RU2704913C1 - Neural network analysis method of cardiac condition - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: invention refers to medicine, particularly to cardiology, and can be used in automatic mode for diagnosing patient's heart condition by electrocardiographic examination of a patient in screening or emergency and emergency care. Diagnosis of the heart condition is suggested, which is carried out by analyzing output signals of neural networks, trained for recognition of direct and reciprocal signs of MI in ECS of 12 common leads, and construction of decision rules. That is ensured by the following: separation of the left ventricular surface into regions for which the revealed direct and reciprocal signs of myocardial infarction (MI) correspond to a specific stage and a specific type of MI by the depth of the lesion; determining the stage of MI and the type of MI by the depth of involvement by heart; determination of localized infarction of the established stages.

EFFECT: invention provides a diagnostic conclusion on the patient's heart condition, in which in the presence of myocardial infarction (MI) of the left ventricle of the heart, its localization is indicated, stage of development (peracute, acute, subacute, cicatricial) and types of cardiac wall involvement (transmural, subepicardial, subendocardial) as well as complete and accurate evaluation of patient's heart condition irrespective of doctor's level and experience of its operation.

1 cl, 12 dwg

Description

The present invention relates to medicine, in particular to cardiology, and can be used automatically for the differential diagnosis of myocardial infarction (MI) according to the electrocardiographic examination of the patient during screening or in emergency and emergency care.

A known method of processing and analysis of an electrocardiogram (EX) for the diagnosis of myocardial infarction [1], including automatic removal of the patient's EX in three standard, three reinforced and six chest leads, its registration, digitization, isolation of the cardiac cycle and combined analysis, with a combined analysis of EX in all leads perform the detection of ECS signs of myocardial infarction, assessing the state of the heart according to the results of the analysis of ECS in the amplitude-time, frequency-time and amplitude-phase regions, as well as the results ompleksnogo contour analysis and neural network analysis ECS ECS, and the formation of a diagnostic conclusion. Moreover, a comprehensive contour analysis of EX [2], which leads to the well-known combined analysis of EX, allows diagnosing myocardial infarction (MI) of certain stages and localizations, but is limited by the set of contour subgraphs and does not allow, in the presence of signs of IM, in the EX to form extended diagnostic conclusions containing comprehensive information about myocardial infarction (that is, information that includes information about the localization, stage of myocardial infarction and the depth of damage to the walls of the affected areas of the heart). Amplitude-time, frequency-time and amplitude-phase analyzes when diagnosing MI do not reveal the stages of its development and the type of depth of the lesion. In addition, when determining the pathology according to the individual elements of the EX and without separating the signs into direct and reciprocal, ambiguities arise when determining the localization of MI (for example, both septal and posterior myocardial infarctions are detected by changes in the EX of leads V1 ... V3, only the first by direct signs, and the second is reciprocal). The disadvantage of the neural network analysis of the pacemaker, which leads to the well-known combined analysis of the pacemaker for the diagnosis of myocardial infarction, is that the number of neural networks is equal to the number of leads and their training can be carried out only for a certain type of myocardial infarction according to the depth of the lesion and a specific stage of its development (since ECS for types and stages of MI significantly differ). In addition, the absence of decisive rules for determining localization does not allow the formation of a diagnostic conclusion about the state of the patient’s heart, relating to one of the k conditions of the heart.

Closest to the proposed invention by the method of processing and analysis of the electrocardiogram (ECS) for the diagnosis of myocardial infarction is the method [3], which consists in the fact that the continuous electrocardiogram is recorded, pre-processed and presented as an n-dimensional vector, create m n- dimensional reference information vectors characterizing one of the heart conditions, and compare the n-dimensional vector of the registered electrocardiogram with the set of m n-dimensional vectors of reference of information characterizing one of the states of the heart, it is decided that the n-dimensional vector is within or outside the threshold range of reference information vectors characterizing one of the states of the heart, they additionally carry out a neural network analysis of k states of the heart, which creates (k-1) * m n-dimensional vectors of reference information, carry out the construction of decision rules, carry out a neural network analysis of the n-dimensional vector of the registered electrocardiogram taking into account the decision rules, and output the result related to Nome from k heart conditions.

Moreover, for the neural network analysis of ECS, the LVQ neural network (NS) is used, which is a development of the structure of the Kohonen network and has the possibility of training with a teacher [4, 5].

To perform a neural network analysis of an n-dimensional vector of a registered electrocardiogram, k * L (L is the number of leads) of neural networks is constructed to analyze each of the k cardiac conditions in each lead.

The construction of decision rules is carried out on the basis of direct and reciprocal signs of topical diagnosis for each of the k conditions of the heart.

The conclusion of the result relating to one of the k states of the heart is carried out on the basis of the choice of the k-th state of the heart, in which the maximum number of signs is revealed.

The figure 1 shows the algorithm of the known method of neural network analysis of the state of the heart.

The figure 2 shows a diagram of the formation of a diagnostic conclusion about the state of the patient’s heart, related to one of two heart conditions (healthy condition and myocardial infarction, anterior and anterior septal localization), according to the results of the known neural network analysis of EX.

According to the claims, a known method of neural network analysis of the state of the heart consists (see figure 1):

firstly, from previously known actions [1], such as “Registering EX”, “Preprocessing EX”, “Presenting EX as an n-dimensional vector”, “Outputting the result”;

secondly, from the introduced actions, such as “Formation of a training sample”, “Training of neural networks”, “Neural network analysis of heart conditions”, “Construction of decision rules” and “Analysis of neural outputs”.

Registration EX. At this stage, the removal and registration of the EX using the registration device EX. When registering an EX, issues of electrical interaction between the electrode and skin tissue at the location of the electrodes are resolved; patient electrical safety; analog filtering and gain EX; discretization and digitization of EX; pairing the registration device EX with the computer. After the registration phase, digitized EXs are received at the computer input.

Pre-processing EX. Here the questions are solved:

- Digital filtering, including noise reduction and artifact elimination. Using digital filtering, irregular beats are removed from the registered EX;

- highlighting the characteristic features of ECS: the amplitudes of P, Q, R, S, T teeth and the values of the midpoints between P and Q teeth, S and T teeth;

- obtaining the values of the coefficients after performing the Fourier transform and / or wavelet.

After the pre-processing stage, data are obtained characterizing the features of the registered EX, necessary for the formation of an n-dimensional vector.

Representation of EX as an n-dimensional vector. At this stage, an n-dimensional vector is formed from the obtained values. Two options for the formation of an n-dimensional vector can be distinguished. In the first case, the values of the n-dimensional vector are the characteristic features of the ECS: the amplitudes of the P, Q, R, S, T teeth and the values of the midpoints between the P and Q teeth, S and T teeth. In the second case, the values of the n-dimensional vector are the values of the coefficients after the Fourier transform and / or the wavelet transform.

The formation of the training sample is based on the formation for each k of the state of the heart of its “own” ECS database for each involved lead and includes:

1. Analysis of registered EX by an experienced cardiologist and assignment of registered EX according to the results of the analysis to one of the k conditions of the heart.

2. Statistical processing of the “own” ECS database generated for each of k heart conditions. For this:

- the amount of ECS of one state of the heart should be at least 50;

- the number of ECS of one state of the heart is divided into two equal samples;

- put forward a null hypothesis about the belonging of the considered samples to one general population;

- t-criterion parameters are calculated [6];

- under the condition t _fact > t _tab , the correlation coefficient is considered significant and the null hypothesis about the belonging of the considered samples to one general population is accepted.

1. The division of the number of ECS of one state of the heart into two parts (training and control) in a ratio of 4: 1.

2. The separation of the training part of the number of ECS of one state of the heart into two components (without noise and noisy) in a ratio of 3: 1.

3. Noise ECS.

4. The use of received EX (noisy and noisy) for training k neural networks.

5. The use of control ECS to test the training of k neural networks.

Neural network training. In the course of this action, k * L (L is the number of leads) of the neural networks are trained, for each of which an “own” ECS database is used, obtained as a result of the previous action.

According to the authors, the adaptation of the structure parameters of the LVQ NS for the analysis of ECS [7] consists in changing such NS parameters as: the dimension of the input vector; the number of neurons in the competing layer; number of inner layers.

The neural network analysis paradigm includes the discretization of the ECS, the isolation of the ECM signs of MI, their processing in the NS, the assessment of the state of the heart.

