CN107680134A

CN107680134A - Vertebra scaling method, device and equipment in medical image

Info

Publication number: CN107680134A
Application number: CN201710912344.3A
Authority: CN
Inventors: 杨越淇; 韩冬
Original assignee: Neusoft Medical Systems Co Ltd
Current assignee: Neusoft Medical Systems Co Ltd
Priority date: 2017-09-29
Filing date: 2017-09-29
Publication date: 2018-02-09
Anticipated expiration: 2037-09-29
Also published as: CN107680134B

Abstract

The invention discloses vertebra scaling method, device and equipment in a kind of medical image, methods described includes：Obtain the medical image of subject；A sampled point is determined respectively on the often section vertebrae in the medical image；At least one sample to be tested is built according to multiple sampled points of determination；Feature extraction is carried out respectively at least one sample to be tested, obtains the sample characteristics of each sample to be tested；The vertebra demarcation that the sample characteristics of each sample to be tested are inputted to training in advance respectively returns device, obtains the calibration result of each sample to be tested；The vertebra calibration result of the subject is determined according to the calibration result of each sample to be tested.The present invention can make full use of the correlation between adjacent ridge vertebra, improve the otherness between the identification of sample and sample, strengthen the accuracy and robustness of calibration result.

Description

Vertebra scaling method, device and equipment in medical image

Technical field

The present invention relates to vertebra scaling method in technical field of image processing, more particularly to a kind of medical image, device and Equipment.

Background technology

Spondylodynia is the common disease of mid-aged population and the high trend fallen ill, show rejuvenation again in recent years.Human body ridge The demarcation of post can help to the diagnosis of spondylodynia, and then establish good base for judgement, the determination of therapeutic scheme of vertebra pathology Plinth.

The method of generally use hand labeled vertebra point carries out backbone demarcation in the prior art, and this method requires that doctor is necessary With enough rich experiences, the accuracy of mark is just can guarantee that, particularly when processing only includes the image of part vertebra, Due to having the repeatability of height between every section vertebrae, on morphologic features, difference is smaller for it, is difficult to discriminate between, and occurs wrong Probability greatly increases by mistake.On the other hand, the method for hand labeled vertebra point usually requires to expend the substantial amounts of energy of doctor, easily Cause doctor's fatigue weak, thus can also increase the possibility of maloperation.

The content of the invention

In view of this, the present invention proposes in a kind of medical image vertebra scaling method, device and equipment to solve above-mentioned skill Art problem.

In order to achieve the above object, the technical solution adopted in the present invention is：

First aspect according to embodiments of the present invention, it is proposed that vertebra scaling method in a kind of medical image, including：

The medical image of subject is obtained, the medical image includes the more piece vertebrae of the subject；

A sampled point is determined respectively on the often section vertebrae in the medical image；

At least one sample to be tested is built according to multiple sampled points of determination；Wherein, each sample to be tested bag Containing the sampled point corresponding to n continuous vertebraes, and there are n-1 overlapping sampled points between adjacent sample to be tested；

Feature extraction is carried out respectively at least one sample to be tested, the sample for obtaining each sample to be tested is special Sign；

The vertebra demarcation that the sample characteristics of each sample to be tested are inputted to training in advance respectively returns device, obtains each The calibration result of the sample to be tested, the calibration result represent corresponding in sample to be tested each affiliated vertebrae of sampled point class Not；

The vertebra calibration result of the subject is determined according to the calibration result of each sample to be tested.

Second aspect according to embodiments of the present invention, it is proposed that vertebra caliberating device in a kind of medical image, including：

Testing image acquisition module, for obtaining the medical image of subject, the medical image includes described tested The more piece vertebrae of body；

Sampled point determining module to be measured, one is determined respectively on the often section vertebrae in the medical image Sampled point；

Sample to be tested builds module, and at least one sample to be tested is built for multiple sampled points according to determination；Its In, each sample to be tested includes the sampled point corresponding to n continuous vertebraes, and has between adjacent sample to be tested N-1 overlapping sampled points；

Characteristic extracting module to be measured, for carrying out feature extraction respectively at least one sample to be tested, obtain each The sample characteristics of the sample to be tested；

Feature input module to be measured, for the sample characteristics of each sample to be tested to be inputted to the ridge of training in advance respectively Vertebra demarcation returns device, obtains the calibration result of each sample to be tested, the calibration result represent corresponding in sample to be tested The classification of each affiliated vertebrae of sampled point；

Calibration result determining module, for determining the ridge of the subject according to the calibration result of each sample to be tested Vertebra calibration result.

The third aspect according to embodiments of the present invention, it is proposed that a kind of electronic equipment, the electronic equipment include：

Processor；

It is configured as storing the memory of processor-executable instruction；

Wherein, the processor is configured as：

Fourth aspect according to embodiments of the present invention, it is proposed that a kind of computer-readable recording medium, be stored thereon with meter Calculation machine program, the program are realized when being processed by the processor：

Compared with prior art, vertebra scaling method in medical image of the invention, by the medical science for obtaining subject Image, a sampled point is determined on the often section vertebrae in medical image respectively, and built according to multiple sampled points of determination At least one sample to be tested, carries out feature extraction at least one sample to be tested respectively, and then by the sample of each sample to be tested The vertebra demarcation that feature inputs training in advance respectively returns device, to determine subject according to the calibration result of each sample to be tested Vertebra calibration result, due to building sample to be tested using sampled point corresponding to multiple continuous vertebraes, thus it can make full use of Correlation between adjacent ridge vertebra, the otherness between the identification of sample and sample is improved, strengthen the accuracy of calibration result And robustness.

Brief description of the drawings

Figure 1A shows the flow chart of vertebra scaling method in the medical image according to one example embodiment of the present invention；

Figure 1B shows vertebra schematic diagram in the medical image according to one example embodiment of the present invention；

Fig. 2A is shown according to one example embodiment of the present invention how on the often section vertebrae in medical image Determine the flow chart of sampled point；

Fig. 2 B show the effect that sampled point is determined on every section vertebrae according to one example embodiment of the present invention Schematic diagram；

Fig. 3 A show the flow of vertebra scaling method in medical image in accordance with a further exemplary embodiment of the present invention Figure；

Fig. 3 B are shown to be shown according to the result that image segmentation is carried out to medical image of one example embodiment of the present invention It is intended to；

How Fig. 4 A are shown according to one example embodiment of the present invention according to the calibration result of each sample to be tested Determine the flow chart of the vertebra calibration result of subject；

Fig. 4 B show the vertebra calibration result schematic diagram according to one example embodiment of the present invention；

Fig. 5 shows the flow for how determining confidence level highest sampled point according to one example embodiment of the present invention Figure；

Fig. 6 A show the flow chart for how training vertebra demarcation to return device according to one example embodiment of the present invention；

Fig. 6 B show the demarcation schematic diagram according to the training sample of one example embodiment of the present invention；

Fig. 7 show according to one example embodiment of the present invention how the often section sample ridge in sample medical image The flow chart of sampled point is determined on vertebra；

Fig. 8 shows the flow for how training vertebra demarcation to return device in accordance with a further exemplary embodiment of the present invention Figure；

Fig. 9 shows the structural frames of vertebra caliberating device in the medical image according to one example embodiment of the present invention Figure；

Figure 10 shows the structure of vertebra caliberating device in medical image in accordance with a further exemplary embodiment of the present invention Block diagram；

Figure 11 shows the structured flowchart of the recurrence device training module according to one example embodiment of the present invention；

Figure 12 shows the structured flowchart of the electronic equipment according to one example embodiment of the present invention.

