CN104757992A - Cardiac sound diagnostic system based on depth confidence network and diagnostic method - Google Patents

Abstract

The invention discloses a cardiac sound diagnostic system based on a depth confidence network and a diagnostic method of the cardiac sound diagnostic system. The cardiac sound diagnostic system comprises a cardiac sound diagnostic terminal, a network interface module, a database module, a data analyzing module, and a data management module; the data management module is connected with the database module, the data analyzing module and the network interface module at the same time, and the database is connected with the data analyzing module at the same time; the network interface module is in signaling connection with the cardiac sound diagnostic terminal. The diagnostic method is based on a depth confidence network model consisting of restricted Boltzmann machine layering; the diagnostic method comprises the steps: using a patient's cardiac sound file database; training the established depth confidence network by using a layer-by-layer greedy algorithm; inputting the cardiac sound signal to be diagnosed to the depth confidence network model after training; obtaining the final diagnostic result at an output layer and returning to the cardiac sound diagnostic terminal. The cardiac sound diagnostic system based on the depth confidence network and the diagnostic method can realize the remote diagnosis of patient's cardiac sound signals; the operation is convenient and simple, the diagnostic accuracy is high, the cost is low; besides, the device is convenient to maintain and upgrade.

Description

A kind of hear sounds diagnostic system based on degree of depth confidence network and diagnostic method thereof

Technical field

The present invention relates to the hear sounds diagnostic techniques of medical instruments field, be specifically related to the hear sounds diagnostic system based on degree of depth confidence network and diagnostic method thereof, belong to the innovative technology of hear sounds diagnostic system based on degree of depth confidence network and diagnostic method thereof.

Background technology

Cardiechema signals is one of physiological signal of wanting of body weight for humans, has contained the original pathological information of a large amount of heart of human body, and therefore hear sounds diagnosis has irreplaceable advantage and clinical value.By the analysis to hear sounds, medically usually the diagnosis of hear sounds can be roughly divided into normal cardiac sound, exception, extra-heart sounds, cardiac murmur, pericardial friction rub 5 kinds of situations occur hear sounds, in the present invention 5 of hear sounds kinds of diagnostic categories are divided into following 14 kinds in detail:

1. normal cardiac sound

2. hear sounds occurs abnormal: the change of hear sounds comprises intensity of heart sounds (strengthen or weaken), the change of character and splitting of heart sounds.

1) intensity of heart sounds changes: caused by diseases such as emphysema, hydrothorax, pericardial effusions.

2) hear sounds character changes: caused by diseases such as large area acute myocardial infarction, severe myocarditis.

3) splitting of heart sounds: by pathology such as valve of pulmonary trunk or aortic stenosis, mitral stenosis, severe hypertensions.

3. extra-heart sounds

1) early stage relaxing period extra-heart sounds: by Left ventricular dysfunction, the diseases such as relaxing period volume load is overweight, myocardial function serious hindrance cause.

2) late period relaxing period extra-heart sounds: cause the heart disease of ventricular hypertrophy to cause, as hypertensive heart disease, hypertrophic cardiomyopathy, aortic stenosis, pulmonary stenosis etc. by overload.

3) artificial valve's sound: by the artificial mechanical valve prosthesis inserted, when opening and close, valve clashes into caused by metal rack.

4. cardiac murmur

1) blood flow accelerates: the blood flow caused by heating, severe anemia, hyperthyroidism accelerates the cardiac murmur caused.

2) narrow: valve orifice narrow (as mitral stenosis) or trunk have stenosis (as Congenital Coarctation of Aorta etc.), or expand the valve orifice relative narrowness produced due to cardiac dilatation or trunk (pulmonary artery or aorta), blood flow by time produce whirlpool and occur noise.

3) valvular insufficiency: valvular insufficiency (as aortic incompetence), or form relative incompetence because trunk or cardiac dilatation make valve orifice expand, blood reflux forms whirlpool, produces noise.

