CN101623198A - Real-time tracking method for dynamic tumor - Google Patents

Abstract

The invention discloses a real-time tracking method for dynamic tumor, which comprises the following steps: (1) inputting an image sequence and a breathing state feature set of a tumor anatomical structure; (2) establishing a state association database; (3) acquiring real-time images and breathing state features; (4) forecasting a delayed breathing state feature, and determining image sequences of a corresponding tumor form change interval; (5) judging a correlation of tumor motion and breathing motion; (6) registering the real-time images and the image sequences of the tumor form change interval one by one in an interesting area; (7) selecting an optimal tumor form image from the registering results; and (8) outputting the image of the tumor anatomical structure. By quantizing and analyzing the breathing state, the method establishes the correlation of the breathing motion and the tumor motion to forecast the breathing and track the motion so as to improve the real-time tracking efficiency for the dynamic tumor and the tracking accuracy.

Description

Real-time tracking method for dynamic tumor

Technical field

The present invention relates to the method that a kind of image guiding is followed the tracks of, especially relate to a kind of method of image guiding real-time tracking.

Background technology

Stereotactic radiotherapy mainly is divided into 60Co stereotactic radiotherapy and electron linear accelerator stereotactic radiotherapy two big classes.Along with progress of science and technology, the enforcement of precise radiotherapy technology has improved the accuracy and the curative effect of oncotherapy greatly, is the developing direction in whole radiotherapy field.In the radiotherapy, tumor tissues dynamically changes in time, as: the radiotherapy of breast abdomen organ tumor is influenced by respiratory movement largely, and consequent uncertain problem becomes the subject matter that current radiotherapy is faced.Technology such as center displacement control, respiratory training, respiration gate control radiation such as traditional radiotherapy pattern based on conformal intensity modulated technology adopts are usually dealt with the problems referred to above.

(1) center displacement control technology such as: as the Primaton system of Siemens Company, the systematic error of detected motion and pendulum position was revised the position of tumor center when several times were put the position before it utilized, and this method is bigger to normal tissue injury as a kind of approximate treatment means.

(2) in the respiratory training method, can force the patient to bear certain breathing constraint, thereby can't guarantee the stability for the treatment of.

(3) in the respiration gate control radiation technique, only utilize local period treatment of breathing cycle, efficient is very low, and phase when being unable to estimate tumor motion accurate.

The present conformal modulating radiotherapy technology that adopts is treated immobilized tumor and can be reached therapeutic purposes substantially, and be difficult to obtain good result for the treatment of the dynamic tumor at positions such as pulmonary, abdominal part, subject matter is the opportunity that is difficult for grasping gate when adopting traditional gate and the treatment of respiratory training method, and the topical therapeutic period of gate is narrow, influences very much therapeutic efficiency.

The patent No. is US 20060074292, name is called the United States Patent (USP) of " Dynamic tracking of moving targets ", introduce a kind of dynamic tumor and followed the tracks of the method for radiotherapy, this method utilizes three dimensional CT (3D-CT) sequence image of collection and the digital reconstruction radioscopic image (tumor form) of tumor to analyze and the position of record tumor in the breathing cycle before treatment; Carry out in the treatment tumor aspect graph picture and real-time image registration, utilize four-dimensional mathematical model to describe the motion and the deformation of tumor anatomical area, thereby finish the tracking and the real-time positioning of tumor target.The target dose control of any time breathing cycle is by the four-dimensional mathematical model decision that connects reference configuration and anatomical area.

The certain calculated load of needs is followed the tracks of in the deformation of the four-dimensional mathematical model that this patent content comprised, lack numerical procedure to the treatment time-delay, owing to do not consider latency issue, in real-time tracking radiotherapy, with producing bigger tracking error, influence radiocurable effect.

Summary of the invention

Above-mentioned defective at prior art, the invention provides a kind of real-time tracking method for dynamic tumor, dynamic characteristic by the analytical calculation target, and by calculating the motion vector field between real time imaging and prediction dynamic area, estimate the tumor form of time-delay in view of the above, realize the real-time tracking of tumor motion, improve the efficient of dynamic tumor real-time tracking and the accuracy of tracking thereof.