Embodiment

Below with reference to embodiment shown in the drawings, the present invention will be described in detail.But these embodiments are simultaneously The present invention is not limited, structure that one of ordinary skill in the art is made according to these embodiments, method or functionally Conversion is comprising within the scope of the present invention.

It is only merely for the purpose of description specific embodiment in terminology used in the present invention, and is not intended to be limiting the present invention. It is also intended in " one kind " of the singulative of the invention with used in appended claims, " described " and "the" including majority Form, unless context clearly shows that other implications.It is also understood that term "and/or" used herein refers to and wrapped Containing the associated list items purpose of one or more, any or all may be combined.

It will be appreciated that though various structures, but these structures may be described using term first, second etc. in the present invention It should not necessarily be limited by these terms.These terms are only used for same type of structure being distinguished from each other out.

Figure 1A shows the flow chart of vertebra scaling method in the medical image according to one example embodiment of the present invention； As shown in Figure 1A, this method comprises the following steps S101-S106：

S101：Obtain the medical image of subject；

In an optional embodiment, Figure 1B shows the medical image median ridge according to one example embodiment of the present invention Vertebra schematic diagram；As shown in Figure 1B, the medical image includes the more piece vertebrae of the subject (at arrow instruction).

In an optional embodiment, above-mentioned medical image can be included by CT (Computer Tomography, electricity Sub- computed tomography), MRI (Magnetic Resonance Imaging, nuclear magnetic resonance), ultrasound (Ultrasound, ) and the medical imaging skill such as digital campus imaging technique (Digital Subtraction Angiography, DSA) US The three-dimensional data for the subject that art obtains.

S102：A sampled point is determined respectively on the often section vertebrae in the medical image；

In an optional embodiment, sampled point, this implementation can be determined in any position on every section vertebrae Example is to the position of sampled point without limiting.

In an optional embodiment, support vector regression (Support Vector Regression), god can be used Through network (Neural Network), decision tree (Decision Tree) scheduling algorithm or manual mode in every section vertebra A sampled point is determined on bone respectively.

S103：At least one sample to be tested is built according to multiple sampled points of determination；

In an optional embodiment, the medical image of m section vertebras is included for a width, n can be chosen successively continuously The sampled point forms each sample to be tested corresponding to vertebrae, and makes have n-1 individual overlapping between adjacent sample to be tested Sampled point, then it can obtain (m-n+1) individual sample to be tested.

For example, as m=22 (i.e. in medical image comprising 22 section vertebras), n=4, (i.e. each sample to be tested is comprising 4 Continuous vertebrae), then the sample chosen is followed successively by：1-2-3-4、2-3-4-5、3-4-5-6、……、18-19-20-21、19- 20-21-22 (digitized representation sampled point therein is subordinate to the classification of vertebrae), then m-n+1=19 are can obtain altogether and treats test sample This.

S104：Feature extraction is carried out respectively at least one sample to be tested, obtains the sample of each sample to be tested Eigen；

In an optional embodiment, the feature extracted can be the characteristics of can protruding the sampled point well, again The feature distinguished with contacting with other sampled points in same sample can be shown, with Enhanced feature identification.

In an optional embodiment, a kind of feature can be extracted to the sample to be tested, or extract a variety of spies simultaneously Sign (corresponding with the feature species for returning the extraction of device training process), fully comprehensively to show the affiliated vertebra class of the sampled point Other feature.

In an optional embodiment, features described above can include single order feature, second order feature, wavelet character (Wavelet Feature) etc., the present embodiment is to this without limiting.

S105：The vertebra demarcation that the sample characteristics of each sample to be tested are inputted to training in advance respectively returns device, obtains To the calibration result of each sample to be tested；

In an optional embodiment, the calibration result can represent each affiliated ridge of sampled point in corresponding sample to be tested The classification of vertebra.

It should be noted that cervical vertebra, thoracic vertebrae, lumbar vertebrae can not differentiated between in the present embodiment, and they are collectively referred to as vertebra, The position that the classification of the spinal bone can be located at according to the block vertebra in backbone determines that calibration process specifically may refer to Fig. 6 A illustrated embodiments, herein first without being described in detail.

In an optional embodiment, as m=22, the calibration result of above-mentioned m-n+1 sample to be tested can form one The matrix that individual (m-n+1) × 22 are tieed up, the row vector of the matrix are followed successively by the calibration result of each sample to be tested, Mei Gehang from top to bottom Vector is respectively provided with n close to 1 element, and 22-n close to 0 element, and this n is worth close to the position of 1 element Put the classification for representing each affiliated vertebrae of sampled point in sample to be tested.

In an optional embodiment, by the sample characteristics of each sample to be tested, (or the sample being made up of various features is special Collection close) respectively input training in advance vertebra demarcation return device, obtain the calibration result of each sample to be tested.

S106：The vertebra calibration result of the subject is determined according to the calibration result of each sample to be tested.

It is continuously arranged due in human vertebra, being under the jurisdiction of different classes of vertebrae, thus test sample is treated known This calibration result, you can extrapolate the classification (i.e. the vertebra calibration result of subject) of all vertebraes in image accordingly.

Vertebra scaling method in the medical image of the present embodiment, by obtaining the medical image of subject, in medical image In often section vertebrae on determine a sampled point respectively, and treat test sample according to multiple sampled points of determination structure is at least one This, feature extraction is carried out at least one sample to be tested respectively, and then the sample characteristics of each sample to be tested is inputted respectively pre- The vertebra demarcation first trained returns device, to determine the vertebra calibration result of subject according to the calibration result of each sample to be tested, Due to using sampled point structure samples to be tested corresponding to multiple continuous vertebraes, thus can make full use of between adjacent ridge vertebra Correlation, improve the otherness between the identification of sample and sample, strengthen the accuracy and robustness of calibration result.

Fig. 2A is shown according to one example embodiment of the present invention how on the often section vertebrae in medical image Determine the flow chart of sampled point；The present embodiment on the basis of Figure 1A illustrated embodiments, with how the often section in medical image It is illustrative exemplified by determination sampled point on vertebrae.As shown in Figure 2 A, in the medical image in above-mentioned steps S102 In the often section vertebrae on determine that a sampled point may comprise steps of S201-S202 respectively：

S201：Down-sampled processing is carried out to the medical image, obtains down-sampled image；

In an optional embodiment, in order to comprising global information, avoid lacking for the high similitude of vertebra image local information Fall into, and avoid impacting contiguous range because pixel is larger in three direction size differences, can be to the doctor Learn image and carry out down-sampled processing；Specifically, three described directions refer to three directions of x, y, z in human body coordinate system, The size difference is mainly what is determined by the machine scanned, and most situation is the pixel size in z directions than x, y side To it is big.And due to needing fixed window size (i.e. regular length) during extraction feature, if pixel is of different sizes, equally long On the premise of degree, comprising pixel number by different (pixel numbers=length/pixel size).