4) abnormal blood flow passage: have abnormal path in heart or between trunk, as ventricular septal defect, patent ductus arteriosus, arteriovenous fistula etc.

5) drift or anomalous structure in the chambers of the heart: in ventricle, the stump of Jiaxuan's Ci or papillary muscles, rupture of chordae tendineae swings, floats in the chambers of the heart, and blood flow is disturbed and produce whirlpool, occurs noise.

6) aneurysm: arterial wall, due to pathological changes or the expansion of wound generation limitation, forms aneurysm.Blood flow, when normal arterial lumen flows through the position of expansion, can produce whirlpool and cause noise.

5. pericardial friction rub: be generally caused by infectious pericarditis (Tuberculous, suppurative etc.), also be due to non-infectious pericarditis sometimes, as after Uremia, tumprigenicity, traumatic, radiation injury, rheumatism and heart and injury, the disease such as syndrome causes.

Traditional hear sounds diagnosis is realized by stethoscope, and stethoscope can provide noninvasive abundant information for the diagnosis of heart disease, is that current medical personnel use the most convenient, frequently and widely one of medical apparatus and instruments.But conventional stethoscope cannot preserve acoustical signal, auscultation process easily by the impact of auscultator's subjective experience, has obvious personalized difference.Therefore there is the deficiency that subjectivity is strong, precision is low in conventional stethoscope.

In recent years, external high-end electronic stethoscope rises gradually, and they can high-fidelity collection, stores and shares auscultative signal with long-range, make auscultation more accurate, objective.But these stethoscopes cannot solve this key issue of pathological characters needing the time of medical practitioner at substantial to differentiate the complexity of hear sounds, auxiliary diagnosis can not be carried out by intelligent extraction pathological characters.And this stethoscope is due to technical monopoly, expensive, cannot extensive use, and upgrading and difficult in maintenance.

Summary of the invention

The object of the invention is to the shortcoming and defect overcoming above-mentioned prior art, a kind of hear sounds diagnostic system based on degree of depth confidence network is provided.The present invention can improve accuracy rate of diagnosis, overcomes the professional dependence to doctor.

Another object of the present invention is to provide a kind of diagnostic method based on degree of depth confidence network.The operation of the easy patient of the present invention, reduces diagnosis cost.

The technical solution used in the present invention is: the hear sounds diagnostic system that the present invention is based on degree of depth confidence network, includes hear sounds diagnosis terminal, Network Interface Module, DBM, data analysis module, data management module; Described data management module is connected with DBM, data analysis module, Network Interface Module simultaneously, and data base is connected with data analysis module simultaneously; Described Network Interface Module is connected with hear sounds diagnosis terminal signal.

The present invention is based on the diagnostic method of the hear sounds diagnostic system of degree of depth confidence network, comprise the following steps:

1), after system start-up, data management module is responsible for dispatching and is completed DBM, data analysis module, the self-inspection of Network Interface Module and relevant initial work;

2) data management module is by Network Interface Module broadcast system initiation message, and receives the response message of hear sounds diagnosis terminal;

3) system starts to receive the cardiechema signals to be diagnosed that hear sounds diagnosis terminal is uploaded, and this signal is sent to data management module by network interface, generates and treats diagnostic signal data base;

4) data analysis module first calls patient's hear sounds archive database from DBM, carries out pretreatment by pretreatment module to the cardiechema signals in patient's hear sounds archive database; Then construct the degree of depth confidence network based on limited Boltzmann machine layering composition, and the number of plies and the nodes of network are set; Then be input in the degree of depth confidence network formed based on limited Boltzmann machine layering by completing pretreated cardiechema signals, successively greedy algorithm training is adopted to obtain a preferably network initial weight, then Softmax grader is added at the top layer of degree of depth confidence network, by using back-propagation algorithm to finely tune network with label data, thus train optimum network model;

5) system is called and is treated diagnostic signal data base from DBM, takes out and treats diagnostic signal, first treat diagnostic signal by pretreatment module and carry out pretreatment; Then pretreatedly treating that diagnostic signal is input to step 4 and trains in the degree of depth confidence network model obtained by completing, exporting diagnostic result from the grader of the top layer of network model;

6) diagnostic result is sent to hear sounds diagnosis terminal by Network Interface Module by data management module, simultaneously also by diagnostic result stored in diagnostic result filing database, if credible result, this diagnostic result and corresponding patient's cardiechema signals are updated to patient's hear sounds archive database simultaneously.