For each of the 12 ECS signals in the conventional leads, an “own” LVQ network is used. The original analogue electrocardiogram (meaning an EX signal in each of 12 common leads) is digitized in accordance with the SCP-ECG standard for digital EX-exchange [8] and is presented in the form of 500 readings per cardiac cycle. Isolation of one cardiocycle [9] is a distinctive feature: one cardiocycle is sufficient to detect myocardial infarction [10].

The values of the readings of one cardiocycle EX go to the inputs of NS LVQ in the form of an input vector x ^h .

определяется стандартом SCP-ECG для обмена цифровыми ЭКС [8] и равна

= {1÷500}. Input Vector Dimension

defined by the SCP-ECG standard for digital EX-exchange [8] and is equal to

= {1 ÷ 500}.

The signal value at the output of the HC LVQ is determined by the following formula:

(one)

where x _i is the i-th element of the input vector,

– i-й элемент вектора весов m-го нейрона конкурирующего слоя,

- the i-th element of the vector of weights of the m-th neuron of a competing layer,

– j-й элемент вектора весов k-го нейрона линейного слоя,

- the jth element of the vector of weights of the kth neuron of the linear layer,

F _compet - transfer function of a competing layer, revealing the winner neuron,

F _lin is a linear function of activation of neurons of the distribution layer,

S ₁ - the dimension of the input vector NS,

S ₂ - the number of neurons in the competing layer,

Y _k is the value of the k-th output of the National Assembly.

When analyzing the input data, the total Euclidean measure should be minimized:

(2)

where h is the number of the input vector,

c (h) is the class of disease corresponding to the vector x ^h .

Analysis of expression (2) shows that the task of finding the minimum of D is equivalent to finding the maximum of the expression:

(3)

The number of neurons in the Kohonen layer N _coh corresponds to the number of clusters into which the input data is supposed to be divided. The following formula is used to determine this value:

(four)

where α is the number of clusters per one class of diseases.

After the determination of the parameters, the training of HC LVQ for each lead is carried out. The purpose of training NS in the analysis of ECS is such an ordering of neurons (selection of their weights) that minimizes the value of the expected distortion, estimated by the approximation error of the input vector x ^h , by the weights w _{x (h) of} the winner neuron in the competition. This error (or recognition error) can be represented as:

(5)

where p is the number of input vectors,

w _{x (h)} is the weight vector of the winner neuron upon presentation of the vector x ^h .

The criterion for completing the learning process of the NS is the fulfillment of the condition E _q <ε, where ε is the error value set by the user before the start of training (for example, ε = 0.01).

To train NS LVQ for the analysis of ECS, the “convex combination” method is used [11]. The NS LVQ learning algorithm works as follows:

.- step 1. The same initial value is assigned to all weights of the Kohonen layer

.

- step 2. The next vector x ^{h is} selected from the training set.

- step 3. Elements of the vector x ^{h are changed} in accordance with formula (6):

(6)

where t is the number of the era of learning,

β (t) is a monotonically increasing function, varying from 0 to 1 as you learn.

- step 4. The Euclidean distance between the vector x 'and each neuron of the competing layer is determined by the formula (7)

(7)

– значение i-го элемента вектора x',Where

Is the value of the i-th element of the vector x ',

– значение i-го веса m-го нейрона конкурирующего слоя.

Is the value of the ith weight of the mth neuron of a competing layer.

- step 5. 2 neurons with the smallest Euclidean distance d _m to the input vector x 'are determined.

- step 6. When condition (8) is fulfilled, the transition to step 7 occurs; otherwise, return to step 2

(8)

where d ₁ is the Euclidean distance between the vector x 'and the weights of the first neuron found,

d ₂ - Euclidean distance between the vector x 'and the weights of the second found neuron,

ε is a value selected from the range 0.2–0.3 [5].

- step 7. The weights of the found neurons change. With the correct classification of the vector x ', a correction is performed according to formula (9). In case of false classification, “rejection” is performed according to the formula (10).

(9)

(10)

where η _t is the value of the speed of learning NS.

- step 8. When minimizing the learning error E _q (5), the learning cycle is terminated, otherwise the transition to step 2.

Neural network analysis is performed simultaneously by parallel analysis of registered ECS by each neural network. The output signal of the NS takes on a value of 1 in the case when the n-dimensional vector of the input EX corresponds to the vector for which this NS is trained, and 0 otherwise. The analysis results in an array of data from the outputs of neural networks.

The construction of decision rules for analyzing the outputs of neural networks is based on the fact that deviations in each of the localizations of MI are not manifested in all leads. To build the decision rules for choosing one of the (k-1) cardiac conditions related to MI, we used combinations of the presence and absence of direct and reciprocal signs of MI in the ECS leads for eight different locations [12]. Decisive rules for constructing logical blocks for eight IM diagnoses:

1) Anterior and anterior septal MI

, (11)

, (eleven)

2) Anterobasal MI

, (12)

, (12)

3) Front Common MI

, (13)

, (13)

4) Lateral MI

, (14)

, (fourteen)

5) Lateral basal MI

, (15)

, (fifteen)

6) The posterior diaphragmatic (lower) MI

, (16)

, (16)

7) Circular MI

, (17)

, (17)

8) Post-basal myocardial infarction

, (18)

, (eighteen)

,

,

,

,

,

,

,

,

,

,

,

– данные с выходов нейронных сетей, выявляющих здоровые ЭКС.here

,

,

,

,

,

,

,

,

,

,

,

- data from the outputs of neural networks that identify healthy EX.

Analysis of the outputs of neural networks. Data from the outputs of neural networks is fed to logical blocks designed to identify the corresponding state of the heart in accordance with the decisive rules, and then to the priority encoder, designed to highlight the corresponding state of the heart. At the output of the priority encoder, an input line number code is generated, to which a positive input signal comes from the output of one of the k logical blocks of neural networks involved in the analysis of ECS. With the simultaneous receipt of several input signals, an output code is generated corresponding to the input with the highest number, i.e. high entries take precedence over low entries. The result of the work at this stage is the number of the diagnostic conclusion about the patient’s heart condition, to which the analyzed EX is assigned.

Conclusion of the result. The number obtained as a result of analysis of the outputs of neural networks is assigned a verbal description of the diagnostic conclusion about the patient’s heart condition, which is communicated to the user.

To justify the shortcomings of the known method of neural network analysis of the state of the heart, information on the distinguishing features (signs) of MI of various stages, types of MI according to the depth of the lesion, as well as information on the main localizations of the MI are given below.

Myocardial infarction occurs during the transition of a normally functioning myocardium to necrotic. As a rule, 4 stages are distinguished in the development of MI: acute (acute coronary syndrome), acute, subacute, and cicatricial [12-18].

In the acute stage, which lasts from several hours to a day, myocardial damage occurs. In the subsequent acute stage of a heart attack (lasting up to 2-3 weeks), the myocardial muscle fibers die and a necrosis site forms. During the subacute stage of myocardial infarction (usually lasting 1-3 months), some of the muscle fibers that have received deep damage go into the necrosis zone, and the rest of the fibers are partially restored and become ischemic. The damage zone gradually disappears, and the necrosis zone stabilizes. In the cicatricial stage of myocardial infarction (months, years), tissue scarring occurs at the site of necrosis, which tightens neighboring healthy areas of the myocardium. Simultaneously with the formation of a scar, compensatory hypertrophy of the remaining muscle fibers occurs, which leads to a decrease in the zone of myocardial infarction (transmural MI can turn into non-transmural). Areas of damage and ischemia are reduced.

In accordance with the depth of damage to the heart muscle, the following types of myocardial infarction are distinguished: transmural (with damage to the entire thickness of the muscle wall of the heart), subepicardial (with myocardial damage in the area adjacent to the epicardium), subendocardial (with myocardial damage in the area adjacent to the endocardium) and intramural ( with a lesion in the thickness of the myocardium).

Due to the fact that the zones of heart attack and scar tissue are not excited, there is a change in the recorded electrocardiogram, which becomes different from normal. The figure 3 shows examples of changes in ECS with ischemia and various stages of MI [16].

The figure 4 in the table shows the simplified forms of the characteristic graphs of EX for three types of MI in the depth of the lesion (transmural, subepicardial and subendocardial) and four stages of MI (acute, acute, subacute and cicatricial) [10, 12-20]. In total, 11 characteristic graphs of EX with direct signs of MI and 6 characteristic graphs of EX with reciprocal signs of MI, which correspond to 9 different EX with direct signs of MI, can be distinguished. This is due to the fact that the signs of intramural MI coincide in many respects with the signs of subendocardial or subepicardial MI, and it is poorly diagnosed. Direct signs of subendocardial myocardial infarction and reciprocal signs of all types of myocardial infarction according to the depth of the lesion for the cicatricial stage are also poorly defined, and reciprocal signs of transmural and subepicardial myocardial infarction are poorly distinguished.

