A construction principle of a technical system for diagnosis and rehabilitation of the musculoskeletal system based on accelerometer method, together with synchronization algorithms measuring patient parameters, is considered. The optimal accuracy estimations of the technical parameters of the accelerometric goniometer system are determined; they are the sample rates of the accelerometer signal converters, the required sensitivity of the sensor, etc. The advantages of the proposed approaches to the construction of rehabilitation and diagnostic systems of the musculoskeletal system are adaptability and reliability of the diagnoses.
Accurate diagnosis and objective assessment of the treatment efficiency of
motor function disorders to date remains one of the urgent problems of modern
traumatology and orthopedics. The large number of evaluation approaches and
techniques reveal a lack of reliability of the proposed criteria for diagnosis and
assessing recovery efficiency. For example, the diversity of human movement is
characterized by a number of parameters: torque, speed, complexity of
trajectories, changes in the level neuromuscular and brain activity. In existing systems,
goniometry and diagnosis of musculoskeletal system mainly take into account
only the kinematic parameters of the skeletal system, regardless of the bone
structure and neurophysiological parameters of state of the patient [
Formation of the goniometric criteria and selection of the optimal working
parameters of the system rehabilitation is carried out on the basis of statistical
clinical studies of patients under normal conditions and in the presence of
deviations. In medical diagnostics, the assessment of the angular movement indicators
and their permissible deviation from the normal ones is done according to the
joints motion estimation table “Regulations on military-medical examinations”
(approved by the Russian Federation Government Decree No 565 dated July 4,
2013) [
On the basis of the data presented in Table 1, it can be said that indicator deviations of joint angles by 1∘ are violations. Consequently, goniometric system must meet the requirements of measurement accuracy, with the threshold sensitivity of the measurement of mutual deviations and measuring range of motion being at least 1∘ . In this case, the measurement error must be smaller than this threshold. It should be noted that at present mechanical goniometers are widely used in medical diagnostics. Their accuracy does not meet the present requirements stated by the system of this class. Low accuracy of the measured parameters might be resulted from the design features of the device, and a high degree of subjectivity of diagnosis due to professional experience and the influence of the human factor [5]. 3
The emergence of the movement results from neuro-cerebral human activity aimed at the implementation of any function. Therefore, the main objective of the design of goniometric systems is to assess the effectiveness of motor actions with respect to the application efforts of their execution [6].
For a comprehensive solution of this problem, we are to study design aspects of an automated diagnostic system of human musculoskeletal apparatus. The design is based on the synthesis of adequate informative physiological methods such as goniometry (accelerometer), computed tomography, electroencephalography (EEG) and electroneuromyography (ENMG). For this purpose, a medicotechnical basis was formed. A block diagram of an automated system of complex real time diagnostics of the musculoskeletal system has been developed. The diagram consists of the circuit of accelerometric goniometer, an electroencephalograph and electromyography imaging is used as well (Fig. 1).
Neural network Evoked potential
DB proScigenssailng NeBuraroinmpuasrcaumlaertoeprstions unit Goniometer angles
Measurements
DB
Mathematical
model Evoked potential processing
Neural network С alibration Patient
Synchronization
EEG
Electroencephalogram
ENMG Electroneuromyography Accelerometers
Acceleration Actuators
Test Test methods
consists of a chain of inverters biokinematic driving parameters of the locomotor apparatus of man (accelerometric goniometer), recording units of psycho- and neurophysiological parameters (EEG and ENMG), and the registration equipment for bone and structural parameters (tomography).
Neural network Decease
DB
Model DB
Diagnosis knowledge base
DSS Diagnosis and is stored in the model data base. The model that most closely matches the time series are instantiated. Pain thresholds and threshold of sensitivity of the patient for generating control signals to the actuators are determined by neural network algorithms. This is possible via a feedback method, whose implementation is carried out by the processing unit of evoked potentials of brain (patient’s reactions to test stimuli). Based on the processed data, operation mode of the actuators is generated and selected from a database of test techniques.
h, cm m, kg αi αi , βi, γi Osd, mkV Psd, mkV Csd, mkV Fsd, mkV CPBm, uV CPBs, uV F-wave, uV
based on databases of the time series, diseases and evoked potentials, an approximate diagnosis of patient is determined.
It should be noted that the above adaptive goniometric control system includes both stationary and mobile measuring systems [6]. The number of monitored parameters is determined according to the severity of the patient’s pathology. In the case of low severity injuries, it is sufficient to use of a several portable accelerometric goniometers, guaranteeing the freedom of the patient’s movement. In the presence of more serious violations in the functioning of the musculoskeletal system is suspected, the accelerometric goniometers coupled with electroneuromyograph (Fig. 3a), EEG data and computing tomography (Fig.