In an optional embodiment, bilinear interpolation method can be used to carry out down-sampled place to the medical image Reason, obtains down-sampled image.

For example, original image size 512*512*328, pixel size are 0.9766*0.9766*2.5, and drop is adopted Image size can be 250*250*410 after sample, and pixel size can be 2*2*2.

S202：A sampled point is determined respectively on the often section vertebrae in the down-sampled image.

, can be by the way of above-mentioned automatic or manual in an optional embodiment, the often section in down-sampled image A sampled point is determined on the vertebrae respectively.One example embodiment of the present invention determines sampling on every section vertebrae The effect diagram of point may refer to Fig. 2 B.

As shown from the above technical solution, the present embodiment realizes the often section ridge in medical image by step S201-S202 Sampled point is determined on vertebra, the defects of vertebra image local information high similitude can be avoided, and avoid because pixel exists Three direction size differences are larger and contiguous range is impacted, and are advantageous to improve the follow-up accuracy for carrying out vertebra demarcation.

Fig. 3 A show the flow of vertebra scaling method in medical image in accordance with a further exemplary embodiment of the present invention Figure；As shown in Figure 3A, this method may comprise steps of S301-S307：

S301：The medical image of subject is obtained, the medical image includes the more piece vertebrae of the subject；

S302：Image segmentation is carried out to the medical image using pre-set image dividing method, extracts the medical image In more piece vertebrae corresponding to image-region；

In an optional embodiment, support vector regression (Support Vector Regression), god can be used Through pre-set image dividing methods such as network (Neural Network), decision trees (Decision Tree) to the medical image Image segmentation is carried out, to extract image-region corresponding to the more piece vertebrae in medical image.Specifically, Fig. 3 B show basis The result schematic diagram that image segmentation is carried out to medical image of one example embodiment of the present invention；As shown in Figure 3 B, more piece ridge Image-region corresponding to vertebra is multiple approximate rectangular regions in figure.

S303：Described image region based on extraction, it is true respectively on the often section vertebrae in the medical image A fixed sampled point；

, can be described in the often section in image-region corresponding to the more piece vertebrae extracted in an optional embodiment A sampled point is determined on vertebrae respectively.

S304：At least one sample to be tested is built according to multiple sampled points of determination；Wherein, it is each described to treat test sample This includes the sampled point corresponding to n continuous vertebraes, and has n-1 overlapping samplings between adjacent sample to be tested Point；

S305：Feature extraction is carried out respectively at least one sample to be tested, obtains the sample of each sample to be tested Eigen；

S306：The vertebra demarcation that the sample characteristics of each sample to be tested are inputted to training in advance respectively returns device, obtains To the calibration result of each sample to be tested, each affiliated vertebra of sampled point in sample to be tested corresponding to the calibration result expression The classification of bone；

S307：The vertebra calibration result of the subject is determined according to the calibration result of each sample to be tested

Wherein, step S301, S304-S307 is identical with step S101, S103-S106 in above-described embodiment, herein not Repeated.

As shown from the above technical solution, the present embodiment carries out image by using pre-set image dividing method to medical image Segmentation, image-region corresponding to the more piece vertebrae in medical image, and the image-region based on extraction are extracted, in medical image In often section vertebrae on determine a sampled point respectively, can improve on vertebrae determine sampled point the degree of accuracy, and then The follow-up accuracy for carrying out vertebra demarcation can be improved.

How Fig. 4 A are shown according to one example embodiment of the present invention according to the calibration result of each sample to be tested Determine the flow chart of the vertebra calibration result of subject；The present embodiment is on the basis of above-described embodiment, with how according to each The calibration result of sample to be tested determines illustrative exemplified by the vertebra calibration result of subject.As shown in Figure 4 A, it is above-mentioned The vertebra calibration result of the subject is determined in step S106 according to the calibration result of each sample to be tested, can be included Step S401-S402：

S401：Determine confidence level highest sampled point in the calibration result of each sample to be tested；

In an optional embodiment, each sampled point can in calibration result that can be by calculating each sample to be tested Reliability, and then determine wherein confidence level highest sampled point.

S402：On the basis of the calibration result of the confidence level highest sampled point, order, which calculates, obtains the subject Vertebra calibration result.

It is continuously arranged due to being under the jurisdiction of different classes of vertebrae in human vertebra, thus can be according to known The calibration position or classification of any one section vertebrae, sequentially extrapolate the vertebrate classification of institute in image.

In the present embodiment, on the basis of the position of the confidence level highest sampled point of determination or classification, order is calculated and obtained Whole vertebra calibration results of subject (referring to Fig. 4 B).

As shown from the above technical solution, confidence level in calibration result of the present embodiment by determining each sample to be tested Highest sampled point, and on the basis of the calibration result of the confidence level highest sampled point, order, which calculates, obtains described be detected The vertebra calibration result of body, due to considering the confidence level of sampled point in calibration result, thus it may insure obtained subject Whole vertebra calibration results confidence level, improve the degree of accuracy of vertebra calibration result.

Fig. 5 shows the flow for how determining confidence level highest sampled point according to one example embodiment of the present invention Figure；The present embodiment carries out example on the basis of Fig. 4 A illustrated embodiments exemplified by how determining confidence level highest sampled point Property explanation.As shown in figure 5, determine that confidence level highest is adopted in the calibration result of each sample to be tested in above-mentioned steps S401 Sampling point, step S501-S504 can be included：

S501：The gradient mean value of the calibration result of each sample to be tested is calculated, to determine that the gradient mean value is maximum Two samples to be tested；

It is each to assess by calculating the gradient mean value of calibration result of each sample to be tested in an optional embodiment The precision of sample to be tested, and take the gradient mean value of wherein calibration result maximum two samples to be tested T1 and T2.

In an optional embodiment, the calibration result of each sample to be tested can be the sparse vector of 1 × m dimension. There is continuous n value close to 1 element and the m-n element being worth close to 0 in the sparse vector, and described n is worth The classification of each affiliated vertebrae of sampled point close in training sample described in the positional representation of 1 element, n are each training sample The number of middle sampled point, m are the joint number for the vertebrae that the medical image includes.

For example, if sample to be tested T1 calibration result is：

[…,0.012773149,0.039999835,0.95791978,1.0259485,1.0500451,

0.95180166,0.053262822,0.015799209,…]

Then the gradient mean value of sample to be tested T1 calibration result can calculate according to following formula：

For example, above formulaThe implication of " 1 " is 0.95791978 in denominator With the difference of the 0.039999835 row ordinal number in calibration result vector；Similarly, formula The implication of " 1 " is also the difference of the 0.95180166 and 0.053262822 row ordinal number in calibration result vector in denominator.