According to localization (topography) and depending on the damage to the branches of the coronary arteries, myocardial infarction happens: right ventricular and left ventricular. In coronary heart disease, the myocardium of the left ventricle is primarily affected.

The left ventricle (LV) can be divided into areas (segments): septal (septal), anteroposterior (apical), lateral (lateral), posterior and lower (diaphragmatic). The first 3 areas make up the front wall, and the last 3 - the back wall. The lateral region, thus, can be involved both in infarction of the anterior wall and infarction of the posterior wall.

The localization of MI can be determined by the deviations in the ECS of the corresponding leads (hereinafter - according to the leads). In each lead, its own projection of the EMF of the heart is recorded, while 3 standard (I, II, III) and 3 reinforced (aVR, aVF, aVL) leads from the limbs reflect the EMF of the heart in the so-called frontal plane, and the chest leads (V1, V2, ...) reflect the EMF of the heart in the horizontal (transverse) plane.

Depending on which areas of the left ventricle of the heart are damaged, there are, as a rule, up to 13 diagnoses of MI of the left ventricle, determined by its localization [12-20].

Anterior septal myocardial infarction (IM of the anterior septal region, IM of the anterior part of the interventricular septum), as a rule, cases are caused by blockage of the septal branch of the anterior descending artery. Characteristic changes in EX are observed in leads V1, V2 (sometimes V3).

Anterior apical MI (apical region myocardial infarction, IM of the anterior wall of the left ventricle), as a rule, is caused by blockage of the anterior descending artery (its distal parts), departing from the left coronary artery. Diagnosed by characteristic changes in EX in leads V3 and V4, reciprocal changes can be observed in leads III, aVF.

Anterior septal myocardial infarction with transition to the apex (IM of the anterior septal region and the anterior wall of the left ventricle), as a rule, is caused by blockage of the left anterior descending artery. Characteristic changes in EX are observed in leads V1-V4, reciprocal changes can be observed in leads III, aVF.

Lateral MI (IM of the lateral region, IM of the lateral wall of the left ventricle) is usually caused by damage to the diagonal artery or posterolateral branches of the left envelope of the artery. The signs of such an MI are determined by the change in the ECS in leads V5, V6, I, aVL, reciprocal changes can be observed in leads V1, V2.

Anterolateral myocardial infarction (IM of the anterolateral wall of the left ventricle) is usually caused by damage to the envelope artery or the anterior descending artery that extend from the left coronary artery. Characteristic changes in EX are observed in leads V3-V6, I, aVL, reciprocal changes can be observed in leads III, aVF, V1 and V2.

High anterolateral myocardial infarction (lateral basal myocardial infarction), as a rule, is associated with damage to the diagonal artery or branch of the left envelope of the artery. Diagnosed by changes in EX in leads aVL and I. Sometimes reciprocal changes can be observed in leads V1, V2.

The common anterior MI (extensive MI of the anterior wall of the left ventricle) is due to blockage of the main trunk of the left coronary artery (most often its branch, the anterior descending artery). Signs of such an MI are recorded in leads V1-V6, I, aVL. Reciprocal changes can be observed in leads III, aVF.

Post-basal myocardial infarction (MI of the upper parts of the posterior wall of the left ventricle, posterior MI) is usually due to blockage of the right posterior descending artery or the left envelope of the artery. Such an MI is very difficult to diagnose and detect, because There are no direct signs of a heart attack in 12 normal leads. The diagnosis of posterior basal myocardial infarction is made by reciprocal changes in EX in leads V1-V3. Direct signs of posterior basal MI are detected in additional leads V7-V9.

Posterior diaphragmatic myocardial infarction (lower myocardial infarction, myocardial infarction of the lower parts of the posterior wall adjacent to the diaphragm) is usually caused by blockage of the descending branch of the right coronary artery, which, with the left type of blood supply, can depart from the left envelope of the artery. With this localization of the heart attack, the lower parts of the posterior wall are affected, often affecting the posterior part of the interventricular septum. Characteristic changes in EX are observed in leads II, III, aVF. Reciprocal changes in the acute stage of MI are observed in leads V1-V3.

The posterior common MI (extensive MI of the posterior wall of the left ventricle) develops as a result of blockage of the right coronary artery, which is localized proximal to the site of discharge of both arteries of the atrioventricular and sinus nodes. An extensive myocardial infarction captures both the lower and upper parts of the posterior wall of the left ventricle, which causes the signs of anterobasal and posterior phrenic myocardial infarction to be recorded on ECS: direct changes in leads II, III, aVF, V5, V6, reciprocal changes in leads V1-V3.

Posterolateral myocardial infarction (IM of the posterolateral wall of the left ventricle) is caused by clogging of the envelope branch of the left coronary artery. With such a localization of a heart attack, the posterior and lateral walls of the left ventricle are simultaneously affected. Characteristic changes in EX are observed in leads II, III, aVF (damage to the diaphragmatic parts of the posterior wall), V5, V6, I, aVL (damage to the lateral wall), reciprocal changes in EX - in leads V1-V3 (with damage to the basal parts of the posterior wall).

Posterior MI is usually caused by damage to the right coronary artery. With this localization of the heart attack, the interventricular septum and the posterior wall of the left ventricle are affected. Direct signs of a heart attack are recorded in leads II, III, aVF, V1 and V2.

Circular MI (circular apical MI) is caused by damage to the envelope artery. With such a localization, a heart attack in a semicircle covers the apex of the heart with simultaneous damage to its front and back parts. In this case, two variants of damage to the apex of the heart are possible: by moving from the back wall of the left ventricle through the apex to the side and front walls, and also a heart attack can cover the lower parts of the top of the left ventricle, spreading from the back wall to its front wall. Therefore, for a circular apical MI, changes in EX are characteristic in two groups of leads:

1) straight - in leads II, III, aVF, occasionally reciprocal - in leads V1, V2 (with damage to the back of the apex of the heart) and straight lines - in leads V3-V6, I, aVL, (with damage to the anterior and lateral walls of the apex of the heart );

2) straight - in leads II, III, aVF (with lesions of the posterior lower parts of the apex of the heart), in leads V3, V4 (with lesions of the anterior part of the apex of the heart), occasionally reciprocal - in leads V1, V2 (lesions of the basal sections of the posterior wall).

Isolated right ventricular MI is extremely rare. As a rule, it is combined with damage to the posterior wall of the left ventricle. Right ventricular myocardial infarction is difficult to diagnose for ECS, since the excitation vector of the right ventricle is normally much smaller than the excitation vector of the left ventricle and the changes in the vector caused by right ventricular MI have little effect on the total emf and have little effect on the QRS complex. Characteristic changes in EX can be observed in the additional right chest leads V3R and V4R.

Figure 5 in the table shows the direct (P) and reciprocal (P) signs of MI for the aforementioned left ventricular myocardial infarction according to generally accepted leads [12-19], with the additional symbol “-” denotes the signs of MI that cannot be diagnosed take into account (hereinafter - secondary signs of MI).

Also, when diagnosing, it is of interest to determine the localization of early (old) myocardial infarction. According to [17] if the new IM:

- located far from the old scar field or on its periphery, the old heart attack usually does not interfere with the diagnosis of the new one;

- develops in the area of old cicatricial changes, then its diagnosis due to minor changes in ECS is not always possible;

 - occurs on the wall opposite to cicatricial changes, then it is diagnosed, as a rule, only if it is larger in size of the old cicatricial field.

Thus, for the formation of a full-fledged (extended) diagnostic conclusion about myocardial infarction (if any), information is needed not only about the localization of myocardial infarction, but also about its stage and depth of myocardial damage, while separating the new (repeated) myocardial infarction from the old one, i.e. comprehensive information about MI is needed.

The known method of processing and analysis of ECS based on neural network analysis [3] in the case of training of NS to identify signs of MI in the characteristic graphs of ECS with direct and reciprocal signs of MI, shown in table 1, will allow for the diagnosis of MI to determine its stage and type according to the depth of the lesion. However:

1. In the known method of processing and analysis of ECS, the construction of decision rules is based on a combination of output signals of NS trained to identify signs of MI in the ECS of all leads at once, therefore, diagnosis is possible only in one state, since the output of a result relating to one of k conditions of the heart , is carried out based on the choice of the k-th state of the heart, for which the maximum number of signs has been revealed, which does not allow diagnosing myocardial infarction of different stages according to localizations (for example, at the same time Mr. peredneverhushechnogo manifestation of the acute phase of MI and MI lateral subacute stage).