In order to solve above technical problem, the present invention includes following steps:

(1) input: the image sequence I (k) of tumor anatomical structure, breathing state feature set R (k);

(2) set up the state relation data base, comprise: the image sequence I (k) of tumor anatomical structure, tumor aspect graph are as sequence D (k), marker feature collection S (k), breathing state feature set R (k), make that T is length normal period, above-mentioned four kinds of status datas are according to phase sequence k=k for the moment ₁, k ₂... k _TSet up related;

(3) obtain t=t in the treatment _iReal time imaging f (the t of moment dynamic tumor _i) and breathing state feature R (t _i);

(4) by forecast model, the breathing state feature R (t behind the prediction time-delay Δ t _i+ Δ t); And in tumor form image sequence D (k), determine R (t _i+ Δ t) the image sequence D (k in Dui Ying tumor metamorphosis interval _j-α)～D (k _j+ α), k _j∈ 1 ..., K};

(5) differentiate t _iThe time tumor motion and the respiratory movement dependency inscribed;

(6) with real time imaging f (t _i) with the image sequence in tumor metamorphosis interval, in region of interest, carry out registration one by one;

(7) from registration results, select the best tumor aspect graph of a width of cloth as D (k _j+ m), m ∈ [k _j-α, k _j+ α], represent t _i+ Δ t tumor form constantly;

(8) t is described in output _i+ Δ t is the image I (k of tumor anatomical structure constantly _j+ m).

The concrete steps of setting up linked database in the step of the present invention (2) are:

(a) body surface is provided with some labels near patient tumors;

(b) standard is under the breathing cycle: obtain moment k in the breathing cycle ₁, k ₂... k _TThe image sequence I (k) of occlusion body list notation thing and tumor anatomical structure;

(c) obtain corresponding down breathing state feature set R (k) constantly of this cycle, k=k synchronously ₁, k ₂... k _T

(d) according to the tumor anatomical structure image sequence I (k) of this breathing cycle, k=k ₁, k ₂... k _T, calculate each corresponding constantly tumor aspect graph respectively as sequence D (k) by the projection attenuation model of target;

(e) from the tumor aspect graph as the sequence D (k), phase sequence k=k on time ₁, k ₂... k _TExtract body surface label and body internal reference thing respectively by Image Segmentation Model, set up marker feature collection S (k) thus, this feature set has comprised the spatial position data of body surface label and the displacement data of the interior static object of reference of relative body;

(f) according to the time corresponding relation, order is set up the tumor anatomical structure image sequence I (k), tumor aspect graph of this breathing cycle linked database as sequence D (k), marker feature collection S (k), breathing state feature set R (k), and the numerical value by R (k) or S (k) in subsequent applications carries out the retrieval of association status.

Obtain the real time imaging of tumor in the treatment in the step of the present invention (3), concrete steps are:

(a) utilize two cover imaging systems to set up the projection imaging mode that space optical path intersects, every cover imaging system is made of source and detecting plate;

(b) above-mentioned light path intersection region has covered body surface label, tumor motion district, equipment isocenter point;

(c) t _iConstantly, i ∈ 1 ..., N} can be by the tumor projected image f of two cover systems collection ₁(t _i) and f ₂(t _i) calculate tumour 3 D real time imaging f (t _i).

Also comprise (t in the step of the present invention (3) to breathing state feature set R _i), i ∈ 1 ..., and N} carries out the differentiation of state, and concrete steps are:

(a) obtain t=t in the treatment _iReal time imaging f (the t of moment dynamic tumor _i) and breathing state feature R (t _i)

(b) with breathing state feature R (t _i) the input state discrimination model;

(c) according to speed and the half period of direction or first derivative zero crossing or time varying signal and the scope in cycle of breathing time varying signal, judge t _iBreathing state constantly,

If breathing state is normal, execution in step (d) then;

Otherwise, enter next t=t constantly _I+1, and execution in step (a);

(d) carry out marker feature coupling and breathing state prediction

Breathing state feature in the step of the present invention (4) behind the prediction time-delay Δ t is determined the image sequence in tumor metamorphosis interval, and concrete steps are:

(a) with t _iBreathing state feature R (t constantly _i) input breathing forecast model, the breathing state feature R (t behind the prediction time-delay Δ t _i+ Δ t);

(b) by R (t _i+ Δ t) among the retrieving reference data base corresponding tumor aspect graph as D (k _j), j ∈ 1 ... T}, and select suitable irrelevance α, thus generate t _iImage sequence D (the k in+Δ t tumor metamorphosis interval constantly _j-α)～D (k _j+ α);

The concrete steps of differentiating tumor motion and respiratory movement dependency in the step of the present invention (5) are:

(a) set up adaptation function F ₁(), input item are real time imaging f (t _i) marker feature S (t _i) and the tumor aspect graph as the feature set S (k) of the label in the sequence D (k), output item is the similarity measure ρ (S (t of label _i), S (k));

(b) utilize F ₁() carries out marker feature S (t _i) with the quick coupling of marker feature collection S (k); Select maximum similarity measure ρ _Max(k _j), the tumor aspect graph of 1≤j≤T correspondence is as D _Max(k _j) as real time imaging f (t _i) corresponding neoplastic state;

(c) set up similarity function F ₂(), input comprises the f ' (t of region of interest _i) and D ' _Max(k _j) gradation data of image, under least mean-square error constraints, be output as tumor form deviation δ (t in the region of interest _i);

If δ (t _i) meet the requirements, execution in step (7) then,

Otherwise execution in step (3) enters next t=t constantly _I+1Calculating.

Select the best tumor aspect graph of width of cloth picture in the step of the present invention (7) from registration results, concrete steps are:

(a) employing is set up the registration Algorithm model based on the method for characteristics of image, and with tumor volume image f ' (t in the region of interest _i) with the image sequence D (k in tumor metamorphosis interval _j-α)～D (k _j+ each image D ' (k in α) _j+ m), and m=-α ... ,+α, registration one by one;

(b) according to the some registration results that calculate, obtain motion vector field V ' (t respectively _i, m), i ∈ 1 ..., N};

(c) structure energy cost function H (), this function be input as t _I-1The moment and t _iThe motion vector field V ' (t of moment tumor target _I-1, m) and V ' (t _i, m), i ∈ 1 ..., and N}, under motion smoothing and continuity constraint condition, motion continuity on average estimated during this function was output as between dynamic area ${\tilde{E}}_m,m\∈\{-\mathrm{\α},...,+\mathrm{\α}\}.$

(d) from registering images, select maximum measure value

Corresponding tumor aspect graph is as D (k _j+ m '), m ' ∈ α ..., and+α }, as best tumor aspect graph picture.

The image sequence I (k) of described tumor anatomical structure can be the medical image of the clinical multiple modalities of obtaining, and comprising: CT, MRI, PET etc.; About tumor aspect graph picture, mainly refer to calculate the digital reconstruction radioscopic image of tumor according to the projection attenuation model from the imaging geometry principle of mathematics simulation CT; Further, under the three-dimensional reconstruction algorithm, obtain the tumor aspect graph as sequence D (k), (k=k ₁, k ₂... k _T).

T in the described content _iThe time f ' (t that inscribes _i), V ' (t _i) be respectively f (t _i), V (t _i) the region of interest expression way; Same D ' (k _j) be respectively D (k _j) region of interest.Dynamic tumor target coverage in the described region of interest maximum region of tumor target travel in all kinds of images, have unified bulk, can be two dimension or three-dimensional data form under in all kinds of calculation procedures and implementation.