It is shown that the dynamic activity of brain neurons relating to the implementation of tool movements, typically starts 50-150 ms. prior to the occurrence of EMG activity and ended after a traffic stop. Thus, the joint reaches equilibrium during the dynamic development of the motor cortex neuronal activity phase long before the establishment of steady equilibrium level of neural activity. The maximum value of the mean frequency of neural activity in one bin of duration is 50 ms. In dynamic phase, reactions of neurons did not correlate with the magnitude of the equilibrium steady-articular angle (Fig. 5). At the same time, a positive correlation was revealed between the average frequency of neural activity in the whole dynamic phase and magnitude response of articular angle [7, 8].
The presented results show that the maximum level of motor neuron activity depends primarily on the joint movement velocity and the duration of the movement. Thus, the obtained data contribute to definition of the criteria of permissible values, characterizing limits of the patient physiological parameters with respect to the normal ones; the limits are determined according to the degree of deviation and the conditions of pathology. Neurophysiological criteria are also formed based on statistical analysis of clinical studies of patients under normal conditions and in the presence of deviations. 4
Development of DSS based on the dynamics analysis of the recorded time series for goniometric, kinematic and neurophysiological parameters is a complex and multicritelial problem. The algorithms of the DSS are based on Bayes’ rule. The rule accounts various heterogeneous types of input data expressing many kinds of deceases of the musculoskeletal system and a large number of symptoms. Bayes’ rule in a generalized form is as follows [9]: ( | ∩ · · · ∩ 1) = ( 1 ∩ · · · ∩ | ) · ( ) , ( 1 ∩ · · · ∩ ) where ( ) is a priori probability of the diagnosis , and 1 . . . are functional physiological parameters. (1)
N=21
N=25 ms 2 ms
N=29 3 ms 4000 2000 0 2000 .
The probability of the hypothesis in the presence of certain abnormalities in the recorded data is calculated based on the prior probability of the hypothesis without confirming abnormalities and the likelihood of having abnormalities in conditions that hypothesis is correct (event ) or incorrect (event ¯ ). Therefore, for the problem of diagnosis of diseases of the musculoskeletal system, it appears that This formula requires ( · )2 + 2 + 2 calculations of probability estimates, where is the number of possible diagnoses, and is the number of different variations. In order to calculate the total probabilities ( 1 ∩ · · · ), we are to calculate ( 1/ 2 ∩ · · · ∩ ) · ( 2/ 3 ∩ · · · ∩ ) · . . . · ( ).
Therefore, the model for automated diagnostic expert system will be based (2) (3)
Let the pathology probability ( ) be equal to . The program generates a condition (parameters in the presence of pathology) and calculates the probability ( | ) depending on the it’s implementation. The answer “Yes” ( ) confirms the above calculations, the answer “No” ( ) does it too but with probability (1 − ), and (1 − ) instead and . Thereafter, the a priori probability ( ) is replaced with ( | ). The program execution is cyclic, with the constant value ( ) refining at each iteration. The general scheme of the diagnosis selection algorithm is shown in Fig. 6.
logical and structural parameters; then the program retrives information on the number of the diseases recorded having the corresponding symptoms from in the database ( is the number of relevant deviations disease, is the number of the disease in question: 0 6 6 ).
Step 2. Set counter of a disease incrementally from the initial state = 0 till 6 .
set and to the selected disease, to prioritize detected deviations. Deviations with the minimal likelihood are excluded from the probability set ( is the selected number of deviations in the set, the number of the current deviation 0 6 6 ).
Step 4. Set the deviation counter incrementally from the initial state = 0 to 6 .
interval [−5, +5] (a scale). If the value belongs to the interval, then the program calculates the proportion of the degree of affiliation to a particular diagnosis parameters, using the corresponding weighting coefficients.
Selecting the max value of an array of prior probabilities of diagnoses and the formation of an advisory diagnosis
Selecting the most probable diagnosis
End
Begin Enter the initial information:physiological parameters, the number of diseases (N)
n=n+1, 0≤n≤N Viewing probability P(d), with the exception of deviations minimum probability
j=j+1, 0≤J≤N Selection of deviation j with max value of a priori probability of the formation of the corresponding issue yes yes yes
Answer «-5» Answer «+5» Answer «0» no no no j=J yes
no Values calculated probabilities based on the works of deviations probabilities yes n=N no Step 8. Figure out new probabilities in Bayes rules. Specify the minimal and maximal values of the probabilities for each disease based on the currently existing a priori probabilities and assumptions that the remaining evidence will speak in favor of the hypothesis or contradict it. This step calculates the total conditional probabilities for each deviations. The hypotheses whose the minimal values are above a certain thresholds are considered as possible outcomes (possible diseases) and are subject to further diagnosis.