2. The decisive rules of the known method of processing and analysis of ECS do not include the possibility of non-manifestation of reciprocal signs of MI [17], which, if there are direct signs of MI and the absence of their reciprocal signs of MI or their unrecognition, will lead to incorrect diagnosis of the patient’s heart condition, that is, non-detection of myocardial infarction (for example, anteroposterior myocardial infarction with non-manifestation of reciprocal signs in leads III, aVF - see table 2, shown in figure 5).

3. The number of neural networks required for diagnosing various MI in the known method depends on the total number of possible signs of MI in the ECS of leads in all localizations. So, if to diagnose MI taking into account its localization, stage and type according to the known scheme [3], then A neural networks will be required, and the value of A is determined by the formula

A = a ₁ * b ₁ + a ₂ * b ₂ '+ L, (19)

where a ₁ , a ₂ - the number of respectively direct and reciprocal signs of MI according to localization,

b ₁ - the number of different EX with direct signs of MI,

b ₂ '- the number of different EX with reciprocal signs of MI, which are compared EX with direct signs,

L is the number of ECS of a healthy patient, equal to the total number of leads.

In this case, for diagnosing MI for 14 localizations according to 63 direct and 30 reciprocal signs of MI (see table 2), taking into account 11 different ECS with direct signs of MI and 9 EX with reciprocal signs of MI, which correspond to EX with direct signs of MI (see. table 1) when using 12 standard leads 975 neural networks will be required.

Obviously, for the formation of an extended diagnostic conclusion about myocardial infarction based on the neural network analysis of ECS, including the output of information about the location of the MI, stage and depth of the lesion, improving the decision rules is required, and to reduce the number of neural networks involved, it is necessary to optimize the structure of the implementation scheme of neural network analysis.

The aim of the invention is the formation of a diagnostic conclusion about the state of the patient’s heart, in which, in the presence of myocardial infarction, its localization, stage of myocardial infarction (acute, acute, subacute or cicatricial) and type of myocardial infarction according to the depth of the lesion (transmural, subepicardial or subendocardial) are indicated.

For this, a neural network analysis of the state of the heart is used, consisting in the fact that a continuous electrocardiogram (EX) is recorded in L leads, pre-processed and presented as an n-dimensional vector for each of L leads, and model n-dimensional vectors E _{et of} various EX for each of L leads, neural networks are trained, neural network analysis of n-dimensional vectors of registered ECS E _reg is carried out, neural network outputs are analyzed based on the construction of decision rules and the result tat on the state of the patient’s heart, characterized in that, firstly, the construction of decision rules is as follows:

- carry out the separation of the surface of the left ventricle of the heart into areas for which the identified direct and reciprocal signs of myocardial infarction (MI) correspond to a specific stage and a specific type of MI according to the depth of the lesion, for this:

- determine p = {1,2, .. 6} of the region of the left ventricle of the heart and assign the corresponding lead groups: for the septal region (p = 1) - leads V1 and V2 (in the case of generally accepted leads), for the posterior region (p = 2) - leads V1 and V2 (in the case of generally accepted leads) with reciprocal signs of MI, for the anterior region (p = 3) - leads V3 and V4 (in the case of generally accepted leads), for lateral lower region (p = 4) - leads V5 and V6 (in the case of conventional leads), for the lateral upper region (p = 5) - leads I and aVL (in the case of conventional leads), for the lower region (p = 5) - leads II, III and aVF (in the case of generally accepted leads);

- take into account that the MI stage within each p = {1,2, .. 6} of the region of the left ventricle of the heart does not change;

- carry out the determination of s = {1,2, .. 4} stage MI on p = {1,2, .. 6} areas of the heart, for this:

- choose the outputs of neural networks trained to identify in the ECS direct and reciprocal signs of MI of a specific stage s = {1,2, .. 4} (acute, acute, subacute, cicatricial);

- form a matrix of ST stages of MI in the regions of the heart, by logical transformation of the signals of the selected outputs of neural networks for s = {1,2, .. 4} of the MI stage within the group of leads corresponding to one of p = {1,2, .. 6 } areas of the heart, while the values of the elements of the matrix ST (p, s) are found as follows:

1) for the septal area (p = 1)

,

,

2) for the posterior region (p = 2)

,

,

3) for the anterior region (p = 3)

,

,

4) for the lateral lower region (p = 4)

,

,

5) for the lateral upper region (p = 5)

,

,

6) for the lower region (p = 6)

,

,

where s = 1 - for the acute stage of MI, s = 2 - for the acute stage of MI, s = 3 - for the subacute stage of MI, s = 4 - for the cicatricial stage of MI;

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF}, j={2,3..18} соответствует ЭКС различных стадий ИМ (острейших, острых, подострых, рубцовых), видов ИМ по глубине поражения (трансмуральноых, субэпикардиальных, субэндокардиальных) и признаков ИМ (прямых, реципрокных);g _{i, j} is the output signal of the j-th neural network of the i-th lead, while i∈L and for conventional leads i

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF}, j = {2,3..18} corresponds to the ECS of various stages of MI (acute, acute, subacute, cicatricial), species MI according to the depth of the lesion (transmural, subepicardial, subendocardial) and signs of MI (direct, reciprocal);

;

;

- they analyze the values of the elements of the matrix of the MI stage in the heart regions ST (p, s) based on the decisive rule: if the matrix element ST (p, s) = 1, then myocardial infarction of the s-th stage in the p-th region of the heart is determined;

- carry out the determination of v = {1,2,3} type of MI according to the depth of the lesion by p = {1,2, .. 6} areas of the heart, for this:

- share the outputs of neural networks trained to identify in ECS direct and reciprocal signs of MI;

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF} отведениях;- select the outputs of neural networks trained to detect in ECS direct signs of MI of a specific type according to the depth of damage v = {1,2,3} (transmural, subepicardial, subendocardial) in i

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF} leads;

- form a matrix of WP types of MI according to the depth of the lesion with direct signs by logical conversion of the signals of the selected outputs of neural networks, while the values of the elements of the matrix WP (i, v) are found as follows:

1) for the transmural type of MI in the depth of the lesion (v = 1):

,

,

2) for subepicardial MI in the depth of the lesion (v = 2):

,

,

3) for subendocardial MI in the depth of the lesion (v = 3):

,

,

where v = 1 - for transmural myocardial infarction, v = 2 - for subepicardial myocardial infarction, v = 3 - for subendocardial myocardial infarction;

{V1, V2} отведениях;- select the outputs of neural networks trained to identify reciprocal signs of MI of a particular type in the ECS according to the depth of the lesion v = {1,2,3} (transmural, subepicardial, subendocardial) in i

{V1, V2} leads;

- form a matrix WR of types of MI according to the depth of the lesion with reciprocal signs by logical conversion of the signals of the selected outputs of the neural networks, while the values of the elements of the matrix WR (i, v) are found as follows:

1) for the transmural (subepicardial) type of MI according to the depth of the lesion (v = 1, v = 2):

,

,

2) for subendocardial MI in the depth of the lesion (v = 3):

;

;

- form vector lines WF _p of the IM type according to the depth of the lesion for p = {1,2, .. 6} areas of the heart:

- for the septal area (p = 1)

,

,

- for the posterior region (p = 2)

,

,

- for the anterior region (p = 3)

,

,

- for the lateral lower region (p = 4)

,

,

- for the lateral upper region (p = 5)

,

,

- for the lower region (p = 6)

,

,

- perform an analysis of the values of vector strings of the MI type according to the lesion depth WF _p for p = {1,2, ... 6} areas of the heart based on decision rules:

if WF _p = [1 X X], then a transmural MI is determined,

if WF _p = [0 1 X], then a subepicardial MI is determined,

if WF _p = [0 0 1], then a subendocardial MI is determined,

wherein the value of X is 0 or 1;

- carry out the determination of the localization of MI installed stages, for this:

- form the row vector SL _S of the MI localizations by transposing the column vector vectors of the matrix ST for s = {1,2, .. 4} stages:

;

;

- perform the analysis of the values of vector lines SL _{S of the} localizations of the MI for s = {1,2, .. 4} stages based on decision rules:

- if SL _S = [1 0 0 0 0 0], then the anterior septal myocardial infarction of the s-th stage is determined,

- if SL _S = [0 0 1 0 0 0], then the antero-apical myocardial infarction of the s-th stage is determined,