Hear sounds diagnostic system based on degree of depth confidence network of the present invention utilizes limited Boltzmann machine layer representation method, simulate the cognitive process of brain multilayer neural network, can according to the heart sound data automatic learning pathological characters of a large amount of patients and classification, compared with prior art, tool has the following advantages in the present invention:

1) the present invention achieves long-range hear sounds automatic diagnosis by the Internet Transmission of data, and patient no matter when and where can carry out collection and the diagnosis of hear sounds, can improve auscultation efficiency, the operation of easy patient, reduces diagnosis cost.

2) with corresponding, the believable diagnostic result of diagnosis can be treated that diagnostic signal adds patient's hear sounds archive database by the present invention, thus reduces the cost setting up heart sound data storehouse.

3) diagnostic method of the present invention is the degree of depth confidence network model based on limited Boltzmann machine layering composition, call patient's hear sounds archive database, successively greedy algorithm is adopted to train the degree of depth confidence network established, and add Softmax grader at the top layer of degree of depth confidence network, and label data is utilized to use back-propagation algorithm to finely tune network.Cardiechema signals to be diagnosed is input in the degree of depth confidence network model of training, just can have obtained final diagnostic result at output layer, and returned to hear sounds diagnosis terminal.The present invention can realize the remote diagnosis of the cardiechema signals to patient, simply easy to operate, and accuracy rate of diagnosis is high, and cost is low, is convenient to safeguard and upgrading.

Accompanying drawing explanation

Fig. 1 is systematic schematic diagram of the present invention;

Fig. 2 is the flow chart of the hear sounds diagnosis that the present invention is based on degree of depth confidence network;

Fig. 3 is degree of depth confidence network diagram in the present invention;

Fig. 4 is the structural representation of limited Boltzmann machine in the present invention.

Detailed description of the invention

In order to make technical scheme of the present invention and advantage more clear understand, below in conjunction with accompanying drawing, technical scheme of the present invention is described in detail.

Figure 1 shows that the schematic diagram of present system, the present invention is based on the hear sounds diagnostic system of degree of depth confidence network, comprise hear sounds diagnosis terminal, Network Interface Module, DBM, data analysis module, data management module; Described data management module is connected with DBM, data analysis module, Network Interface Module simultaneously, and data base is connected with data analysis module simultaneously.Described Network Interface Module is connected with fault diagnosis terminal signaling.Described Network Interface Module, DBM, data analysis module and data management module are integrated in background server.

Described hear sounds diagnosis terminal, is used for receiving and uploading cardiechema signals to be diagnosed, and receives and display diagnostic result;

Described Network Interface Module, is used for receiving the cardiechema signals to be diagnosed of automatic network, and diagnostic result is returned to hear sounds diagnosis terminal;

Described DBM, being used for storage treats diagnostic signal data base, patient's hear sounds archive database and diagnostic result filing database;

Described data analysis module, for running the diagnosis algorithm of hear sounds.

Described data management module, transmits for Systematical control and the data managed between each module.

