CN104036505A - Heart failure detection method based on combined semantic technology and medical image segmentation - Google Patents

Abstract

The invention discloses a heart failure detection method based on combined semantic technology and medical image segmentation. The method comprises the following steps: segmenting the inner membrane and the outer membrane of a left ventricle magnetic resonance image; calculating a segmentation result so as to convert image information into triple information; storing the triple information into a heart failure body established by protete; and then, automatically detecting the state of a heart by the powerful reasoning function of a semantic technology. According to the heart failure detection method based on the combined semantic technology and medical image segmentation, an input heart magnetic resonance image sequence is subjected to myocardial mass calculation and left ventricular ejection fraction calculation almost without manual intervention, the extraction precision of the inner membrane and the outer membrane of a left ventricle can be greatly improved, wrong segmentation brought by a fact that the extraction of an inner membrane profile and an outer membrane profile is affected by papillary muscles and an artifact can be reduced.

Description

The detection method in heart failure of being combined with medical image segmentation based on semantic technology

Technical field

The present invention relates to digital home technical field, be specifically related to a kind of detection method in heart failure of being combined with medical image segmentation based on semantic technology.

Background technology

Heart failure is one of major disease of generally acknowledged harm humans life and health.According to statistics, approximately there is in the world 1200～1500 patient to suffer from heart failure decocting; Nearly 1,000,000,000 population centers of the America and Europe patient nearly 5% that declines; And U.S.'s patients with heart failure approximately 5,000,000, every year newly-increased 500,000 people; In China, existing patients with heart failure reaches more than 400 ten thousand, and medical expense approaches 10,000,000,000 every year, and this brings serious financial burden to country and society.In addition, although treatment in heart failure obtains development at full speed, but its mortality ratio is still high, and as in China, the admission rate of heart failure accounts for 20% of the angiocardiopathy same period, and mortality ratio lacks up to 40%; The mortality of NYHA III-IV level is up to 30%～40%, and the mankind's life and health in serious threat.In addition according to statistics, more than 50% heart failure patient is all accompanied by the phenomenon of left ventricular hypertrophy, and abnormal Left Ventricular Ejection Fraction and myocardial mass, if can make detection to this from medical image, that brings very large facility will to early diagnosis and therapy in heart failure, also can reduce the input of a large amount of manpower and materials.

Technology related to the present invention mainly contains both: the automatic segmentation of heart left ventricle and semantic medical skill.Wherein, the automatic partition segmentation method of heart left ventricle is mainly the characteristic study such as shape facility and gray feature for heart, wherein relatively the algorithm in forward position have dividing method based on cluster analysis, based on statistical image partition method, dividing method based on graph theory and based on can deformation model and the dividing method of the differential equation etc.

The general meeting of dividing method based on cluster analysis and gauss hybrid models (Gaussian Mixture model, the GMM) combination based on gradation of image feature, be applied to cutting apart of cardiac image.Gauss hybrid models is that the frequency that certain gray-scale value of reflecting by image grey level histogram occurs is come foreground area in differentiate between images and background area, and peak corresponding to different regions.For complicated image, especially medical image is due to the unevenness of its gray scale, so general performance is multimodal.By histogrammic multimodal characteristic being regarded as to the stack of multiple Gaussian distribution, adopt subsequently the method for cluster analysis that these peaks are separated, for example use K-means etc., just can solve the problem of cutting apart of image.

The method that is Statistics Application based on statistical method is carried out mathematical modeling, the gray scale of each pixel in digital picture is regarded as to the stochastic variable with certain probability distribution, thus correct cutting apart image or recover actual object from the image of observing.From the angle of Bayes statistics, obtain exactly the distribution with maximum a posteriori probability, wherein famous statistical models has markov random file (Markov Random Field, MRF) model and condition random field (Conditional Random Field, CRF) model., therefore adopt statistical method to cut apart and can obtain reasonable effect because the gray scale contact between pixel is larger due to the image in reality scene.And for medical image, cut apart demand calculated amount very greatly and owing to being easy to cause excessively by force losing edge details information because of correlativity, have very large challenge so cut apart on the medical image of intensity profile inequality due to statistical.

Dividing method based on graph theory comes from Clustering, its principle is image to be regarded as to the figure G=(V of cum rights, E), V is vertex set, E is the limit collection of figure, and G refers to the set that the unordered pair (or ordered pair) of different summits composition forms, and wherein can utilize gradation of image information etc. to construct the weighted value on limit, after having built vertex set and limit collection, complete cutting apart of image by finding optimum cut set.