- if SL _S = [1 0 1 0 0 0], then the anterior septal myocardial infarction is determined with the transition to the apex of the s-th stage,

- if SL _S = [1 0 1 1 1 0], then the front common myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 0 1 1 0] or SL _S = [0 0 0 1 1 0], then the lateral myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 1 1 1 0] or SL _S = [0 0 1 1 1 0], then the anterolateral myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 0 0 1 0] or SL _S = [0 0 0 0 1 0], then the lateral basal myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 0 0 0 0], then the posterior basal myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 0 0 0 1] or SL _S = [0 0 0 0 0 1], then the posterior diaphragmatic myocardial infarction of the s-th stage is determined,

- if SL _S = [1 0 0 0 0 1], then the posterior septal myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 0 1 0 1], then the posterior common myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 0 1 1 1] or SL _S = [0 0 0 1 1 1], then the posterolateral myocardial infarction of the s-th stage is determined,

- if SL _S = [0 1 1 1 1 1] or SL _S = [0 0 1 1 1 1] or SL _S = [0 1 1 0 0 1] or SL _S = [0 0 1 0 0 1], then the circular myocardial infarction of the s-th stage is determined;

secondly, the conclusion of the result of a neural network analysis of the state of the heart is as follows:

- determine the parameters of the state of the heart D1-D3:

,

,

,

,

,

,

where g _{i, 1} is the output signal j = 1 of the neural network of the i-th lead trained for the analysis of ECS for compliance with a healthy state of the heart;

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF, aVR};i

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF, aVR};

ST (p, s) - elements of the matrix of stages of MI in the regions of the heart, p = {1,2, .. 6}, s = {1,2, .. 4};

- form the conclusion of the result about the general condition of the patient’s heart, while:

- if D ₁ = 1, then output the result "Within normal limits",

- if D ₂ = 1, then output the result "Suspicion of myocardial infarction",

- if D ₃ = 1, then output the result "Deviation from the norm";

- form the conclusion of the results on the state of the patient’s heart in the presence of signs of MI (D ₂ = 1) obtained using the decision rules, while:

- indicate the name of the MI, determined by the localization (anteroposterior, anteroposterior, anterior septal with a transition to the apex, lateral, anterolateral, lateral basal, anterior common, anterobasal, posterior diaphragmatic, anterior septal, posterior common, anterolateral, circular) for a specific stage (acute, acute cicatricial);

- list the affected MI regions (septal, anterior, lateral superior, lateral inferior, inferior, posterior) and the corresponding MI types according to the lesion depth (transmural, subepicardial, subendocardial).

According to the authors of the proposed invention, as a result of the implementation of the proposed method of neural network analysis of the state of the heart for automated electrocardiographic examination of the patient during screening or in emergency and emergency care when diagnosing myocardial infarction, the doctor is provided with comprehensive information on the localization, stages of myocardial infarction and the depth of damage to the affected areas of the heart.

The figure 6 shows the algorithm of the proposed method of neural network analysis of the state of the heart.

The figure 7 shows the structural diagram of the implementation of the proposed method of neural network analysis of the state of the heart.

Figure 8 in the table shows the indices of neural networks of the i-th assignment and the corresponding EX with signs of MI, the n-dimensional vectors of which are trained data NS.

The figure 9 conditionally shows the position of the heart in space and shows the direction of conventional leads in relation to the areas of the heart.

The figure 10 is a diagram showing the selection of the outputs of neural networks trained to identify signs of a specific stage in the ECS for the purpose of further logical conversion of the values of the output signals of the NS according to the decision rules for determining the stage of the MI.

The figure 11 is a diagram showing the selection of the outputs of neural networks trained to detect signs of MI of a particular type in the ECS according to the depth of the lesion in the case of direct and reciprocal signs of MI within the leads in order to further logically convert the values of the output signals of NS according to the decision rules for determining the type of MI by the depth of the lesion.

The figure 12 in the table shows a comparison of the functionality of the prototype and the present invention.

According to the claims, the proposed method of neural network analysis of the state of the heart consists (see figure 6):

firstly, from well-known actions, such as “Registering EX”, “Preprocessing EX”, “Presenting EX as an n-dimensional vector”, “Forming a training sample”, “Training of neural networks”, “Neural network analysis”;

secondly, from the introduced actions, such as “Building decision rules for determining the areas of the heart and their corresponding leads”, “Building decision rules for determining the stage of MI in the areas of the heart”, “Building decision rules for determining the type of MI in the depth of the lesion by regions heart ";

thirdly, from altered actions, such as “Building decision rules for determining the localization of MI of established stages” (analogous to the action “Building decision rules for k heart conditions”), “Analysis of the outputs of neural networks”, “Output of the result”.

The description of the known actions “Registration of EX”, “Pre-processing of EX”, “Presentation of EX as an n-dimensional vector”, “Formation of a training sample”, “Training of neural networks”, “Neural network analysis” is given above, but at the same time:

- within the framework of the “Formation of a training sample” action, n-dimensional ECS vectors are formed with direct and reciprocal signs, by which it is possible to diagnose MI of different types (transmural, subepicardial, subendocardial) and stages (acute, acute, subacute, cicatricial). The number of n-dimensional vectors of ECS is determined by the required number of neural networks from the leads;

- within the framework of the Neural Network Analysis action, the necessary number of neural networks required for diagnosing various MI is determined by the formula

B = b ₁ * l ₁ + b ₂ * l ₂ + L, (20)

where b ₁ , b ₂ - the number of different EX, respectively, with direct and reciprocal signs of MI,

l ₁ , l ₂ - the number of leads in the ECS which are observed, respectively, direct and reciprocal signs of MI,

L is the number of ECS of a healthy patient, equal to the total number of leads.

In this case, for diagnosing myocardial infarction according to 11 different ECS with direct signs of MI in 11 leads and 6 different ECS with reciprocal signs of MI in 2 necessary leads V ₁ , V ₂ , (see tables 1 and 2) with 12 generally accepted leads, 145 neural networks, namely (see figures 7 and 8):

- 1 NS for each of 12 leads for the analysis of ECS for compliance with a healthy state of the heart (NS _{i, 1} , i = 1, 2..12);

- 11 NS for each of the leads V1 ... V6, I, II, III, aVF, aVL (without aVR) to identify direct signs of MI in ECS (NS _{i, 2} ... NS _{i, 12} , i = 1, 2 .. eleven);

- 6 NS on leads V1 and V2 to identify reciprocal signs of MI in ECS (NS _{i, 13} ... NS _{i, 18} , i = 1, 2).

Let us explain the features of the introduced and changed actions.

Building decision rules for determining areas of the heart and their corresponding leads. It is known that heart attack lesions affect one or another region of the heart (segment), which corresponds to changes in the ECS of certain leads, therefore, each region of the heart (or its part) can be associated with a certain combination of the presence of signs of MI in a certain group of leads [16, 19].

The chest leads "show" the heart in the horizontal plane with a kind of semicircle (see figure 9): leads V1 and V2 - septal (septal) region (interventricular septum) with direct signs of MI or posterior region with reciprocal signs of MI, V3 and V4 - front ( apical) region (anterior apex), V5 and V6 - lateral (lateral) region in its middle and lower parts.

Leads from the extremities “show” the heart in the frontal plane: I and aVL - lateral (lateral) region in its upper part, leads II, III and aVF - lower region. Lead aVR, as a rule, does not reflect changes in LV myocardial infarction.

The construction of decision rules for determining the stage of MI in the regions of the heart. These decision rules include the following actions:

- the choice of the outputs of neural networks trained to identify in the ECS signs of MI of a specific s = {1,2, .. 4} stage (see figure 10);

- the formation of the matrix ST stages of MI in the regions of the heart by logical transformation of the signals of the selected outputs of the neural networks g _{i, j} for s = {1,2, .. 4} of the MI stage within the group of leads corresponding to one of p = {1,2, ..6} areas of the heart;

- analysis of the values of the elements of the matrix ST of the MI stages in the regions of the heart.

