injury risk of college student male football players by using the neural network model as claimed in claim 1, wherein the step S1 of collecting the functional action evaluation index data specifically comprises the following steps:

5. The method for predicting the sports injury risk of college student male football players according to the neural network model of claim 1, wherein the predetermined period of time is 1 year in step S2;

6. The method for predicting sports injury risk of college student male football players according to the neural network model as claimed in any one of claims 3 or 4, wherein the P values of the basic information data and the functional action evaluation index data between the injury group and the non-injury group are calculated in step S3, specifically:

7. The method for predicting sports injury risk of college student male football players according to the neural network model as claimed in claim 1, wherein the criterion that the P value is used as the criterion for determining whether various types of basic information and various types of functional action evaluation indexes are screened as injury risk factors in step S3 is specifically as follows:

8. The method for predicting the sports injury risk of college student male football players by using the neural network model as claimed in claim 1, wherein the step S4 of constructing the multi-layer sensor model specifically comprises the following steps:

9. The method for predicting the sports injury risk of college student male football players according to the neural network model as claimed in claim 8, wherein the determining neurons of the input layer according to the independent variables is specifically:

10. The method for predicting the sports injury risk of college student men according to the neural network model of claim 9, wherein the optimal probability diagnosis point for diagnosing the non-contact injury is determined to be 0.86439 by using the ROC curve method according to the data of the output layer.