Medical information system, which implements this algorithm, produces a finite set of the recommendations for doctor, emphasizing the presence of deviations registered with the diagnostics system sensors. Limiting the selection decision by a finite set of possible diagnoses is to reduce the probability of setting wrong preliminary diagnosis, eliminate human factor and increase the objectivity of the diagnosis of diseases. 5
In order to develop the control unit of diagnostics system of the musculoskeletal system, we propose a method of computer support of diagnosis setting based on fuzzy logic and artificial neural networks. The method is represented in the form of two structural units: decision-making unit and knowledge base.
The decision (a diagnosis) is produced in two stages [10]. At the first stage – the preliminary diagnosis–, the system determines in advance the possible pathology and generate diagnostic recommendations based on data from the medical records and X-ray images. Then, at the stage of the goniometric diagnosis, together with EMNG, the recommendations are confirmed or rejected on the basis of the analysis of the obtained information, with immediate neural network processing by the diagnosis system.
system by means of fuzzy production rules (FPR) [11, 12]. The initial values of the parameters (used in FPR) for the normal cases or the pathological ones are determined at the beginning on the base of experts’ opinions. The values are adjusted with neural network engines.
of so-called factors of pain diagnosed by EEG. Pain is one of the most important factors in the diagnosis of diseases of the musculoskeletal system. On the basis of these features we justify the choice of the rules of fuzzy productions of the form:
IF <Condition1 = true> AND ... AND <ConditionN = true> THEN <Consequent1 = true> AND ... AND <ConsequentN = true>
) are selected according to a statistical base of the analysis of medical records and printed materials [13].
responding to one of the values of the input linguistic variables) were determined by means of the following expert evaluation techniques. Let expert 1 believe that the specific value of * belongs to the fuzzy term set at 1 6 * 6 1; 21 22 · · .· 2... ⎟⎟⎠ .
... ... . . 1 2 · · · (4)
In the figure, the horizontal axis shows the value of strict variable under fuzzification, where
= 1, is an expert number. The vertical axis displays fraction of all the experts who believe that its value of a belongs to this linguistic value of the linguistic variable. This initially plots the MF’s, which is obtained at ∈ ( ( ),
( )) ∪ ( and will get a linear form if the least squares method is used. ( ), ( )); the plot is curvilinear
IF <IN_LV1=VALUE11> OR ... OR <IN_LV1=VALUE1m> AND <IN_LVn=VALUEn1> OR ... OR <IN_LVn=VALUEnm>
Each of the terms <IN LVi=VALUEi1> OR ... OR <IN LVi=VALUEij> consists of the -th values subconditions <IN LVi=VALUEij>, where VALUEij is the -th value of in subconditions. Its number is determined by the number of input : . Let the truth degree of subconditions with the number be, respectively, . The following RFP matrix is formed for all the subconditions: =
/ ︁∫ ; .
︁∫ where set = 1, MF Using this matrix at the 3-6-th stages, we get the formula for calculating the confidence coefficient – the correctness precision of the system solution, – which is calculated for each of the possible diseases identified by DSS. · ︂(
︂( ˜ · ︀∑ =1 ·
( ︀∑ =1 ( ( ))) )︂ / ˜ · ︀∑ =1 ·
( ︀∑ =1 ( ( ))) )︂ , (5) and
are the left and right limits of the carrier interval of fuzzy under consideration; are the weighting coefficients of the rules, is the number of RFP, which is determined in the consequent of the -th term of MF
; ˜ ( ) is the antecedent MF -th term of -th
It should be noted that the weights of the rules vary depending on the occurrence of new facts and results of fuzzy inferences at the previous stages. To resolve this uncertainty in form of an adjustment of RFP weights, an inference system is represented as a hybrid, i.e., fuzzy neural network (Fig. 8). Its structure is identical to the multilayer network, but its layers correspond to the stages of fuzzy inference, which has continuously carry out the following procedures: - Input layer performs fuzzification function based on the specified input membership functions; - Output layer implements the defuzzification function; - Hidden layers reflect the totality of the RFP and the output stages: aggregation, activation and accumulation.
Inputs layer 1 layer 2 layer 3 layer 4 layer 5 layer 6 layer 7 layer 8 Outputs ~ ̀ umax1 ~ ̀ umaxvmax max
min between
and an intermediate Neurons , and
indicated in Fig. 8 act as appropriate mathematical functions. Neurons, marked with “×” is transmitted to the output product of the input signals. Symbol “≡” marks neurons establishing a correspondence , and symbols “⋀︀ ” are neurons realizing operation fuzzification for each term of INLV. Node “/ ” denotes division of the input value by the sum of the weights of active rules. Neuron “ ” implements the function of defuzzification, with applying the method of gravity center.