The values of the elements of the matrix ST (p, s) are found as follows:

1) for the septal region (p = 1):

- in the acute stage of MI (s = 1):

, (21)

, (21)

- in the acute stage of MI (s = 2):

, (22)

, (22)

- with a subacute stage of MI (s = 3):

, (23)

, (23)

- in the cicatricial stage of MI (s = 4):

, (24)

, (24)

2) for the posterior region or another region with a reciprocal sign (p = 2):

- in the acute stage of MI (s = 1):

, (25)

, (25)

- in the acute stage of MI (s = 2):

, (26)

, (26)

- with a subacute stage of MI (s = 3):

, (27)

, (27)

- in the cicatricial stage of MI (s = 4):

(признак ИМ отсутствует), (28)

(there is no sign of MI), (28)

3) for the anterior region (p = 3):

- in the acute stage of MI (s = 1):

, (29)

, (29)

- in the acute stage of MI (s = 2):

, (30)

, (thirty)

- with a subacute stage of MI (s = 3):

, (31)

, (31)

- in the cicatricial stage of MI (s = 4):

, (32)

, (32)

4) for the lateral lower region (p = 4):

- in the acute stage of MI (s = 1):

, (33)

, (33)

- in the acute stage of MI (s = 2):

, (34)

, (34)

- with a subacute stage of MI (s = 3):

, (35)

, (35)

- in the cicatricial stage of MI (s = 4):

, (36)

, (36)

5) for the lateral upper region (p = 5):

- in the acute stage of MI (s = 1):

, (37)

, (37)

- in the acute stage of MI (s = 2):

, (38)

, (38)

- with a subacute stage of MI (s = 3):

, (39)

, (39)

- in the cicatricial stage of MI (s = 4):

, (40)

, (40)

6) for the lower region (p = 6):

- in the acute stage of MI (s = 1):

, (41)

, (41)

- in the acute stage of MI (s = 2):

, (42)

, (42)

- with a subacute stage of MI (s = 3):

, (43)

, (43)

- in the cicatricial stage of MI (s = 4):

. (44)

. (44)

If the matrix element is ST (p, s) = 1, then the myocardial infarction of the sth stage in the pth region of the heart is determined.

Building decision rules for determining the type of MI according to the depth of the lesion in the areas of the heart. These decision rules include the following actions:

- separation of the outputs of neural networks trained to identify in ECS direct and reciprocal signs of MI;

- the choice of the outputs of neural networks trained to identify in ECM direct signs of a specific MI in the lesion depth v = {1,2,3} (transmural, subepicardial, subendocardial) in i = {1, 2..11} leads (see figure 11);

- the formation of a matrix of WP types of MI in the depth of the lesion with direct signs;

{1,2} отведениях (см. фигуру 11);- selection of the outputs of neural networks trained to identify reciprocal signs of MI of a particular type in the ECS according to the depth of the lesion v = {1,2,3} (transmural, subepicardial, subendocardial) in i

{1,2} leads (see figure 11);

- the formation of the matrix WR types of MI according to the depth of the lesion with reciprocal signs;

- the formation of vector lines WF _p of the IM type according to the depth of the lesion for p = {1,2, .. 6} areas of the heart;

- analysis of the values of vector lines WF _p of the IM type according to the depth of the lesion for p = {1,2, ... 6} areas of the heart.

The values of the elements of the matrix WP (i, v) are found as follows:

1) for the transmural type of MI in the depth of the lesion (v = 1):

, (45)

, (45)

, (46)

, (46)

... ...

, (55)

, (55)

2) for subepicardial MI in the depth of the lesion (v = 2):

, (56)

, (56)

, (57)

, (57)

... ...

, (66)

, (66)

3)) for subendocardial MI in the depth of the lesion (v = 3):

, (67)

, (67)

, (68)

, (68)

... ...

, (77)

, (77)

The values of the elements of the matrix WR (i, v) are found as follows:

1) for the transmural (subepicardial) type of MI according to the depth of the lesion (v = 1, v = 2):

, (78)

, (78)

, (79)

, (79)

Since EX with reciprocal signs of transmural and subepicardial MI practically coincide, in the presence of such EX, the type of MI in the depth of the lesion is diagnosed as transmural;

2) for subendocardial MI in the depth of the lesion (v = 3):

, (80)

, (80)

, (81)

, (81)

The row vectors WF _p of the MI type according to the depth of the lesion for p = {1,2, .. 6} areas of the heart are found by adding the corresponding vector rows of the matrices WP (with direct signs of MI) and WR (with reciprocal signs of MI) for conventional leads within a group corresponding to one of the areas of the heart:

- for the septal area (p = 1):

, (82)

, (82)

- for the posterior region (p = 2):

, (83)

, (83)

- for the anterior region (p = 3):

, (84)

, (84)

- for the lateral lower region (p = 4):

, (85)

, (85)

- for the lateral upper region (p = 5):

, (86)

, (86)

- for the lower region (p = 6):

, (87)

, (87)

The analysis of the values of the vector lines WF _p of the MI type according to the depth of the lesion for p = {1,2, ... 6} areas of the heart is performed on the basis of a given hierarchy of the depth of the MI lesion (transmural, subepicardial, subendocardial):

In this case, the following types of MI are determined by the depth of the lesion:

- transmural MI in the p-th region of the heart, if

WF _p = [1 X X], (88)

- subepicardial MI in the pth region of the heart, if

WF _p = [0 1 X], (89)

- subendocardial MI in the pth region of the heart, if

WF _p = [0 0 1], (90)

wherein the value of the element X is 0 or 1.

The construction of decision rules for determining the localization of the MI of the established stages. These decision rules include the following actions:

- the formation of vector lines SL _{S of the} localizations of MI for s = {1,2, .. 4} stages;

- analysis of the values of vector lines SL _{S of the} localizations of the MI for s = {1,2, .. 4} stages.

The row vectors SL _{S of the} localizations of the MI for the sth stage are formed by transposing the corresponding column vectors of the matrix ST:

- for the acute stage of MI (s = 1):

, (91)

, (91)

- for the acute stage of MI (s = 2):

, (92)

, (92)

- for the subacute stage of MI (s = 3):

, (93)

, (93)

- for the cicatricial stage of MI (s = 4):

, (94)

, (94)

The analysis of the values of the vector rows of SL _{S is} carried out taking into account that secondary signs of MI are redundant, and reciprocal signs of MI in the ECS from leads V1 and V2 may or may not appear. In this case, the following localization of the MI for the established stage is determined:

- anterior septal MI of the s-th stage, if

SL _S = [1 0 0 0 0 0], (95)

- anteroposterior MI of the s-th stage, if

SL _S = [0 0 1 0 0 0], (96)

- anterior septal MI with transition to the top of the s-th stage, if

SL _S = [1 0 1 0 0 0], (97)

- anterior common MI of the s-th stage, if

SL _S = [1 0 1 1 1 0], (98)

- lateral MI of the s-th stage, if

SL _S = [0 1 0 1 1 0] or SL _S = [0 0 0 1 1 0], (99)

- anterolateral myocardial infarction of the s-th stage, if

SL _S = [0 1 1 1 1 0] or SL _S = [0 0 1 1 1 0], (100)

- lateral basal MI of the s-th stage, if

SL _S = [0 1 0 0 1 0] or SL _S = [0 0 0 0 1 0], (101)

- posterior basal MI of the s-th stage, if

SL _S = [0 1 0 0 0 0], (102)

- posterior diaphragmatic MI of the s-th stage, if

SL _S = [0 1 0 0 0 1] or SL _S = [0 0 0 0 0 1], (103)

- posterior MI of the s-th stage, if

SL _S = [1 0 0 0 0 1], (104)

- posterior common MI of the s-th stage, if

SL _S = [0 1 0 1 0 1], (105)

- posterolateral MI of the s-th stage, if

SL _S = [0 1 0 1 1 1] or SL _S = [0 0 0 1 1 1], (106)

- circular MI of the s-th stage, if

SL _S = [0 1 1 1 1 1] or SL _S = [0 0 1 1 1 1] or SL _S = [0 1 1 0 0 1]

or SL _S = [0 0 1 0 0 1], (107)

Analysis of the outputs of neural networks. First, the logical analyzer determines the parameters of the state of the heart D1-D3, on the basis of which a result is formed about the state of the patient’s heart, namely:

1) heart condition parameter “Within normal limits”

, (108)

, (108)

where g _{i, 1} is the output signal j = 1 of the neural network of the i-th lead trained for the analysis of ECS for compliance with a healthy state of the heart;

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF, aVR}.i

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF, aVR}.

2) heart condition parameter “Suspicion of myocardial infarction”

, (109)

, (109)

where ST (p, s) are the elements of the matrix of MI stages in the regions of the heart, p = {1,2, .. 6}, s = {1,2, .. 4};

3) heart condition parameter “Deviation from the norm”

. (110)

. (110)

The patient’s heart condition is within normal limits if the neural network analysis shows that the pacemaker from all 12 leads is within normal limits, i.e. D ₁ = 1.