The obtaining means of described breathing state feature R (t) comprises:

(a) displacement signal by the static relatively object of reference of measuring body list notation thing changes, and extracts the breathing state feature;

(b) under the breathing condition, obtain breath signal, extract the breathing state feature by respiration measurement equipment;

The technical solution used in the present invention comprises: (1) carries out the anatomy form of tumor and the accurate estimation and the real-time tracking of locus on the basis of correct decision breathing state, really reaches dynamically suitable shape, transfers strong therapeutic effect in real time; (2) replace traditional time variable space with state space, thus make respiratory movement in the therapeutic process need not to meet specific rule with the time retrain mutually; (3) adopt non-linear respiratory movement trace model to handle local motion and dynamic range variation.Owing to adopted this technical scheme, produced following technique effect:

The first, can handle the tumor of any part and type feature

The present invention can solve under any breathing state, the tumor-localizing of arbitrary shape and form tracking problem.Can be continuously, automatically, complete period ground accurate treatment is subjected to respiratory movement to influence the tumor of big pulmonary/abdominal part, also comprises the immobilized tumor in the treatment any common position of human body.

The second, local control rate is higher, track and localization is more accurate

Reason is: (1) utilizes breath signal to carry out condition discrimination and prediction, compared with utilizing real time imaging and tumor form/3D-CT sequence to carry out the registration/tracking of the motion of tumor merely, reduce the calculating time-delay and then reduce tracking error thereby can greatly reduce calculated load; (2) non-linear respiratory movement trace model is different from the conventional linear estimation model, can estimate and calculate the local details of more doing more physical exercises that takes place, and therefore can reduce the deviation between estimated result and actual result.(3) do not add under any breathing constraints, breathing state may show the situation of very complicated and forecasting inaccuracy certainly, and the condition discrimination model and the state control strategy of the present invention's proposition can provide real-time feedback to therapy system, occurs than mistake avoiding.(4) under the respiratory movement, all shapes and the anatomy structural information of tumor are stored in advance, and in the treatment, any version of tumor can retrieve from previously stored data fast.Therefore, really reached real-time dynamically suitable shape effect.

The 3rd, speed is faster, efficient is higher

Speed advantage of the present invention is embodied in: breathing on the base of prediction, " by n knub position/form constantly; estimate n+1 knub position/form constantly " in fact finished a kind of only at the coupling between local dynamic area in the sequence, thereby simpler and more direct and efficient than adopting real-time X image and tumor form complete sequence coupling in the United States Patent (USP).

Description of drawings

Fig. 1 is a FB(flow block) of the present invention

Fig. 2 is signal and the image acquisition before the treatment of the present invention

Fig. 3 is that target of the present invention and object of reference concern sketch map

Fig. 4 is the perspective view of radiotherapy system of the present invention;

Fig. 5 is the side direction structural representation of radiotherapy system of the present invention;

Fig. 6 is the sketch map that concerns between breath signal and the anatomic form;

Fig. 7 is pulmonary's cross-sectional image of occlusion body list notation thing and tumor;

Fig. 8 is the moving wave shape of body surface label and tumor among Fig. 7;

Fig. 9 is a respiratory movement prediction sketch map

Figure 10 is a respiratory movement condition discrimination sketch map

The mechanical part of 30. multi-diaphragm collimators among the figure, 31. the X ray bulb of the first cover imaging system, 32. the treatment beam of linear accelerator emission, 33. be used for the body surface Mk system track record instrument of breath signal collection and processing, 34. therapeutic bed, 35. the drive system of therapeutic bed, 36. the flat panel detector of the second cover imaging system, the flat panel detector of 37. first cover imaging systems, the X ray bulb of 38. second cover imaging systems, 39. linear accelerator, 40. the body surface label, the isocenter point of 41. treatments, 42. patients, 44. tumor, 50. object of reference skeleton, 110-160. are respectively the picture number of six different conditions in the respiratory, 401. first body surface labels, 402. the second individual list notation thing, 403. the 3rd individual list notation things.

The specific embodiment

The invention will be further described below in conjunction with accompanying drawing.