S502：Determine to demarcate end value two sampled points of highest in described two samples to be tested；

In an optional embodiment, calibration result value can be the element value in above-mentioned sparse vector.For example, above-mentioned treat This T1 of test sample calibration result value is up to 1.0500451, then calibration result value highest sampled point is 1.0500451 institute Corresponding sampled point.

S503：According to the calibration position difference of described two sampled points and the physical location difference of described two sampled points Comparison result, judge whether the calibration result of described two samples to be tested is accurate；

In an optional embodiment, if finding demarcation end value highest point p1, p2 in sample to be tested T1 and T2 respectively, Can now determine to exist in sample to be tested T1 and T2 calibration result continuous n it is individual close to 1 value, and remaining (m-n) Close to 0 value.On this basis, measuring point p1, p2 corresponding vector (vector by continuous n it is individual close to 1 value institute group Into) in inverted order index d1, d2.

On this basis, the difference of point p1, p2 abscissa and d1, d2 difference and value w (i.e. w=(p1-p2)+(d1- are calculated D2)), should and value w expressions two calibration results most accurate sample to be tested T1 and T2 in confidence level highest test sample point s1, Alternate position spikes of the s2 in original image.

Wherein, the abscissa of sampled point is the sequence number of the affiliated sample to be tested of the sampled point.

Further, the difference y of point p1, p2 ordinate is calculated, it represents the position of test sample point s1, s2 in calibration result Put difference.

Whether and then it is identical with w values to compare above-mentioned y values；It is if identical, then it represents that above-mentioned calibration result is accurate；Otherwise, Represent that above-mentioned calibration result is inaccurate.It is understood that above-mentioned w values are represented in real image, two samplings of s1, s2 The range difference of point, this is exactly ground truth (goldstandard), and the expression of y values is prediction result in the present embodiment, such as Two values of fruit are identical, then it represents that above-mentioned calibration result is accurate.Otherwise, it means that above-mentioned calibration result mistake.

S504：If judgement knows that the calibration result of described two samples to be tested is accurate, described two sampled points are got the bid Determine end value highest sampled point and be defined as the confidence level highest sampled point.

As shown from the above technical solution, the present embodiment is by calculating the gradient mean value of the calibration result of each sample to be tested, To determine two maximum samples to be tested of gradient mean value, and then determine that end value highest two is demarcated in two samples to be tested adopts Sampling point, with the comparison result of the calibration position difference according to two sampled points and the physical location difference of two sampled points, judge Whether the calibration result of described two samples to be tested is accurate, can accurately determine confidence level highest sampled point, due to considering The confidence level of sampled point in calibration result, thus the whole vertebra calibration results of subject that may insure to obtain is credible Degree, improves the degree of accuracy of vertebra calibration result.

Fig. 6 A show the flow chart for how training vertebra demarcation to return device according to one example embodiment of the present invention； The present embodiment is illustrative exemplified by how training vertebra demarcation to return device on the basis of above-described embodiment.Such as figure Shown in 6A, the method comprising the steps of S601-S606：

S601：Several sample medical images are obtained, include more piece sample vertebrae in sample medical image described in every width；

In an optional embodiment, above-mentioned sample medical image can include passing through CT (Computer Tomography, CT scan), MRI (Magnetic Resonance Imaging, nuclear magnetic resonance), ultrasound And the obtained three-dimensional data of sample subject of the medical imaging technology such as digital campus imaging technique (DSA) (US).

In the prior art, according to the division methods of human anatomic structure, vertebra can be divided into 24 classes (the wherein class of cervical vertebra 7, chest The class of vertebra 12, the class of lumbar vertebrae 5), along with the class of background 1 can be divided into 25 classes altogether.Because the uppermost two classes image of cervical vertebra is not easy to obtain, because And vertebra can be divided into：22 classes (class of cervical vertebra 5, the class of thoracic vertebrae 12 and the class of lumbar vertebrae 5), along with the class of background 1 totally 23 class.Can one In the embodiment of choosing, cervical vertebra, thoracic vertebrae, lumbar vertebrae can not be made a distinction in follow-up calibration, but they are collectively referred to as ridge Vertebra, the position being located at according to each piece of vertebra in backbone determine the classification of the block vertebrae, and using label L1~L22 according to from Order under up to represents the classification of each piece of spinal bone successively.

Inventor has found through overtesting, the vertebra in medical image be in human body it is irregular, and lumbar vertebrae relative to Difference in size is larger for cervical vertebra, therefore it is difficult to ensure that the number of training proportion per class is identical, thus can cause to influence The precision and robustness of follow-up vertebra scaling method.In view of this, in an optional embodiment, it may be referred to human medical figure The segmentation figure of vertebra, the ratio for making every class training sample account for whole number of training are identical as in.It should be strongly noted that this Referenced segmentation figure and there need not be very accurately segmentation effect in embodiment, only needing can be by the major part of every class vertebra Content segmentation goes out.

S602：A sampling is determined on the often section sample vertebrae in sample medical image described in every width respectively Point；

In an optional embodiment, sampled point, this implementation can be determined in any position on every section sample vertebrae Example is to the position of sampled point without limiting.

In an optional embodiment, support vector regression (Support Vector Regression), god can be used Through network (Neural Network), decision tree (Decision Tree) scheduling algorithm or manual mode in every section sample vertebra A sampled point is determined on bone respectively.

S603：Multiple training samples are built according to multiple sampled points of determination；Wherein, each training sample bag Containing the sampled point corresponding to n continuous vertebraes；

In an optional embodiment, the sample medical image of L section vertebras is included for a width, n can be chosen continuously The sampled point forms each training sample corresponding to vertebrae, then can obtain (L-n+1) individual training sample.

It should be noted that there is n-1 weight between each training sample that need not ensure to build in the training stage in this step Multiple sampled point, and it is to include n continuous sampling point only to need to ensure inside each training sample.

S604：Feature extraction is carried out respectively to the multiple training sample, obtains the respective sample of the multiple training sample Eigen；

In an optional embodiment, a kind of feature can be extracted to the training sample, or extract a variety of spies simultaneously Sign, the characteristics of fully comprehensively to show the sampled point affiliated vertebra classification.

S605：Each self-corresponding label of the multiple training sample is determined, the label is used to represent corresponding training sample The classification of each affiliated vertebrae of sampled point in this；

Can be the sparse vector conduct that each training sample sets that a 1 × m is tieed up in an optional embodiment Corresponding label.Fig. 6 B show the demarcation schematic diagram according to the training sample of one example embodiment of the present invention；Such as Fig. 6 B It is shown, can have the element (n=4 in this example) that continuous n value is 1 and the member that m-n value is 0 in the sparse vector Element, and the n value is that the position of 1 element can represent the classification of each affiliated vertebrae of sampled point in the training sample, n For the number of sampled point in each training sample, m is the joint number for the vertebrae that the medical image includes.