Summary of the invention

The technical problem to be solved in the present invention mainly contains two: one be by which kind of method can extract well the interior adventitia profile of left ventricle magnetic resonance image (MRI) and calculate left ventricular ejection fraction and myocardial mass with respect to very Useful Information of diagnosing patients with heart failure; The two is to obtain after cardiac magnetic resonance image information, how it is combined with semantic technology, is applied among early diagnosis in heart failure.

The embodiment of the present invention provides a kind of detection method in heart failure of being combined with medical image segmentation based on semantic technology, comprises as follows:

Cut apart by the inner membrance to left ventricle magnetic resonance image (MRI) and adventitia, segmentation result is calculated, thereby convert image information to triplet information, be deposited in the body in heart failure of setting up by prot é g é, then utilize the powerful inference function of semantic technology, can robotization heart state is detected.

Described method comprises:

First input the sequence of cardiac magnetic resonance image, need subsequently, using first figure in this sequence as sample, need to sketch the contours of diaphragm area in heart left chamber by artificial mode;

Then, adopt protruding relaxed algorithm to calculate cardiac intima region by this sample, iteration initial curve using this interior diaphragm area as active contour model subsequently, thereby extract the adventitia profile of heart left ventricle, calculate the parameter such as ejection fraction and myocardial mass of heart left ventricle by the interior adventitia segmentation result obtaining;

Then, set up body in heart failure by prot é g é, then the parameter such as ejection fraction and myocardial mass calculating is stored in this body by the form of tlv triple, then in conjunction with the information relevant to this patient's diagnosing patients with heart failure storing in this body, thereby the illness of this patient's heart failure is learnt in reasoning.

The present invention has following beneficial effect, by the present invention, can, in the situation that needing manual intervention hardly, carry out the calculating of myocardial mass and the calculating of Left Ventricular Ejection Fraction to the cardiac magnetic resonance image sequence of input.The extraction accuracy of adventitia is also greatly improved in left ventricle, and the extraction that can reduce interior adventitia profile is subject to the impact of papillary muscle and artifact and the erroneous segmentation brought.Finally, set up body in heart failure, in conjunction with the parameter such as myocardial mass and Left Ventricular Ejection Fraction calculating, by the inference engine of body, can carry out feature extraction to the cardiac magnetic resonance image of input in robotization ground, then, in conjunction with other health information of patient, under the interference that needs hardly doctor, can make medical diagnosis, or, can provide very valuable advisory opinion (as making state of an illness diagnosis in heart failure and medication advice etc.) to diagnosis.

Brief description of the drawings

In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, to the accompanying drawing of required use in embodiment or description of the Prior Art be briefly described below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, do not paying under the prerequisite of creative work, can also obtain according to these accompanying drawings other accompanying drawing.

Fig. 1 is that in the heart left ventricle in the embodiment of the present invention, adventitia is cut apart process flow diagram;

Fig. 2 is the medical assistance decision system process flow diagram in heart failure in the embodiment of the present invention;

Fig. 3 is body in heart failure and the object properties schematic diagram in the embodiment of the present invention;

Fig. 4 is the interface structure figure of the semantic medical diagnosis system of heart failure in the embodiment of the present invention.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiment.Based on the embodiment in the present invention, those of ordinary skill in the art, not making all other embodiment that obtain under creative work prerequisite, belong to the scope of protection of the invention.

The testing process in heart failure of being combined with medical image segmentation based on semantic technology that the present invention proposes is as follows: the sequence of first inputting cardiac magnetic resonance image, need subsequently using first figure in this sequence as sample (being that heart minor axis is cut figure heart base portion), need to sketch the contours of diaphragm area in heart left chamber by artificial mode.Then, adopt protruding relaxed algorithm to calculate cardiac intima region by this sample, iteration initial curve using this interior diaphragm area as active contour model subsequently, thereby extract the adventitia profile of heart left ventricle, calculate the parameter such as ejection fraction and myocardial mass of heart left ventricle by the interior adventitia segmentation result obtaining.Then, set up body in heart failure by prot é g é, then the parameter such as ejection fraction and myocardial mass calculating is stored in this body by the form of tlv triple, then in conjunction with the information relevant to this patient's diagnosing patients with heart failure storing in this body, thereby the illness of this patient's heart failure is learnt in reasoning.Wherein Fig. 1 is the process flow diagram that in heart left ventricle, adventitia is cut apart, and Figure 2 shows that the realization flow figure of medical assistance diagnostic system in heart failure.