As Fig. 4, shown in Figure 5, in the radiotherapy system, the mechanical part 30 of multi-diaphragm collimator is connected with linear accelerator 39, mechanical part 30 belows of multi-diaphragm collimator are provided with the therapeutic bed 34 of drive system 35 supports of therapeutic bed, be loaded with patient 42 on the therapeutic bed 34, after the drive system 35 of therapeutic bed receives control instruction, can drive therapeutic bed 34 and carry out multi-dimensional movement, during treatment, patient's 42 intravital tumors are positioned on the isocenter point of treatment 41, be distributed with body surface label 40 near patient's the tumor, the body surface Mk system track record instrument 33 that is used for breath signal collection and processing is positioned at by the therapeutic bed, be connected with patient 42 by sniffer, the X ray bulb 38 of the X ray bulb 31 of the first cover imaging system and the second cover imaging system is positioned at the both sides of the mechanical part 30 of multi-diaphragm collimator, respectively with therapeutic bed 34 belows first the cover imaging system flat panel detector 37 and second the cover imaging system flat panel detector 36 corresponding, the X ray that bulb sends passes intravital tumor of patient and body surface label, received by corresponding flat panel detector, the imaging system of forming by bulb and flat panel detector is obtained the tumor 44 position forms in the real-time image, beam that linear accelerator 39 sends 32 is through the mechanical part 30 of multi-diaphragm collimators, passes on the isocenter point of treatment 41.

The step of implementing the dynamic tumor real-time tracking is as follows:

1. as 10 among Fig. 1: before the treatment, body surface is placed some labels near patient tumors, this label is the higher material of the opaque density of X ray, its effect: the one, by body surface labelling tracking system recording respiration signal, and convert the time series of respiratory characteristic data to; The 2nd, by obtaining anatomical structure image-CT volume images, tumor aspect graph picture-tumor aspect graph picture, breath signal, the marker feature collection of objects of reference such as comprising spinal column, label, structural regime linked database; The 3rd, by the object of reference in the Real Time Image System record therapeutic process, the space situation of change of tumor, to judge the degree of relevancy of tumor motion and breathing.

2. input: the image sequence I (k) of tumor anatomical structure, breathing state feature set R (k);

Shown in 11 among Fig. 1: the image sequence I (k) of anatomical structure;

Before the treatment, a standard is under the breathing cycle: obtain moment k in the breathing cycle ₁, k ₂... k _TThe image sequence I (k) that comprises body surface object of reference and tumor anatomical structure.This data effect has two: one, is used to generate tumor form image sequence-tumor form image sequence, and the 2nd, treatment planning systems positions reference with Rapid Dose Calculation with it.

Shown in 14 among Fig. 1: the breath signal of same period, breathing state feature set R (k);

In this breathing cycle, obtain corresponding breath signal constantly under the same period synchronously, extract breathing state feature set R (k), k=k ₁, k ₂... k _T, the breathing state feature set is used for carrying out differentiation, the dependence state relation storehouse of breathing state and carries out the retrieval of tumor form, breathes prediction and dynamic interval estimation; The obtaining means of breathing state feature R (t) comprises:

(a) displacement signal by the static relatively object of reference of measuring body list notation thing changes, and extracts the breathing state feature;

(b) under the breathing condition, obtain breath signal, extract the breathing state feature by respiration measurement equipment;

3. set up the state relation data base

Shown in 12 among Fig. 1: the tumor aspect graph is as sequence D (k);

According to the tumor anatomical structure image sequence I (k) of this breathing cycle, k=k ₁, k ₂... k _T, calculate each corresponding constantly tumor aspect graph respectively as sequence D (k) by the projection attenuation model of target; This computational process has been utilized existing sophisticated projection, decay, reconstruction model.Tumor aspect graph picture and X ray real time imaging have consistent projection attenuation properties and some diffusion principle, therefore can be used to carry out series of computation operations such as coupling, registration, condition discrimination, estimation in the therapeutic process with real-time image;

Shown in 13 among Fig. 1: marker feature collection S (k);

From the tumor aspect graph as the sequence D (k), phase sequence k=k on time ₁, k ₂... k _TExtract body surface label and body internal reference thing-spinal column respectively by Image Segmentation Model, change in displacement and the conversion of mutual edge distance mutually according to the relative body internal reference of body surface label thing-spinal column, set up marker feature collection S (k), this feature set has comprised the spatial position data of body surface label and the displacement data of the interior static object of reference-spinal column of relative body;

As shown in Figure 3, cut apart and strengthen algorithm according to attenuation model, tumor target, the tumor aspect graph is looked like to handle the image of static object of reference-spinal column, tumor in the occlusion body list notation thing that obtains, the body, make X, Y, Z represent respectively side direction, front and back to and a foot to, then corresponding three t constantly _j(j=l), t _j(j=m), t _j(j=n), subgraph a, b, c are the three-dimensional relationship sketch map; Subgraph a ', b ', the plane relation signal of c ' for obtaining along the Y-axis projection.Produce at the main phase relation analysis of obtaining of the 3D-CT image sequence of breathing cycle according to breath signal.