S606：The default device that returns, which is performed, according to each self-corresponding label of the multiple training sample and sample characteristics trains calculation Method, obtain the vertebra demarcation and return device.

In an optional embodiment, the training sample and its corresponding demarcation that can utilize above-mentioned construction are performed and preset back Return device training algorithm, device is returned with one human vertebra demarcation of training, for the continuous n vertebra sampled point for arbitrarily inputting Feature, export the demarcation of corresponding vertebral location.

Vertebra scaling method in the medical image of the present embodiment, by obtaining several sample medical images, in every width sample A sampled point is determined on often section sample vertebrae in medical image respectively, and it is multiple according to multiple sampled points of determination structure Training sample, feature extraction is carried out respectively to multiple training samples, obtain the respective sample characteristics of multiple training samples, and then really Fixed multiple each self-corresponding labels of training sample, it is pre- to be performed according to each self-corresponding label of multiple training samples and sample characteristics If returning device training algorithm, obtain vertebra demarcation and return device, due to using sampled point structure instruction corresponding to multiple continuous vertebraes Practice sample to train vertebra demarcation to return device, thus the correlation between adjacent ridge vertebra can be made full use of, improve sample Otherness between identification and sample, strengthen subsequently based on training vertebra demarcation return device carry out vertebra demarcation accuracy with And robustness.

Fig. 7 show according to one example embodiment of the present invention how the often section sample ridge in sample medical image The flow chart of sampled point is determined on vertebra；The present embodiment is on the basis of Fig. 6 A illustrated embodiments, with how in sample medical science figure It is illustrative exemplified by determination sampled point on often section sample vertebrae as in.As shown in fig. 7, in above-mentioned steps S602 A sampled point is determined respectively on the often section sample vertebrae in sample medical image described in every width, can include step S701-S702：

S701：Down-sampled processing is carried out to the sample medical image, obtains down-sampled sample image；

In an optional embodiment, in order to comprising global information, avoid lacking for the high similitude of vertebra image local information Fall into, and avoid impacting contiguous range because pixel is larger in three direction size differences, can be to the sample This medical image carries out down-sampled processing.

S702：Using the default method of sampling, divide on the often section sample vertebrae in the down-sampled sample image Que Ding not a sampled point.

, can be by the way of above-mentioned automatic or manual, in down-sampled sample image in an optional embodiment Often save and determine a sampled point on the vertebrae respectively.

As shown from the above technical solution, the present embodiment is realized every in sample medical image by step S701-S702 Sampled point is determined on section vertebrae, the defects of vertebra image local information high similitude can be avoided, and avoid due to pixel It is o'clock larger in three direction size differences and contiguous range is impacted, be advantageous to improve the follow-up essence for carrying out returning device training Exactness.

Fig. 8 shows the flow for how training vertebra demarcation to return device in accordance with a further exemplary embodiment of the present invention Figure；As shown in figure 8, this method may comprise steps of S801-S807：

S801：Several sample medical images are obtained, include more piece sample vertebrae in sample medical image described in every width；

S802：Image segmentation is carried out to sample medical image described in every width using pre-set image dividing method, described in extraction Image-region corresponding to more piece sample vertebrae in sample medical image；

In an optional embodiment, support vector regression (Support Vector Regression), god can be used Through pre-set image dividing methods such as network (Neural Network), decision trees (Decision Tree) to the sample medical science Image carries out image segmentation, to extract image-region corresponding to the more piece vertebrae in sample medical image.

S803：Described image region based on extraction, the often section sample ridge in sample medical image described in every width A sampled point is determined on vertebra respectively；

In an optional embodiment, image corresponding to the more piece vertebrae that can be extracted in every width sample medical image A sampled point is determined respectively on the often section vertebrae in region.

S804：Multiple training samples are built according to multiple sampled points of determination；Wherein, each training sample bag Containing the sampled point corresponding to n continuous vertebraes；

S805：Feature extraction is carried out respectively to the multiple training sample, obtains the respective sample of the multiple training sample Eigen；

S806：Each self-corresponding label of the multiple training sample is determined, the label is used to represent corresponding training sample The classification of each affiliated vertebrae of sampled point in this；

S807：The default device that returns, which is performed, according to each self-corresponding label of the multiple training sample and sample characteristics trains calculation Method, obtain the vertebra demarcation and return device.

Wherein, step S801, S804-S807 is identical with step S601, S603-S606 in above-described embodiment, herein not Repeated.

As shown from the above technical solution, the present embodiment is carried out by using pre-set image dividing method to sample medical image Image is split, and extracts image-region corresponding to the more piece vertebrae in sample medical image, and the image-region based on extraction, A sampled point is determined respectively on often section vertebrae in sample medical image, can improve and sampled point is determined on vertebrae The degree of accuracy, and then the follow-up accuracy for carrying out returning device training can be improved.

Fig. 9 shows the structural frames of vertebra caliberating device in the medical image according to one example embodiment of the present invention Figure；As shown in figure 9, the device includes testing image acquisition module 110, sampled point determining module 120 to be measured, sample to be tested structure Module 130, characteristic extracting module to be measured 140, feature input module 150 to be measured and calibration result determining module 160, wherein：

Testing image acquisition module 110, for obtaining the medical image of subject, the medical image includes the quilt The more piece vertebrae of a corpse or other object for laboratory examination and chemical testing；

Sampled point determining module 120 to be measured, determined respectively on the often section vertebrae in the medical image One sampled point；

Sample to be tested builds module 130, and at least one sample to be tested is built for multiple sampled points according to determination； Wherein, each sample to be tested includes the sampled point corresponding to n continuous vertebraes, and has between adjacent sample to be tested There are n-1 overlapping sampled points；

Characteristic extracting module 140 to be measured, for carrying out feature extraction respectively at least one sample to be tested, obtain each The sample characteristics of the individual sample to be tested；

Feature input module 150 to be measured, for the sample characteristics of each sample to be tested to be inputted into training in advance respectively Vertebra demarcation return device, obtain the calibration result of each sample to be tested, the calibration result represent corresponding to treat test sample The classification of each affiliated vertebrae of sampled point in this；

Calibration result determining module 160, for determining the subject according to the calibration result of each sample to be tested Vertebra calibration result.

Vertebra caliberating device in the medical image of the present embodiment, by obtaining the medical image of subject, in medical image In often section vertebrae on determine a sampled point respectively, and treat test sample according to multiple sampled points of determination structure is at least one This, feature extraction is carried out at least one sample to be tested respectively, and then the sample characteristics of each sample to be tested is inputted respectively pre- The vertebra demarcation first trained returns device, to determine the vertebra calibration result of subject according to the calibration result of each sample to be tested, Due to using sampled point structure samples to be tested corresponding to multiple continuous vertebraes, thus can make full use of between adjacent ridge vertebra Correlation, improve the otherness between the identification of sample and sample, strengthen the accuracy and robustness of calibration result.