(1) adventitia extracting method in left ventricle

The extraction key of inner membrance is to set up and the gray feature of sample image and the distribution matching equation of distance feature contrast.The cardiac magnetic resonance image of input, the part that its gray feature and distance feature and sample similarity are higher will be retained, in addition, be subject to the impact of papillary muscle and artifact in order to prevent extracting region, need to obtain its marginal information by the gradient information that calculates cardiac magnetic resonance image, then make the interior diaphragm area of convergence be subject to the joint effect of edge feature, gray feature and distance feature.Place similar to inner membrance area grayscale in sample in cardiac magnetic resonance image will be extracted; And distance vector makes to only have the region approaching in inner membrance central point certain limit just to have higher weight, will be filtered away from the region of this central point like this; Marginal vectors this patent adopts gradient vector flow force field model, only has so the relatively region near border just to have higher weight.So, obtained by the common iteration of three by the inner membrance profile that is left chamber.

In the method for this patent important a bit, using the initial curve of as active contour model, left ventricular epicardium being cut apart by improving inner membrance contour curve that protruding lax distribution matching algorithm obtains.Because the iteration effect of active contour model is easily subject to the impact of initial curve, and inner membrance profile itself is more close with adventitia profile, the initial curve generating by inner membrance profile, can make active contour model can accurately converge to fast outer diaphragm area.And for the extraction of adventitia profile, this patent key point is the constraint to active contour model, this constraint is divided into local arc constraint and the conversion of external force field parameter adaptive.Local arc constraint refer to into each point in active contour curve partially, need to form part-circular with its 3～5 points around, to reach local round and smooth object.Simultaneously, also need to add 4 corrections, 4 corrections and local arc constraint comparing class are seemingly, guarantee each point and its around nearest 2 place straight lines and nearlyer 2 place straight lines distance within the specific limits, but not to be both not be that the each end points of curve judges while moving to its maximum, but carry out overall correction finish (a little just computing is complete) during iteration.In addition, also facing a problem is to encounter on border at curve, due to the effect of external force, curve may be restrained, but if border more complicated, near curve, border is more, because outer force parameter is constant, therefore while running into identical border, there will be the variation of curve to be controlled by the situation of internal force completely, thereby cause curve to cross convergence.In order to address this problem, this patent is that outer force parameter adds constraint, and this constraint is subject to the distance on the border running into first with outline line and the intensity effect on border.Shown in 2.1:

${\mathrm{\ω}}_{\mathrm{\τ}}=\frac{\log \mathrm{\θ}\·E_{\mathrm{ext}}}{\log [{(x-x_f)}^2+{(y-y_f)}^2]}---\left(2.1\right)$

(x in formula _f, y _f) be the boundary position running into for the first time, θ E _extfor current external force field intensity.

(2) semantic technology is applied to early diagnosis in heart failure

Just can calculate left chamber myocardial mass and Left Ventricular Ejection Fraction by the interior adventitia profile calculating in 2.2.1, ensuing work is exactly in early diagnosis in heart failure by these market demands.Body in heart failure should comprise following part concept in heart failure, patient, patient characteristic, detection and therapy, can utilize Ontology Editing Tool prot é g é to complete this work.Need to be class the object stores that does not relate to concrete things, similar medicine treatment saves as class, and subclass under drug therapy is if treatment group in heart failure is as subclass, and certain concrete class medicine is established as individuality as captopen.Can see from above-mentioned, the shangguan of contacting directly at class and class is subordinate relation, show as the relation of class and subclass, between class and class, also having a kind of closing is in addition object properties, exactly because this object properties, allow the tlv triple database be no longer simple hierarchical structure, but have the snowflake type pattern in a similarity relation database.For example need the individuality in patient class and patient characteristic class to link together, need object properties " health ", as " patient a-> health-> patient characteristic ".As shown in Figure 3, detect list item class and be subordinated to and detect class but be not subordinated to patient class, but by object properties and its contact together with.In body, except class, also have a kind of data type to be called individuality (instance).Individuality refers to the instantiation of class, and such as angiotensin converting enzyme inhibitor class comprises concrete medicine enalapril, captopen, Trandolapril, lisinopril and Ramipril etc.; As comprised concrete certain patient in patient's class as Jack and Tom etc.This means that individuality occurs as an example in body in heart failure, that is to say, the parameter such as Left Ventricular Ejection Fraction and volume of calculating according to the CMRI of input by cardiac image disposal system stores as this heart failure body, and need to be stored under patient's class.Inference function is more efficient on to the reasoning of information with respect to relational database in tlv triple database.In tlv triple database, reasoning is to be based upon on the basis of class, individuality, object properties and data attribute, the subordinate attribute by object properties and class by and according to the Inference Conditions of input, just can easily infer destination object.