Shown in 15 among Fig. 1: set up R (k), S (k) and I (k), D (k) state relation data base; Order is set up the tumor anatomical structure image sequence I (k), tumor aspect graph of this breathing cycle as sequence D (k), body surface marker feature collection S (k), breathing state feature set R (k), and according to identical order composition sequence time phase, form linked database, the adjacent data form has the seriality and the flatness characteristics of respiratory variations in the sequence, and the numerical value by breathing state feature set R (k) or feature set S (k) in subsequent applications carries out the retrieval of association status.

As shown in Figure 2, at first, before treatment, utilize high row's number CT scanner to obtain patient's 3D-CT image sequence between each state area in breathing cycle, and the respiratory movement waveshape signal that synchronous recording should breathing cycle lower body list notation thing, set up marker feature collection, respiratory characteristic collection; Secondly,, corresponding CT image transitions is become corresponding tumor form image sequence, the state relation data base of structure breathing/tumor form according to projection attenuation model and reconstruction model.The organ that this data base is reflected, tumor aspect graph picture and anatomical structure image are one to one, do not change with treatment environment and treatment change at interval; Generation at interval is local to be changed and the corresponding relation of marker feature collection S (k), breathing state feature set R (k) and above reference image data can and be treated along with the treatment environment, reason is that the reason of tumor has two kinds: the one, breathe, and the 2nd, pressure changes between the intracorporeal organ.

4. shown in 16 among Fig. 1: obtain real time imaging f (t=t _i); Real-time breathing state feature R (t=t _i) obtain the real time imaging f (t of tumor in the treatment _i) step be:

(a) utilize two cover imaging systems to set up the projection imaging mode that space optical path intersects, every cover imaging system is made up of X ray bulb 31,38 and flat panel detector 36,37;

(b) the bulb intersection region of sending X ray has covered body surface label, tumor motion district, equipment isocenter point;

(c) t _iConstantly, i ∈ 1 ..., N} can be by the tumor projected image f of two cover systems collection ₁(t _i) and f ₂(t _i) calculate tumour 3 D real time imaging f (t _i).

5. shown in 17 among Fig. 1: whether normal, concrete steps are if differentiating breathing state:

(a) with breathing state feature R (t _i) the input state discrimination model;

Under the eupnea state model, can use feature set R (k _j), j=1 ..., T has write down each sampled point characteristic of correspondence vector in a general breathing cycle; If sampling number is T, characteristic vector length is M, and then feature set has been described the breathing state matrix of the T * M size of an object.Each classifies the respiratory characteristic vector as matrix, and its composition is defined as follows:

[status indicator s]: corresponding normal and abnormal condition; [state direction d]: [EX, EOE, IN]; [amplitude of variation r]: the distance between the object of reference; [pace of change v]: the change direction of distance and speed;

[first derivative zero crossing]: the half period of reaction time varying signal and the scope in cycle.

(b) according to speed and the half period of direction or first derivative zero crossing or time varying signal and the scope in cycle of breathing time varying signal, judge t _iBreathing state constantly,

If breathing state is normal, execution in step (6) then;

Otherwise, enter next t=t constantly _I+1, and execution in step (4), the real time imaging and the breathing state feature of gathering next dynamic tumor constantly

T in the treatment _iConstantly, if satisfy the breathing state normal condition, neoplastic state is influenced by respiratory movement merely, and motion of body surface label fully synchronously and have a specific phase contrast near tumor and its; The tumor motion track has certain dependency with respect to body surface label movement locus.If at improper breathing state, as: cough, asthma and anxious state of mind etc., can cause the makings variation of the intravital tumor form of patient, described dependency can variation.By differentiating breathing state, can improve accuracy of predicting.