Figure 2 shows that the flow chart of the hear sounds diagnostic system based on degree of depth confidence network, the work process of native system comprises the steps:

Step 1: after system start-up, data management module is responsible for dispatching and is completed DBM, data analysis module, the self-inspection of Network Interface Module and relevant initial work;

Step 2: data management module by Network Interface Module broadcast system initiation message, and receives the response message of hear sounds diagnosis terminal;

Step 3: system starts to receive the cardiechema signals to be diagnosed that hear sounds diagnosis terminal is uploaded, and this signal is sent to data management module by network interface, generates and treats diagnostic signal data base;

Step 4: data analysis module first calls patient's hear sounds archive database from DBM, carries out pretreatment by pretreatment module to the cardiechema signals in patient's hear sounds archive database; Then construct the degree of depth confidence network based on limited Boltzmann machine layering composition, and the number of plies and the nodes of network are set; Then be input in the degree of depth confidence network formed based on limited Boltzmann machine layering by completing pretreated cardiechema signals, successively greedy algorithm training is adopted to obtain a preferably network initial weight, then Softmax grader is added at the top layer of degree of depth confidence network, by using back-propagation algorithm to finely tune network with label data, thus train optimum network model;

Step 5: system is called and treated diagnostic signal data base from DBM, takes out and treats diagnostic signal, first treat diagnostic signal by pretreatment module and carry out pretreatment; Then pretreatedly treat that diagnostic signal is input in step 4 by completing and complete in the degree of depth confidence network model of training, the probability of 14 kinds of hear sounds diagnostic categories that the grader of the top layer of computing network model exports, gets classification corresponding to maximum probability as diagnostic result;

Step 6: diagnostic result is sent from hear sounds diagnosis terminal by Network Interface Module by system, simultaneously also diagnostic result stored in diagnostic result filing database, if credible result, this diagnostic result and corresponding patient's cardiechema signals are updated to patient's hear sounds archive database simultaneously.

In described step 4, pretreatment module is carried out pretreatment to the cardiechema signals in patient's hear sounds archive database and is comprised 3 flow processs, and detailed process is as follows:

1) for the heart sound data in patient's hear sounds archive database selects identical data length, and medium filtering removal baseline drift and iir digital filter is utilized to carry out the process of 50Hz notch filter;

2) computation of mean values average standardization is done to the average of data;

3) data principal component is analyzed, then whitening processing is carried out to data.

In described step 4, degree of depth confidence network forms based on limited Boltzmann machine layering, is described in detail below by the training of network.As shown in Figure 3,4: we adopt successively greedy algorithm to carry out training network.Namely first utilize the input of pretreated cardiechema signals to carry out the ground floor of training network, obtain the output of the limited Boltzmann machine of ground floor; Then limited for ground floor Boltzmann machine is exported and carry out re-training as the input of the limited Boltzmann machine of the second layer, obtain the model parameter of the second layer; Finally, to the strategy that each layer below adopts equally, the mode that the output by front layer inputs as lower one deck is trained successively.Therefore, entire depth confidence network is from bottom to top superposed by multiple limited Boltzmann machine, and successively training obtains.Limited Boltzmann machine comprise two layers-: visible layer (visible-layer) and hidden layer (hidden-layer), as shown in Figure 4, wherein v is visual layers to the connection between neuron, and h is hidden layer.Because Boltzmann machine is a model based on energy, therefore, for one group of given state (v, h), can be defined as follows energy function:

$\begin{array}{c}{E}_{\mathrm{\θ}}(v,h)=-\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{a}_i{v}_i-\underset{j=1}{\overset{1}{\mathrm{\Σ}}}{b}_j{h}_j-\underset{i=1}{\overset{n_v}{\mathrm{\Σ}}}\underset{j=1}{\overset{n_h}{\mathrm{\Σ}}}{v}_i{w}_{\mathrm{ij}}{h}_j\\ =-v^T\mathrm{Wh}-a^{T}v-b^{T}h\end{array}---\left(1\right)$

Utilize the energy function that formula (1) defines, can to the joint probability distribution of do well (v, h):

$P_{\mathrm{\θ}}(v,h)-\frac{1}{z_{\mathrm{\θ}}}{e}^{-E_{\mathrm{\θ}}(v,h)}---\left(2\right)$