Suspicion of heart myocardial infarction will occur if a neural network analysis shows that in the group of ECS corresponding to any of the areas of the heart, there are signs of MI, i.e. D ₂ = 1.

The deviation from the normal state of the heart will be determined if the ECS of any of the 12 leads are not within the normal range, but no signs of MI were detected, i.e. D ₃ = 1.

In the case of identifying signs of MI (D ₂ = 1), the logical analyzer determines:

- stage MI in areas (see formulas 21-44);

- types of myocardial infarction according to the depth of lesion by region (see formulas 45-90);

- localization of MI for established stages (see formulas 91-107).

For example, in the following neural networks, the values of the output signals turned out to be different from 0:

g _1,3 = 1 (in assignment V1 defined EX with direct signs of subepicardial MI of the acute stage);

g _2,3 = 1 (in assignment V2 defined EX with direct signs of subepicardial MI of the acute stage);

g _3.4 = 1 (in assignment V3, an EX is determined with direct signs of subendocardial MI of the acute stage);

g _4,4 = 1 (in assignment V4 defined EX with direct signs of subendocardial MI of the acute stage);

g _5.11 = 1 (in assignment V5 defined EX with direct signs of transmural myocardial scar stage);

g _6.11 = 1 (in assignment V6 an EX was determined with direct signs of transmural myocardial infarction);

g _7.11 = 1 (in lead I, an EX is determined with direct signs of transmural myocardial scar stage);

g _11,12 = 1 (in the assignment of aVL, an EX was determined with direct signs of subepicardial myocardial infarction of the cicatricial stage).

Then, based on the neural network analysis, it is determined:

1) lateral myocardial infarction (because SL ₄ = [000110]) in the cicatricial stage (since ST (4,4) = 1, ST (5,4) = 1) and at the same time the type of MI is specified by the depth of the lesion in the regions of the heart: the lateral lower region (since p = 4) - transmural myocardial infarction (because rowWL (5,) = [100], rowWL (6,) = [100] and WF ₄ = [one hundred]); lateral upper region (because p = 5) - transmural myocardial infarction (because rowWL (7,) = [100], rowWL (11,) = [010] and WF ₅ = [110]);

2) anterior septal myocardial infarction with transition to the apex (since SL ₁ = [101000]) in the acute stage (because ST (1,1) = 1, ST (3,1) = 1) and at the same time the type of MI is determined according to the depth of the lesion in the regions of the heart: the septal region (since p = 1) is subepicardial myocardial infarction (because rowWL (1,) = [010], rowWL (2,) = [010] and WF ₁ = [010]); the anterior region (because p = 3) is subendocardial myocardial infarction (because rowWL (3,) = [001], rowWL (4,) = [001] and WF ₃ = [001]).

Conclusion of the result. The extended diagnostic report on the patient’s heart condition consists of two parts:

- conclusion of the conclusion on the general condition of the patient’s heart;

- the conclusion of the results on myocardial infarction in the presence of its signs.

The general conclusion about the patient’s heart condition includes the following entries: “The condition is within normal limits”, “There are deviations from the norm” and “Suspicion of myocardial infarction”.

If there is a suspicion of myocardial infarction, its name is displayed, which is determined by the localization for a particular stage (anteroseptal, anteroposterior, anterior septum with a transition to the apex, lateral, anterolateral, lateral basal, anterior widespread, anterobasal, posterior diaphragmatic, posterior septal, posterior common, posterolateral, also the name of the stage of MI (acute, acute, subacute, cicatricial).

The following is a listing of the affected MI areas with the corresponding type of MI. To display the results, the following LV areas were determined: septal (septal), anterior (apical), lateral superior (lateral basal), lateral inferior (lateral inferior), inferior (diaphragmatic) and posterior. To output a part of the result by type of MI (depth of lesion), the following records are used: transmural myocardial infarction, subendocardial myocardial infarction and subepicardial myocardial infarction.

Examples of a diagnostic conclusion in the presence of signs of MI in leads I, aVL, V1, V2 ... V6 (depending on the different stages and types of MI on the depth of the lesion):

Example 1

Suspicion of an anterior widespread acute myocardial infarction. Affected areas and depth of myocardial infarction:

1) septal region - subepicardial MI;

2) anterior region - subepicardial MI;

3) lateral upper region - transmural MI;

4) the lateral lower region is transmural MI.

Example 2

Suspicion of lateral myocardial infarction in the cicatricial stage. Affected areas and depth of myocardial infarction:

1) lateral upper region - subepicardial MI;

2) the lateral lower region is transmural MI.

Suspicion of anterior septal myocardial infarction with transition to the apex in the acute stage. Affected areas and depth of myocardial infarction:

1) septal region - subendocardial MI;

2) anterior region - subendocardial MI.

In conclusion, it should be noted that this invention is aimed at automating the electrocardiographic diagnosis of the patient’s heart condition and in case of suspicion of myocardial infarction will help the doctor more fully and accurately assess the patient’s heart condition, regardless of their level of qualification and work experience.

Information sources:

1. RF patent No. 2383295. A method of processing an electrocardiogram for the diagnosis of myocardial infarction / Bodin O.N., Zaitseva O.A., Loginov D.S., Moiseev A.E. // Official Bulletin “Inventions. Utility models. ” No. 7, 2010.

2. RF patent No. 2217045. A method of processing an electrocardiogram in dynamics for the diagnosis of myocardial infarction. / Belotserkovsky O.M., Vinogradov A.V., Shebeko S.V., Samgin Yu.S., Vinogradov N.A., Gamolya S.V., Konotop Yu.I. // Official Bulletin “Inventions. Utility Models. ” No. 11, 2003.

3. RF patent No. 2461877. Method of neural network analysis of the state of the heart / Bodin ON, Volkova NA, Loginov DS, Ryabchikov RV, Funtikov VA // Official Bulletin “Inventions. Utility models. ” No. 26, 2012.

4. Kohonen, S., Self-Organizing Maps, Second Edision, Berlin: Springer-Verlag, 1997.

5. Osovsky S. Neural networks for information processing / Per. from Polish I.D. Rudinsky. - M.: Finance and Statistics, 2002.

6. Bagat A.V., Konkina M.M., Simchera V.M. Statistics. - M.: Finance and Statistics, 2005. - 368 p.

7. Bodin O.N., Loginov D.S. Adaptation of the LVQ neural network structure for the analysis of an electrocardiogram // Neurocomputers: development, application, 2008, No. 5-6, P.88-92.

8. Plotnikov A.V., Prilutsky D.A., Selishchev S.V. SCP-ECG standard for digital EX-exchange // Abstracts of reports of the International Conference on Biomedical Instrumentation “Biomedpribor-98” - Moscow, VNIIMP RAMS. 1998, C. 213-215.

9. RF patent No. 2294139. A method for isolating the beginning of a cardiocycle and a device for its implementation / Bodin O.N., Zhulev I.O., Loginov D.S., Mitroshin A.N., Proshkin V.V. // Official Bulletin “Inventions. Utility models. ” No. 6, 2007.

10. Zudbinov Yu.I. ABC EX. Ed. 4th, Spanish and add. - Rostov-on-Don: Phoenix, 2003, - 240 p.

11. Wassermen F. Neurocomputer technology: Theory and practice. - M.: Mir, 1992 .-- 240 p.

12. Roitberg G.E., Strutinsky A.V. Internal illnesses. The cardiovascular system. - M .: Binom-press, 2007 .-- 856 s.

13. Briskin V.R. Electrocardiogram of myocardial infarction. - Vinnitsa: PP "Vidavnitsvo" Thesa ", 2006. - 144 p.

14. Goldberger A.L. Clinical electrocardiography. Visual approach. / A.L. Goldberger trans. from English Yu.V. Furmenkova; under the editorship of A.V. Strutinsky. - M.: GEOTAR-Media, 2009 .-- 328 p.

15. Meshkov A.P. The ABC of clinical electrocardiography. - N. Novgorod: Publisher: NGMA, 1998, - 149 p.

16. Murashko V.V., Strutinsky A.V. Electrocardiography: Textbook. - M.: MEDpress-inform, 2008 .-- 320 p.

17. Orlov V.N. Guidelines for electrocardiography / V.N. Orlov. 9th ed., Rev. - M .: Medical Information Agency LLC, 2017. - 560 p.

18. Syvolap V.D. Clinical electrocardiography. - Zaporozhye: IPO Zaporizhzhia, 2008. - 264 p.

19. Yartsev S.S. Electrocardiography: a practical guide for doctors. - M.: RUDN, 2014 .-- 227 p.

20. Wagner Galen S., Ssrauss David G. Marrioss's pracsical elecsrocardiography. 12sh Revised edision. Lippincoss Williams and Wilkins, 2013 .-- 552 pages.