As Fig. 7, shown in Figure 8, utilize in the real-time image spatial relationship of the dynamically labeled thing 401,402,403 of the respirometric body surface of reaction, knub position form, object of reference skeleton-spinal column, by the moving wave shape of the X ray opaque mark thing that measures; 443 is that tumor 44 is in the terminal position of exhaling; 442 is the position of tumor 44 at air-breathing end; 441 refer to that tumor is with respirometric waveform.

As shown in figure 10, utilize Finite State Model to differentiate the process of breathing state and control tracking, the change of state of breathing is divided into four kinds of patterns, that is: expiration, end-tidal, air-breathing, irregular status, wherein R _Ex, R _Eoe, R _In, R _IrrBe respectively the jump condition of expiration EX, end-tidal EOE, air-breathing IN, irregular status IRR; Each pattern can be described with the respiratory characteristic collection.During condition discrimination, the present invention proposes discrimination model, this model is controlled according to the state transitions condition, and the state transitions condition is according to the parameter distribution range of respiratory characteristic.Condition discrimination has run through the whole process of breathing prediction and tumor tracking, and when the respiratory characteristic that measures belonged to the pattern of a certain definite state, the jump condition of state model worked, and finished from the saltus step of a state to this state.Under the situation that does not satisfy the saltus step condition, then proceed tracking and prediction under the current state.

Need measure the parameter distribution range of the status flag of individual subject under the eupnea condition during differentiation of breathing state, in actual measurement and the computational process, can describe whole respiratory with the feature set that above-mentioned characteristic vector constitutes.According to the body surface label, breathe position and shape that forecast model can the real-time tracking tumor.

6. shown in 21 among Fig. 1,22:, estimate next breathing state feature constantly by forecast model;

Determine the image sequence in tumor metamorphosis interval, concrete steps are:

(a) with t _iBreathing state feature R (t constantly _i) input breathing forecast model, the breathing state feature R (t behind the prediction time-delay Δ t _i+ Δ t); Be used to estimate the region of search of tumor form.

(b) by R (t _i+ Δ t) among the retrieving reference data base corresponding tumor aspect graph as D (k _j), j ∈ 1 ... T}, and select suitable irrelevance α, thus generate t _iImage sequence D (the k in+Δ t tumor metamorphosis interval constantly _j-α)～D (k _j+ α);

Wherein, the tumor aspect graph is as sequence D (k _j), j=1 ..., T, in comprised all neoplastic states of a breathing cycle, when treatment in real time, only be concerned about D (k between the tumor dynamic area that dopes _j-α)～D (k _j+ α).Determine the tumor image D (k at center between dynamic area _j) time: utilize in the sub-piece 21 predicting the outcome in the state relation data base, to retrieve and obtaining of producing.Interval generation is the k in tumor form sequence _jThe place, about respectively increase the range of observation α of equal length.

Adopt nonlinear model can reach when as shown in Figure 9, breathing prediction than conventional linear model tracking effect more accurately.The nonlinear motion algorithm for estimating has adopted maximum posteriori criterion:

P (X/Y)=P (X) P (Y/X), wherein on behalf of real-time respiratory characteristic, Y, X represent the standard respiratory characteristic.Local restriction rule in the probabilistic Modeling process is: define prior-constrained condition P (X) and be the seriality or the similarity of actual respiratory adjacent moment; Definition likelihood constraints P (Y/X) is the status flag dependency of the standard respiratory of current respiratory and observation.The respiratory movement prediction is to carry out continuously according to the order of sampling under each breathing state.Maximize the product of above-mentioned two local probability, can so that results estimated approach to flatness, actual similarity with measurement.On this basis, if estimated result couples together effectively piecemeal (be meant neighbouring sample point place results estimated is coupled together with straight line or curve, or claim line segment or curve fitting), just can describe out to continuous whole the respiratory movement track.The respiratory movement prediction result was evenly distributed in the track after the match by the time.