Wherein

$Z_{\mathrm{\θ}}={\mathrm{\Σ}}_{v,h}{e}^{-E_{\mathrm{\θ}}(v,h)}---\left(3\right)$

Wherein n is visual layers unit number, and l is hidden layer unit number, and θ is model parameter, is the weight coefficient between visual layers i and hidden layer unit j; A and b is respectively the biased of the neural unit of correspondence.During certain node layer state given, the status condition between another node is separate, that is:

P _θ(h|v)＝Π _jp(h _j|v) (4)

P _θ(v|h)＝Π _ip(v _i|h) (5)

After the state of known visual layers node, the activation probability of a jth hidden layer node can be expressed as:

P _θ(h _j＝1|v)＝δ(b _j+Σ _iW _ijv _ij) (6)

Wherein δ (.) is sigmoid activation primitive, is defined as sigmoid (x)=1/ (1+e ^-x), in like manner, after trying to achieve all hidden layer nodes, based on the symmetrical structure of Boltzmann machine, the activation probability of i-th visual layers node can be expressed as:

P _θ(v _i＝1|h)＝δ(a _i+Σ _jW _ijh _ij) (7)

Wherein, delta-function is activation primitive.δ(x)＝1/[1+e ^-x]。Hypothesis only has a training sample below, carrys out labelling P respectively with data and model _θand P (h|v) _θ(v|h) these two probability distribution, in order to train limited Boltzmann machine, ask local derviation to each model parameter, can obtain log-likelihood function and be respectively about the partial derivative of the biased a of connection matrix W, visual layers node and the biased b of hidden layer node:

$\frac{\&PartialD;\log P_{\mathrm{\&theta;}}}{\&PartialD;W_{\mathrm{ij}}}={<v_i{h}_j>}_{\mathrm{data}}-{<v_i{h}_j>}_{\mathrm{mode}l}---\left(8\right)$

$\frac{\&PartialD;\log P_{\mathrm{\&theta;}}\left(v\right)}{\&PartialD;a_i}={<v_i>}_{\mathrm{data}}-{<v_i>}_{\mathrm{mode}l}---\left(9\right)$

$\frac{\&PartialD;\log P_{\mathrm{\&theta;}}\left(v\right)}{\&PartialD;b_j}{<h_j>}_{\mathrm{data}}-{<h_j>}_{\mathrm{mode}l}---\left(10\right)$

Because restriction Boltzmann machine meets the condition using Gibbs sampling, the sample data of obeying the distribution of restriction Boltzmann machine finally can be obtained.Therefore can hocket Gibbs sampling in restriction Boltzmann machine, if v ₀for visual layers original state, concrete sampling process is described below:

h ₀～p(h|v ₀) (11)

v ₁～p(v|h ₀) (12)

h ₁～p(h|v ₁) (13)

v ₂～p(v|h ₁) (14)

Wherein, x ~ p (h|v ₁) represent that x is from Probability p (h|v ₁) on the stochastical sampling that obtains.Utilize the fast learning algorithm of sdpecific dispersion algorithm as restriction Boltzmann machine.This algorithm, by pre-training learning data, obtains v ₀after initial value, then only need to carry out Gibbs sampling one to twice, just can complete last probability and be similar to.Given learning data v ₀, calculate all hidden layer node j binary conditions, after hidden layer node is all obtained, determine visible node v conversely _itwo state of value, and then the reconstruct producing visual layers.When using stochastic gradient rise method to maximize log-likelihood function value on the training data, the replacement criteria of each parameter is:

ΔW _ij＝γ(<v _ih _j> _data-<v _ih _j> _recon) (15)

Δa _i＝γ(<v _i> _data-<v _i> _recon) (16)

Δb _j＝γ(<h _j> _data-<h _j> _recon) (17)

Wherein, γ is learning rate, < ~ > _reconrepresent the distribution of a rear model of reconstruct.