Claims (90)

The method of neural network analysis of the state of the heart, which consists in the fact that a continuous electrocardiogram (EX) is recorded in L leads, pre-processed and presented as an n- dimensional vector for each of L leads, create model n- dimensional vectors E _et different EX for each L of leads, trained neural networks, neural network analysis performed n -dimension vectors for the pacemaker E _reg perform analysis of neural network outputs based on constructing decision rules and output the result of comprising uu patient's heart, characterized in that, firstly, the construction of decision rules is as follows: - carry out the separation of the surface of the left ventricle of the heart into areas for which the identified direct and reciprocal signs of myocardial infarction (MI) correspond to a specific stage and a specific type of MI according to the depth of the lesion, for this: - determine p = {1,2, .. 6} of the region of the left ventricle of the heart and assign the corresponding lead groups: for the septal region ( p = 1) - leads V1 and V2, for the posterior region ( p = 2) - leads V1 and V2 with reciprocal signs of MI, for the anterior region ( p = 3) - leads V3 and V4, for the lateral lower region ( p = 4) - leads V5 and V6, for the lateral upper region ( p = 5) - leads I and aVL, for the lower region ( p = 5) - leads II, III and aVF; - take into account that the MI stage within each p = {1,2, .. 6} of the region of the left ventricle of the heart does not change; - carry out the determination of s = {1,2, .. 4} stage MI on p = {1,2, .. 6} areas of the heart, for this: - select the outputs of neural networks trained to detect in the ECS direct and reciprocal signs of MI of the most acute ( s = 1), acute ( s = 2), subacute ( s = 3) and cicatricial stages ( s = 4); - form a matrix of ST stages of MI in the regions of the heart, by logical transformation of the signals of the selected outputs of neural networks for s = {1,2, .. 4} of the MI stage within the group of leads corresponding to one of p = {1,2, .. 6 } areas of the heart, while the values of the elements of the matrix ST ( p, s ) are found as follows: 1) for the septal area ( p = 1)

, 2) for the posterior region ( p = 2)

, 3) for the anterior region ( p = 3)

, 4) for the lateral lower region ( p = 4)

, 5) for the lateral upper region ( p = 5)

, 6) for the lower region ( p = 6)

, Wheres= 1 - for the acute stage of MI,s= 2 - for the acute stage of MI,s= 3 - for the subacute stage of MI,s= 4 - for cicatricial stage of myocardial infarction; g _{i, j} is the output signal of the j- th neural network of the i- th lead, with i ∈ L and for the conventional leads i

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF}, j = {2,3..18} corresponds to the ECS of various stages of MI, various types of MI according to the depth of lesion, direct and reciprocal signs of MI;

; - perform an analysis of the values of the elements of the matrix stage MI in the regions of the heartST(p, s) based on the decision rule:if matrix elementST(p, s) = 1,then myocardial infarction is determinedsstage atp1st area of the heart; - carry out the determination of v = {1,2,3} type of MI according to the depth of the lesion by p = {1,2, .. 6} areas of the heart, for this: - share the outputs of neural networks trained to identify in ECS direct and reciprocal signs of MI; - select the outputs of neural networks trained to identify in the ECS direct signs of MI of transmural ( v = 1), subepicardial ( v = 2) and subendocardial ( v = 3) types of MI according to the depth of the lesion in i

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF} leads; - form a matrixWP types of myocardial infarction with direct signs by logical conversion of the signals of the selected outputs of neural networks, while the values of the elements of the matrixWP(i, v) are found as follows: 1) for the transmural type of MI in the depth of the lesion ( v = 1):

, 2) for subepicardial MI in the depth of the lesion ( v = 2):

, 3) for subendocardial MI in the depth of the lesion ( v = 3):

, where v = 1 - for transmural myocardial infarction, v = 2 - for subepicardial myocardial infarction, v = 3 - for subendocardial myocardial infarction; - select the outputs of neural networks trained to detect reciprocal signs of MI in transmural ( v = 1), subepicardial ( v = 2) and subendocardial ( v = 3) types of myocardial infarction according to the depth of the lesion in i

{V1, V2} leads; - form a matrixWr types of myocardial infarction with reciprocal features by logical transformation of the signals of the selected outputs of neural networks, while the values of the elements of the matrixWr(i, v) are found as follows: 1) for a transmural or subepicardial type of MI according to the depth of the lesion ( v = 1, v = 2):

, 2) for subendocardial MI in the depth of the lesion ( v = 3):

; - form vector lines WF _p of the IM type according to the depth of the lesion for p = {1,2, .. 6} areas of the heart: - for the septal area ( p = 1)

, - for the posterior region ( p = 2)

, - for the anterior region ( p = 3)

, - for the lateral lower region ( p = 4)

, - for the lateral upper region ( p = 5)

, - for the lower region ( p = 6)

, - perform an analysis of the values of vector strings of the MI type according to the lesion depth WF _p for p = {1,2, ... 6} areas of the heart based on decision rules: if WF _p = [1 X X], then a transmural MI is determined, if WF _p = [0 1 X], then a subepicardial MI is determined, if WF _p = [0 0 1], then a subendocardial MI is determined, wherein the value of X is 0 or 1; - carry out the determination of the localization of MI installed stages, for this: - form the row vector SL _S of the MI localizations by transposing the column vector vectors of the matrix ST for s = {1,2, .. 4} stages:

; - perform the analysis of the values of the vector lines SL _{S of the} localizations of the MI for s = {1,2, .. 4} stages based on the decision rules: - if SL _S = [1 0 0 0 0 0], it is determined peredneperegorodochny myocardial infarction s -th stage, - if SL _S = [0 0 1 0 0 0], it is determined peredneverhushechny myocardial infarction s -th stage, - if SL _S = [1 0 1 0 0 0], then the anterior septal myocardial infarction is determined with the transition to the top of the s- th stage, - if SL _S = [1 0 1 1 1 0], define a common front myocardial infarction s -th stage, - if SL _S = [0 1 0 1 1 0] or SL _S = [0 0 0 1 1 0], then the lateral myocardial infarction of the s- th stage is determined, - if SL _S = [0 1 1 1 1 0] or SL _S = [0 0 1 1 1 0], then it is determined anterolateral myocardial infarction s -th stage, - if SL _S = [0 1 0 0 1 0] or SL _S = [0 0 0 0 1 0], then the lateral basal myocardial infarction of the s- th stage is determined, - if SL _S = [0 1 0 0 0 0], it is determined zadnebazalny myocardial infarction s -th stage, - if SL _S = [0 1 0 0 0 1] and SL _S = [0 0 0 0 0 1], it is determined zadnediafragmalny myocardial infarction s -th stage, - if SL _S = [1 0 0 0 0 1], it is determined zadneperegorodochny myocardial infarction s -th stage, - if SL _S = [0 1 0 1 0 1], then the posterior common myocardial infarction of the s- th stage is determined, - if SL _S = [0 1 0 1 1 1] or SL _S = [0 0 0 1 1 1], then the posterolateral s- stage myocardial infarction is determined, - if SL _S = [0 1 1 1 1 1] or SL _S = [0 0 1 1 1 1] or SL _S = [0 1 1 0 0 1] or SL _S = [0 0 1 0 0 1], then s- stage circular myocardial infarction is determined; secondly, the conclusion of the result of a neural network analysis of the state of the heart is as follows: - determine the parameters of the state of the heart D 1- D 3:

,

,

, where g _{i, 1} is the output signal j = 1 of the neural network of the i- th lead trained for the analysis of ECS for compliance with a healthy heart condition; i

{V1, V2, V3, V4, V5, V6, I, II, III, aVL, aVF, aVR}; ST ( p, s ) - elements of the matrix of stages of MI in the regions of the heart, p = {1,2, .. 6}, s = {1,2, .. 4}; - form the conclusion of the result about the general condition of the patient’s heart, while: - if D ₁ = 1, then output the result "Within normal limits", - if D ₂ = 1, then output the result "Suspicion of myocardial infarction", - if D ₃ = 1, then output the result "Deviation from the norm"; - form the conclusion of the results on the state of the patient’s heart in the presence of signs of MI ( D ₂ = 1) obtained using the decision rules, while: - indicate the name of the MI, determined by the location for a particular stage; - list the affected IM areas of the heart and the corresponding types of MI according to the depth of the lesion.

Images