The fuzzy rule selection engine is represented as INLV inputs having “0” (a rule is selected) and “1” (a rule is not selected) logic levels weights multiplied by the corresponding membership functions ( ). Here index ∈ 1, is the number of and index ∈ 1,
is the number of its term.
A neural network is trained by the algorithm of error back-propagation modified for use in the fuzzy neural networks. The layers of neurons with specified parameters are represented by one layer with a complex activation function, fuzzy artificial neural network (ANN) is identical a three-layer ANN with one hidden layer. Thus, network training is reduced to a three-layer perceptron learning. It is worth noting that the fuzzy neural network is used only in the case of changes of the DSS structure (change aggregate RFP, input or output LV), and in the case of the appearance of new evidence proving or disproving the previously known data in the literature or medical practice. 6
This section is devoted to the results of the benchmark tests of the calculated model; the results are obtained with an installed bodily machinery of the human skeleton. During the investigation, we used the method of mechanical goniometry in collaboration with an orthopedic doctor. The method of accelerometer goniometry usage is proposed in [5].
In addition, the diagnosis was carried on 20 patients who underwent rehabilitation after complicated shoulder injury and wrist. It should be noted that in all cases the diagnosis was a medical opinion on the normal rates. However, based on the neural network processing of the recorded data of the goniometric control, the electroneuromyographic and the tomographic control, a dial of a estimation of a motion in the joints from the Table 1, it was found that in 14 cases the medical diagnosis coincided with the diagnosis of DSS, in 4 cases were diagnosed of a functional deviation of the work wrist in a small extent, and in 2 cases were more prominent deviation of the combining elbow-shoulder joint in small extent. 7
Data processing algorithms and approaches to designing a system of diagnostics of the musculoskeletal system are presented in the article. The obtained results allow one to
- define the evaluation criteria of the “current state” of the musculoskeletal system;
- determine the severity of biomechanical disorders with a high degree of confidence; - predict the biomechanical disorders of the musculoskeletal system; - have the possibility of a science-based rehabilitation prognosis; - create and optimize individual training programs that promote the advancement of technical training of athletes and prevent the diseases.
loskeletal system based on the proposed system is an informative method of detecting violations. This technique is recommended to be used as a supplement to conventional methods of examination of the musculoskeletal system condition, as well as a stand alone technique.
work was supported by
Acknowledgments. 16-38-00704 mol a. 5. Grecheneva, A.V., Kuzichkin, O.R., Dorofeev, N.V., Konstantinov, I.S.: The use of the accelerometer in the goniometric measurement system. In: Information systems and technologies, ISSN 2072-8964, No 4 (90) 2015 July-August, pp. 5-10. 6. Grecheneva, A.V., Dorofeev, N.V., Kuzichkin, O.R.: The use of the accelerometer in the goniometric measurement systems. In: Mechanical engineering and life safety, ISSN 2222-5285, No 1, 2015. pp 55-58. 7. Sarnadskiy, V.N., Fomichev, H.G., Monitoring of spinal deformity by computer optical topography. - Benefit For Physicians Health Ministry. -Novosibirsk, NIITO 2001. 8. Zokirhodzhaev, M.A.: Diagnostic criteria and methods of rehabilitation therapy flat feet in children. In: The doctor-graduate student, No 5.2 (54), 2012. - P. 234-239. 9. Kobrinsky, B.A.: The logic of the argument in the decision-making in medicine. In:
STI ser.2. - 2001. - No 9. - P. 1-8. 10. Podolny, M.A., Taperova, L.N.: Designing medical diagnostic system based on the model of fuzzy inference. In: Eighth Nat. Conf. by Art. Intelligence with int. uch .: mat. Conf. V.2. M .: FIZMATLIT, 2002. - P. 641-646. 11. Babkin, E.A., Kozyrev, O.R., Kurkina, I.V.:The principles and algorithms of artificial intelligence, Nizhny Novgorod: Nizhegorod. state. tehn. University Press, 2006. 132 p. 12. Chernov, A.V., Shtanko, S.I., Rogozin, M.A., Podvigin, S.N.: The use of mathematical analysis for the diagnosis of borderline mental disorders in medical students.
In: Doctor-graduate student, No 1 (50), 2012. P. 43-49. 13. Choporov, O.N., Razinkin, K.A.: Optimization model of choice of the initial plan of control actions for modeling information systems. In: Control Systems and Information Technology, 2011. Vol. 46, N 4.1. P. 185-187.