Add Softmax grader at the top layer of degree of depth confidence network, and train this layer, obtain the parameter of this layer, process is as follows:

$S_i^c=\mathrm{Soft}\max \left(f_i^c\right)=\frac{e^{f_i^c}}{{\mathrm{\&Sigma;}}_{c=1}^k{e}^{f_i^c}}---\left(18\right)$

Wherein, k is the classification number of hear sounds, and i corresponds to i-th patient in file store,

x is each neuron state of hidden layer, and θ is model parameter.For above-mentioned training method, when each layer parameter of training, other each layer parameter can be fixed and remain unchanged.So, if expect better result, after above-mentioned training process completes, can by calling the label data y of patient's hear sounds archive database, use back-propagation algorithm to adjust the parameter of all layers to improve result, this process is claimed " fine-tuning " simultaneously.The label data y corresponding to i-th patient _irepresent, y _i∈ R ^{14 × 1}, work as y _iduring corresponding to a certain classification in 14 kinds of hear sounds diagnostic categories, y _i=1, otherwise y _i=0; System goal function then based on Softmax grader is:

${\min }_{w,b}J=-\frac{1}{m}\left[{\mathrm{\&Sigma;}}_{i=1}^m{\mathrm{\&Sigma;}}_{j=1}^{k}1\right\{y_i=j\left\}\log \frac{e^{f_i^c}}{{\mathrm{\&Sigma;}}_{j=1}^k{e}^{f_i^c}}\right]---\left(19\right)$

1 (.) is wherein an indicative function, and namely when the value in braces is true time, the result of this function is just 1, otherwise its result is just 0.With gradient descent method, can obtain the partial derivative of object function, differentiate result is as follows:

$\frac{\&PartialD;J_i}{\&PartialD;W_c}=-\frac{1}{m}{\mathrm{\&Sigma;}}_{i=1}^m\left[x_i\left(1\right\{y_i=j\}-p_{\mathrm{\&theta;}}(y_i=j|x_i))\right]---\left(20\right)$

$\frac{\&PartialD;J_i}{\&PartialD;b_c}=-\frac{1}{m}{\mathrm{\&Sigma;}}_{i=1}^m\left[1\right\{y_i=j\}-p_{\mathrm{\&theta;}}(y_i=j|x_i)]---\left(21\right)$

According to the partial derivative obtained, to model parameter θ={ W _c, b _ccarry out fine setting θ={ W _c, b _c; According to this renewal principle, successively forward direction, finely tunes the model parameter of every one deck, to obtain optimum network weight.

In described step 5, the cardiechema signals that pretreatment module treats diagnosis carries out pretreatment and comprises 3 flow processs, and detailed process is as follows:

1) for cardiechema signals to be diagnosed selects identical data length, and medium filtering removal baseline drift and iir digital filter is utilized to carry out the process of 50Hz notch filter;

2) computation of mean values average standardization is done to the average of data;

3) data principal component is analyzed, then whitening processing is carried out to data.

Above-described embodiment is the present invention's preferably embodiment; but embodiments of the present invention are not restricted to the described embodiments; change, the modification done under other any does not deviate from spirit of the present invention and principle, substitute, combine, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (10)

1., based on a hear sounds diagnostic system for degree of depth confidence network, it is characterized in that including hear sounds diagnosis terminal, Network Interface Module, DBM, data analysis module, data management module; Described data management module is connected with DBM, data analysis module, Network Interface Module simultaneously, and data base is connected with data analysis module simultaneously; Described Network Interface Module is connected with hear sounds diagnosis terminal signal.

2. the hear sounds diagnostic system based on degree of depth confidence network according to claim 1, is characterized in that: described hear sounds diagnosis terminal is used for receiving and uploading cardiechema signals to be diagnosed, and receives and display diagnostic result; Described Network Interface Module is used for receiving the cardiechema signals to be diagnosed of automatic network, and diagnostic result is returned to hear sounds diagnosis terminal; Described DBM is used for storage and treats diagnostic signal data base, patient's hear sounds archive database and diagnostic result filing database; Described data analysis module is for running the diagnosis algorithm of hear sounds, and described data management module, transmits for Systematical control and the data managed between each module.