The interface structure of semantic medical diagnosis system in heart failure as shown in Figure 4, because what system adopted is restful framework, shown in this interface layer aggregated(particle) structure set up based on restful framework, be divided into RESTFUL event receiving layer, user information authentication layer, parameter acquiring layer, event handling layer, RDF data operation layer and data and return to layer.Total has completed the structure to system total interface.In data-switching-view data translation interface of the interface that wherein view data converts tlv triple data in event handling layer.This interface level shows, in the time sending request of data, can be received by RESTFUL event receiving layer, then complete a series of certification by user information authentication layer, then the type of diagnosis request is carried out event handling (as medication advice, call the inference system of body in heart failure, provide medication advice by existing data according to request), after receiving request, need RDF data operation layer to carry out the operation of concrete ontology data, finally return to layer by data and return to the data that user or system are specified.

The present invention has following beneficial effect, by the present invention, can, in the situation that needing manual intervention hardly, carry out the calculating of myocardial mass and the calculating of Left Ventricular Ejection Fraction to the cardiac magnetic resonance image sequence of input.The extraction accuracy of adventitia is also greatly improved in left ventricle, and the extraction that can reduce interior adventitia profile is subject to the impact of papillary muscle and artifact and the erroneous segmentation brought.Finally, set up body in heart failure, in conjunction with the parameter such as myocardial mass and Left Ventricular Ejection Fraction calculating, by the inference engine of body, can carry out feature extraction to the cardiac magnetic resonance image of input in robotization ground, then, in conjunction with other health information of patient, under the interference that needs hardly doctor, can make medical diagnosis, or, can provide very valuable advisory opinion (as making state of an illness diagnosis in heart failure and medication advice etc.) to diagnosis.

One of ordinary skill in the art will appreciate that all or part of step in the whole bag of tricks of above-described embodiment is can carry out the hardware that instruction is relevant by program to complete, this program can be stored in a computer-readable recording medium, storage medium can comprise: ROM (read-only memory) (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or CD etc.

A kind of detection method in heart failure of being combined with medical image segmentation based on semantic the technology above embodiment of the present invention being provided is described in detail, applied specific case herein principle of the present invention and embodiment are set forth, the explanation of above embodiment is just for helping to understand method of the present invention and core concept thereof; , for one of ordinary skill in the art, according to thought of the present invention, all will change in specific embodiments and applications, in sum, this description should not be construed as limitation of the present invention meanwhile.

Claims (2)

1. a detection method in heart failure of being combined with medical image segmentation based on semantic technology, is characterized in that, described method comprises:

Cut apart by the inner membrance to left ventricle magnetic resonance image (MRI) and adventitia, segmentation result is calculated, thereby convert image information to triplet information, be deposited in the body in heart failure of setting up by prot é g é, then utilize the powerful inference function of semantic technology, can robotization heart state is detected.

2. detection method in heart failure of being combined with medical image segmentation based on semantic technology as claimed in claim 1, is characterized in that, described method comprises:

First input the sequence of cardiac magnetic resonance image, need subsequently, using first figure in this sequence as sample, need to sketch the contours of diaphragm area in heart left chamber by artificial mode;

Then, adopt protruding relaxed algorithm to calculate cardiac intima region by this sample, iteration initial curve using this interior diaphragm area as active contour model subsequently, thereby extract the adventitia profile of heart left ventricle, calculate the parameter such as ejection fraction and myocardial mass of heart left ventricle by the interior adventitia segmentation result obtaining;

Then, set up body in heart failure by prot é g é, then the parameter such as ejection fraction and myocardial mass calculating is stored in this body by the form of tlv triple, then in conjunction with the information relevant to this patient's diagnosing patients with heart failure storing in this body, thereby the illness of this patient's heart failure is learnt in reasoning.