Figure 10 shows the structure of vertebra caliberating device in medical image in accordance with a further exemplary embodiment of the present invention Block diagram；Wherein, testing image acquisition module 210, sampled point determining module 230 to be measured, sample to be tested structure module 240, to be measured In characteristic extracting module 250, feature input module 260 to be measured and calibration result determining module 270 and embodiment illustrated in fig. 9 Testing image acquisition module 110, sampled point determining module 120 to be measured, sample to be tested structure module 130, feature extraction mould to be measured Block 140, feature input module 150 to be measured and calibration result determining module 160 are identical, herein without repeating.Such as Figure 10 institutes Show, sampled point determining module 230 to be measured can include：

The down-sampled unit 231 of testing image, for carrying out down-sampled processing to the medical image, obtain down-sampled figure Picture；

Sampled point determining unit 232 to be measured, for using the default method of sampling, the often section institute in the down-sampled image State and determine a sampled point on vertebrae respectively.

In an optional embodiment, described device may also include：

Testing image splits module 220, for carrying out image point to the medical image using pre-set image dividing method Cut, extract image-region corresponding to the more piece vertebrae in the medical image；

Correspondingly, sampled point determining module 230 to be measured can be also used for the described image region based on extraction, described in execution The operation for determining a sampled point on the vertebrae respectively is often saved in the medical image.

In an optional embodiment, characteristic extracting module 250 to be measured can be also used at least one treating test sample to described This carries out the feature extraction of a variety of preset kinds respectively, obtains the sample feature set of each sample to be tested；

Correspondingly, feature input module 260 to be measured can be also used for the sample feature set of each sample to be tested The vertebra demarcation for inputting training in advance respectively returns device.

In an optional embodiment, calibration result determining module 270 can include：

Credible sampled point determining unit 271, confidence level highest in the calibration result for determining each sample to be tested Sampled point；

Calibration result projected unit 272, on the basis of the calibration result of the confidence level highest sampled point, order Reckoning obtains the vertebra calibration result of the subject.

In an optional embodiment, credible sampled point determining unit 271 can be also used for：

The gradient mean value of the calibration result of each sample to be tested is calculated, to determine two of the gradient mean value maximum Sample to be tested；

Determine to demarcate end value two sampled points of highest in described two samples to be tested；

According to the comparison of the calibration position difference of described two sampled points and the physical location difference of described two sampled points As a result, judge whether the calibration result of described two samples to be tested is accurate；

When judgement knows that the calibration result of described two samples to be tested is accurate, by calibration result in described two sampled points Value highest sampled point is defined as the confidence level highest sampled point.

In an optional embodiment, described device can also include：

Device training module 280 is returned, for training the vertebra demarcation to return device.

Figure 11 shows the structured flowchart of the recurrence device training module according to one example embodiment of the present invention；This implementation Example is illustrative exemplified by returning the structure of device training module on the basis of above-described embodiment.As described in Figure 11, return Device training module 280 is returned to include：

Sample image acquiring unit 281, for obtaining several sample medical images, in sample medical image described in every width Including more piece sample vertebrae；

Specimen sample point determining unit 283, for the often section sample vertebra in sample medical image described in every width A sampled point is determined on bone respectively；

Training sample construction unit 284, multiple training samples are built for multiple sampled points according to determination；Its In, each training sample includes the sampled point corresponding to n continuous vertebraes；

Training characteristics extraction unit 285, for carrying out feature extraction respectively to the multiple training sample, obtain described more The individual respective sample characteristics of training sample；

Sample label determining unit 286, for determining each self-corresponding label of the multiple training sample, the label is used The classification of each affiliated vertebrae of sampled point in training sample corresponding to expression；

In an optional embodiment, sample label determining unit 286 can be also used for setting for each training sample Put label corresponding to the sparse vector conduct of 1 × m dimension；Wherein, there is the member that continuous n value is 1 in the sparse vector The element that element and m-n value are 0, and the n value is each sampled point institute in training sample described in the positional representation of 1 element Belong to the classification of vertebrae, n is the number of sampled point in each training sample, and m is the vertebrae that the medical image includes Joint number.

Device training unit 287 is returned, for being held according to each self-corresponding label of the multiple training sample and sample characteristics Row is default to return device training algorithm, obtains the vertebra demarcation and returns device.

In an optional embodiment, specimen sample point determining unit 282 can be also used for the sample medical image Down-sampled processing is carried out, obtains down-sampled sample image；And

Using the default method of sampling, determined respectively on the often section sample vertebrae in the down-sampled sample image One sampled point.

In an optional embodiment, returning device training module 280 can also include：

Sample image cutting unit 282, for being entered using pre-set image dividing method to sample medical image described in every width Row image is split, and extracts image-region corresponding to the more piece sample vertebrae in the sample medical image；

Correspondingly, specimen sample point determining unit 283 can be also used for the described image region based on extraction, described in execution The operation for determining a sampled point on the sample vertebrae respectively is often saved in sample medical image described in every width.

In an optional embodiment, training characteristics extraction unit 285 can be also used for the multiple training sample point The feature extraction of a variety of preset kinds is not carried out, obtains the sample feature set of each training sample；

Correspondingly, device training unit 287 is returned to can be also used for according to each self-corresponding label of the multiple training sample And characteristic set performs default recurrence device training algorithm.

For device embodiment, because it corresponds essentially to embodiment of the method, so related part is real referring to method Apply the part explanation of example.Device embodiment described above is only schematical, wherein described be used as separating component The unit of explanation can be or may not be physically separate, can be as the part that unit is shown or can also It is not physical location, you can with positioned at a place, or can also be distributed on multiple NEs.Can be according to reality Need to select some or all of module therein to realize the purpose of the present invention program.Those of ordinary skill in the art are not paying In the case of going out creative work, you can to understand and implement.

The embodiment of vertebra caliberating device can be applied on network devices in medical image of the present invention.Device embodiment can To be realized by software, can also be realized by way of hardware or software and hardware combining.Exemplified by implemented in software, as one Device on logical meaning, it is to be referred to corresponding computer program in nonvolatile memory by the processor of equipment where it Order reads what operation in internal memory was formed.For hardware view, as shown in figure 12, for vertebra mark in the medical image of the present invention A kind of hardware structure diagram of electronic equipment where determining device, except the internal bus 310 shown in Figure 12, memory 320, processor 330 and network interface 340 outside, the equipment in embodiment where device can also generally include other hardware, such as be responsible for place Manage forwarding chip of message etc.；The equipment is also possible to be distributed equipment from hardware configuration, may include multiple Interface card, to carry out the extension of Message processing in hardware view.

The embodiment of the present invention additionally provides a kind of computer-readable recording medium, is stored thereon with computer program, the journey Following task processing method is realized when sequence is processed by the processor：

Compared with prior art, vertebra scaling method in medical image of the invention, by the medical science for obtaining subject Image, a sampled point is determined on the often section vertebrae in medical image respectively, and built according to multiple sampled points of determination At least one sample to be tested, carries out feature extraction at least one sample to be tested respectively, and then by the sample of each sample to be tested The vertebra demarcation that feature inputs training in advance respectively returns device, obtains the matrix of the calibration result composition of all samples to be tested, then The confidence level of test result is determined according to this matrix, it is more due to using so as to extrapolate the final vertebra calibration result of subject Sampled point structure sample to be tested corresponding to individual continuous vertebrae, thus the correlation between adjacent ridge vertebra can be made full use of, The otherness between the identification of sample and sample is improved, strengthens the accuracy and robustness of calibration result.