3. the hear sounds diagnostic system based on degree of depth confidence network according to claim 1, is characterized in that: described Network Interface Module, DBM, data analysis module and data management module are integrated in background server; Described Network Interface Module, DBM, data analysis module and data management module run on background server.

4. the hear sounds diagnostic system based on degree of depth confidence network according to claim 2, is characterized in that: described data analysis module adopts background server to run cardiechema signals diagnotor.

5. the hear sounds diagnostic system based on degree of depth confidence network according to claim 1, is characterized in that: described DBM stores patient's hear sounds archive database, treats diagnostic signal data base and diagnostic result filing database.

6. the hear sounds diagnostic system based on degree of depth confidence network according to claim 1, is characterized in that: described Network Interface Module adopts Wifi, 3G or 4G network to be connected with hear sounds diagnosis terminal signal.

7. the hear sounds diagnostic system based on degree of depth confidence network according to claim 1, is characterized in that: described data management module adopts the communication between control sequence control diagnostic system modules and data to transmit.

8. the hear sounds diagnostic system based on degree of depth confidence network according to claim 1, it is characterized in that: described data analysis module is by multiple limited Boltzmann machine layered combination, obtain the degree of depth confidence network of multiple-input and multiple-output, cardiechema signals inputs from bottom, and the top layer of network exports the classification diagnostic result of hear sounds.

9., based on a diagnostic method for the hear sounds diagnostic system of degree of depth confidence network, it is characterized in that comprising the following steps:

1), after system start-up, data management module is responsible for dispatching and is completed DBM, data analysis module, the self-inspection of Network Interface Module and relevant initial work;

2) data management module is by Network Interface Module broadcast system initiation message, and receives the response message of hear sounds diagnosis terminal;

3) system starts to receive the cardiechema signals to be diagnosed that hear sounds diagnosis terminal is uploaded, and this signal is sent to data management module by network interface, generates and treats diagnostic signal data base;

4) data analysis module first calls patient's hear sounds archive database from DBM, carries out pretreatment by pretreatment module to the cardiechema signals in patient's hear sounds archive database; Then construct the degree of depth confidence network based on limited Boltzmann machine layering composition, and the number of plies and the nodes of network are set; Then be input in the degree of depth confidence network formed based on limited Boltzmann machine layering by completing pretreated cardiechema signals, successively greedy algorithm training is adopted to obtain a preferably network initial weight, then Softmax grader is added at the top layer of degree of depth confidence network, by using back-propagation algorithm to finely tune network with label data, thus train optimum network model;

5) system is called and is treated diagnostic signal data base from DBM, takes out and treats diagnostic signal, first treat diagnostic signal by pretreatment module and carry out pretreatment; Then pretreatedly treating that diagnostic signal is input to step 4 and trains in the degree of depth confidence network model obtained by completing, exporting diagnostic result from the grader of the top layer of network model;

6) diagnostic result is sent to hear sounds diagnosis terminal by Network Interface Module by data management module, simultaneously also by diagnostic result stored in diagnostic result filing database, if credible result, this diagnostic result and corresponding patient's cardiechema signals are updated to patient's hear sounds archive database simultaneously.

10. the diagnostic method of the hear sounds diagnostic system based on degree of depth confidence network according to claim 9, it is characterized in that in described step 4), pretreatment module is carried out pretreatment to the cardiechema signals in patient's hear sounds archive database and is comprised 3 flow processs, and detailed process is as follows:

1) for the heart sound data in patient's hear sounds archive database selects identical data length, and medium filtering removal baseline drift and iir digital filter is utilized to carry out the process of 50Hz notch filter;

2) computation of mean values average standardization is done to the average of data;

3) data principal component is analyzed, then whitening processing is carried out to data.