Those skilled in the art will readily occur to the present invention its after considering specification and putting into practice invention disclosed herein Its embodiment.The application be intended to the present invention any modification, purposes or adaptations, these modifications, purposes or Person's adaptations follow the general principle of the present invention and including undocumented common knowledges in the art of the invention Or conventional techniques.Description and embodiments are considered only as exemplary, and true scope and spirit of the invention are by the application Claim point out.

It should be appreciated that the invention is not limited in the precision architecture for being described above and being shown in the drawings, and And various modifications and changes can be being carried out without departing from the scope.The scope of the present invention is only limited by appended claim.

Claims

A kind of 1. vertebra scaling method in medical image, it is characterised in that including：

The medical image of subject is obtained, the medical image includes the more piece vertebrae of the subject；

A sampled point is determined respectively on the often section vertebrae in the medical image；

At least one sample to be tested is built according to multiple sampled points of determination；Wherein, each sample to be tested includes n The sampled point corresponding to continuous vertebrae, and there are n-1 overlapping sampled points between adjacent sample to be tested；

Feature extraction is carried out respectively at least one sample to be tested, obtains the sample characteristics of each sample to be tested；

The vertebra demarcation that the sample characteristics of each sample to be tested are inputted to training in advance respectively returns device, obtains each described The calibration result of sample to be tested, the calibration result represent corresponding in sample to be tested each affiliated vertebrae of sampled point classification；

The vertebra calibration result of the subject is determined according to the calibration result of each sample to be tested.
2. according to the method for claim 1, it is characterised in that the often section vertebrae in the medical image Upper one sampled point of determination respectively, including：

Down-sampled processing is carried out to the medical image, obtains down-sampled image；

A sampled point is determined respectively on the often section vertebrae in the down-sampled image.
3. according to the method for claim 1, it is characterised in that methods described also includes：

Image segmentation is carried out to the medical image using pre-set image dividing method, extracts the more piece ridge in the medical image Image-region corresponding to vertebra；

Described image region based on extraction, perform and determined respectively on the often section vertebrae in the medical image The operation of one sampled point.
4. according to the method for claim 1, it is characterised in that described that spy is carried out respectively at least one sample to be tested Sign extraction, obtains the sample characteristics of each sample to be tested, including：

Carry out the feature extraction of a variety of preset kinds respectively at least one sample to be tested, obtain each sample to be tested Sample feature set；

The vertebra demarcation that the sample characteristics by each sample to be tested input training in advance respectively returns device, including：

The vertebra demarcation that the sample feature set of each sample to be tested is inputted to training in advance respectively returns device.
5. according to the method for claim 1, it is characterised in that the calibration result according to each sample to be tested is true The vertebra calibration result of the fixed subject, including：

Determine confidence level highest sampled point in the calibration result of each sample to be tested；

On the basis of the calibration result of the confidence level highest sampled point, order, which calculates, obtains the vertebra demarcation of the subject As a result.
6. according to the method for claim 5, it is characterised in that in the calibration result for determining each sample to be tested Confidence level highest sampled point, including：

The gradient mean value of the calibration result of each sample to be tested is calculated, to determine that the gradient mean value maximum two is to be measured Sample；

Determine to demarcate end value two sampled points of highest in described two samples to be tested；

According to the comparison result of the calibration position difference of described two sampled points and the physical location difference of described two sampled points, Judge whether the calibration result of described two samples to be tested is accurate；

If judgement knows that the calibration result of described two samples to be tested is accurate, end value will be demarcated most in described two sampled points High sampled point is defined as the confidence level highest sampled point.
7. according to the method for claim 1, it is characterised in that the vertebra demarcation returns device and trained by following steps Arrive：

Several sample medical images are obtained, include more piece sample vertebrae in sample medical image described in every width；

A sampled point is determined respectively on the often section sample vertebrae in sample medical image described in every width；

Multiple training samples are built according to multiple sampled points of determination；Wherein, each training sample includes n continuously The sampled point corresponding to vertebrae；

Feature extraction is carried out respectively to the multiple training sample, obtains the respective sample characteristics of the multiple training sample；

Each self-corresponding label of the multiple training sample is determined, the label is used to represent respectively to sample in corresponding training sample The classification of vertebrae belonging to point；

Default recurrence device training algorithm is performed according to each self-corresponding label of the multiple training sample and sample characteristics, obtains institute State vertebra demarcation and return device.
8. according to the method for claim 7, it is characterised in that the often section institute in sample medical image described in every width State and determine a sampled point on sample vertebrae respectively, including：

Down-sampled processing is carried out to the sample medical image, obtains down-sampled sample image；

Using the default method of sampling, one is determined respectively on the often section sample vertebrae in the down-sampled sample image Sampled point.
9. according to the method for claim 7, it is characterised in that methods described also includes：

Image segmentation is carried out to sample medical image described in every width using pre-set image dividing method, extracts the sample medical science figure Image-region corresponding to more piece sample vertebrae as in；

Described image region based on extraction, perform the often section sample vertebra in sample medical image described in every width The operation of a sampled point is determined on bone respectively.
10. according to the method for claim 7, it is characterised in that described that feature is carried out respectively to the multiple training sample Extraction, obtains the respective sample characteristics of the multiple training sample, including：

Carry out the feature extraction of a variety of preset kinds respectively to the multiple training sample, obtain the sample of each training sample Eigen set；

It is described that default recurrence device training algorithm is performed according to each self-corresponding label of the multiple training sample and feature, including：

Default recurrence device training algorithm is performed according to each self-corresponding label of the multiple training sample and characteristic set.
11. according to the method for claim 7, it is characterised in that described to determine that the multiple training sample is each self-corresponding Label, including：

Label corresponding to the sparse vector conduct for setting a 1 × m to tie up for each training sample；Wherein, it is described it is sparse to There is the element that continuous n value is 1 and the element that m-n value is 0, and the n value is the positional representation of 1 element in amount The classification of each affiliated vertebrae of sampled point in the training sample, n are the number of sampled point in each training sample, and m is described The joint number for the vertebrae that medical image includes.
A kind of 12. vertebra caliberating device in medical image, it is characterised in that including：

Testing image acquisition module, for obtaining the medical image of subject, the medical image includes the subject More piece vertebrae；

Sampled point determining module to be measured, determine a sampling respectively on the often section vertebrae in the medical image Point；

Sample to be tested builds module, and at least one sample to be tested is built for multiple sampled points according to determination；Wherein, often The individual sample to be tested includes the sampled point corresponding to n continuous vertebraes, and has n-1 between adjacent sample to be tested Overlapping sampled point；

Characteristic extracting module to be measured, for carrying out feature extraction respectively at least one sample to be tested, obtain each described The sample characteristics of sample to be tested；

Feature input module to be measured, for the sample characteristics of each sample to be tested to be inputted to the vertebra mark of training in advance respectively Surely device is returned, the calibration result of each sample to be tested is obtained, is respectively adopted in sample to be tested corresponding to the calibration result expression The classification of the affiliated vertebrae of sampling point；

Calibration result determining module, for determining the vertebra mark of the subject according to the calibration result of each sample to be tested Determine result.
13. device according to claim 12, it is characterised in that the sampled point determining module to be measured includes：

The down-sampled unit of testing image, for carrying out down-sampled processing to the medical image, obtain down-sampled image；

Sampled point determining unit to be measured, for using the default method of sampling, the often section vertebra in the down-sampled image A sampled point is determined on bone respectively.
14. device according to claim 12, it is characterised in that described device also includes：

Testing image splits module, for carrying out image segmentation, extraction to the medical image using pre-set image dividing method Image-region corresponding to more piece vertebrae in the medical image；

The sampled point determining module to be measured is additionally operable to the described image region based on extraction, performs described in the medical image In often save the operation for determining a sampled point on the vertebrae respectively.
15. device according to claim 12, it is characterised in that the characteristic extracting module to be measured be additionally operable to it is described extremely A few sample to be tested carries out the feature extraction of a variety of preset kinds respectively, obtains the sample characteristics collection of each sample to be tested Close；

The feature input module to be measured is additionally operable to the sample feature set of each sample to be tested inputting advance instruction respectively Experienced vertebra demarcation returns device.
16. device according to claim 12, it is characterised in that the calibration result determining module includes：

Credible sampled point determining unit, confidence level highest samples in the calibration result for determining each sample to be tested Point；

Calibration result projected unit, on the basis of the calibration result of the confidence level highest sampled point, sequentially calculating To the vertebra calibration result of the subject.
17. device according to claim 16, it is characterised in that the credible sampled point determining unit is additionally operable to：

The gradient mean value of the calibration result of each sample to be tested is calculated, to determine that the gradient mean value maximum two is to be measured Sample；

Determine to demarcate end value two sampled points of highest in described two samples to be tested；

According to the comparison result of the calibration position difference of described two sampled points and the physical location difference of described two sampled points, Judge whether the calibration result of described two samples to be tested is accurate；

When judgement knows that the calibration result of described two samples to be tested is accurate, end value will be demarcated most in described two sampled points High sampled point is defined as the confidence level highest sampled point.
18. device according to claim 12, it is characterised in that also include：

Device training module is returned, for training the vertebra demarcation to return device；Wherein, the recurrence device training module includes：

Sample image acquiring unit, for obtaining several sample medical images, include in sample medical image described in every width more Save sample vertebrae；

Specimen sample point determining unit, on the often section sample vertebrae in sample medical image described in every width respectively Determine a sampled point；

Training sample construction unit, multiple training samples are built for multiple sampled points according to determination；Wherein, Mei Gesuo State training sample and include the sampled point corresponding to n continuous vertebraes；

Training characteristics extraction unit, for carrying out feature extraction respectively to the multiple training sample, obtain the multiple training The respective sample characteristics of sample；

Sample label determining unit, for determining each self-corresponding label of the multiple training sample, the label is used to represent The classification of each affiliated vertebrae of sampled point in corresponding training sample；

Device training unit is returned, is preset back for being performed according to each self-corresponding label of the multiple training sample and sample characteristics Return device training algorithm, obtain the vertebra demarcation and return device.
19. device according to claim 18, it is characterised in that the specimen sample point determining unit is additionally operable to：

Down-sampled processing is carried out to the sample medical image, obtains down-sampled sample image；

Using the default method of sampling, one is determined respectively on the often section sample vertebrae in the down-sampled sample image Sampled point.
20. device according to claim 18, it is characterised in that the recurrence device training module also includes：

Sample image cutting unit, for carrying out image point to sample medical image described in every width using pre-set image dividing method Cut, extract image-region corresponding to the more piece sample vertebrae in the sample medical image；

The specimen sample point determining unit is additionally operable to the described image region based on extraction, performs described in sample described in every width The operation for determining a sampled point on the sample vertebrae respectively is often saved in medical image.
21. device according to claim 18, it is characterised in that the training characteristics extraction unit is additionally operable to described more Individual training sample carries out the feature extraction of a variety of preset kinds respectively, obtains the sample feature set of each training sample；

The recurrence device training unit is additionally operable to according to each self-corresponding label of the multiple training sample and characteristic set execution It is default to return device training algorithm.
22. device according to claim 18, it is characterised in that the sample label determining unit is additionally operable to as each institute State label corresponding to the sparse vector conduct that training sample sets a 1 × m to tie up；Wherein, there is continuous n in the sparse vector The element that individual value is 1 element and m-n value is 0, and the n value is training sample described in the positional representation of 1 element In each affiliated vertebrae of sampled point classification, n be each training sample in sampled point number, m be the medical image in wrap The joint number of the vertebrae included.
23. a kind of electronic equipment, it is characterised in that the electronic equipment includes：

Processor；

It is configured as storing the memory of processor-executable instruction；

Wherein, the processor is configured as：

The medical image of subject is obtained, the medical image includes the more piece vertebrae of the subject；

A sampled point is determined respectively on the often section vertebrae in the medical image；

At least one sample to be tested is built according to multiple sampled points of determination；Wherein, each sample to be tested includes n The sampled point corresponding to continuous vertebrae, and there are n-1 overlapping sampled points between adjacent sample to be tested；

Feature extraction is carried out respectively at least one sample to be tested, obtains the sample characteristics of each sample to be tested；

The vertebra demarcation that the sample characteristics of each sample to be tested are inputted to training in advance respectively returns device, obtains each described The calibration result of sample to be tested, the calibration result represent corresponding in sample to be tested each affiliated vertebrae of sampled point classification；

The vertebra calibration result of the subject is determined according to the calibration result of each sample to be tested.
24. a kind of computer-readable recording medium, is stored thereon with computer program, it is characterised in that the program is by processor Realized during processing：

The medical image of subject is obtained, the medical image includes the more piece vertebrae of the subject；

A sampled point is determined respectively on the often section vertebrae in the medical image；

At least one sample to be tested is built according to multiple sampled points of determination；Wherein, each sample to be tested includes n The sampled point corresponding to continuous vertebrae, and there are n-1 overlapping sampled points between adjacent sample to be tested；

Feature extraction is carried out respectively at least one sample to be tested, obtains the sample characteristics of each sample to be tested；

The vertebra demarcation that the sample characteristics of each sample to be tested are inputted to training in advance respectively returns device, obtains each described The calibration result of sample to be tested, the calibration result represent corresponding in sample to be tested each affiliated vertebrae of sampled point classification；

The vertebra calibration result of the subject is determined according to the calibration result of each sample to be tested.