In the tumor tracing process, the control strategy of breathing prediction and the alternation of condition discrimination process is as follows:

(a) a bit begin to follow the tracks of starting point from air-breathing end Register as initial seed point, Finite State Model record current state classification is EX.

(b) Candidate Set { Y} of definition probability tracking _EXCollect for certain one piece of data in the standard breath signal cycle distributes, the starting point of the collection that distributes has identical amplitude with the current characteristic point of actual breath signal, and the terminal point of the collection that distributes is the first derivative zero crossing place that equal state direction d is arranged with actual breath signal.

(c) according to current point

Characteristic parameter and probability Candidate Set { Y} _EX, with the value of any under the probabilistic model prediction be

If

Still belong to sample set { Y} _EX, the probability of then proceeding under the current state is followed the tracks of.

(d) when estimating t+1 constantly, if ${\tilde{x}}_{t+1}\∉{\left\{Y\right\}}_{\mathrm{EX}}$ And ${\tilde{x}}_{t+1}\∈{\left\{Y\right\}}_{\mathrm{EOE}},$ Then jump to corresponding state EOE, and follow-up probability Candidate Set uses { Y} _EOEIn view of the rule of change of state, if ${\tilde{x}}_{t+1}\∈{\left\{Y\right\}}_{\mathrm{IN}},$ Then jump to IRR.

(e) the state transition logical condition of control procedure is defined as: $R_{\mathrm{ex}}\↔{\tilde{x}}_{t+1}\∈{\left\{Y\right\}}_{\mathrm{ex}};$ $R_{\mathrm{eoe}}\↔{\tilde{x}}_{t+1}\∈{\left\{Y\right\}}_{\mathrm{eoe}};$ $R_{\mathrm{in}}\↔{\tilde{x}}_{t+1}\∈{\left\{Y\right\}}_{\mathrm{in}};$ $R_{\mathrm{irr}}\↔{\tilde{x}}_{t+1}\∉\left\{Y\right|Y\∈\mathrm{EX}|Y\∈\mathrm{EOE}|Y\∈\mathrm{IN}\}.$

By that analogy, utilize above-mentioned strategy to finish and breathe the process control of following the tracks of.

7. differentiate t _iThe time tumor motion and the respiratory movement dependency inscribed, concrete steps are:

(a) set up adaptation function F ₁(), input item are real time imaging f (t _i) marker feature S (t _i) and the tumor aspect graph as the feature set S (k) of the label in the sequence D (k), output item is the similarity measure ρ (S (t of label _i), S (K)).

(b) utilize F ₁() carries out marker feature S (t _i) with the quick coupling of marker feature collection S (k); Select maximum similarity measure ρ _MaxCorresponding tumor aspect graph is as D _Max(k _j) as corresponding to f (t _i) neoplastic state image, the i.e. relevant target image of current time tumor motion state; Shown in 18 among Fig. 1: marker feature S (t _i) and S (k) mate;

(c) obtain real time imaging f (t respectively _i) and the tumor aspect graph as D _Max(k _j) in region of interest data f ' (t _i), D ' _Max(k _j);

(d) set up similarity function F ₂(), input comprises the f ' (t of region of interest _i) and D ' _Max(k _j) gradation data of image, region of interest real time imaging f ' (t _i) and region of interest target image D ' _Max(k _j) by similarity function F ₂() mates, and under least mean-square error constraints, is output as shown in 19 among Fig. 1: tumor form deviation δ (t in the region of interest _i);

(e) shown in 20 among Fig. 1: judge tumor form deviation δ (t _i), if meet the requirements δ (t _i)＜ε, execution in step (8) then, otherwise execution in step (4) enters next t=t constantly _I+1Calculating.Judge tumor form deviation, utilize a threshold parameter ε to estimate the difference of region of interest target, as δ (t _i) think that tumor motion is from eupnea and body internal pressure equilibrium condition during less than threshold value; As δ (t _i) during greater than threshold value, think that the tumor motion part is from situations such as improper breathing and the changes of body internal pressure.

As shown in Figure 6, among the last figure, under the inhomogeneous breathing condition, near the tumor body surface label plurality of continuous in the cycle with respirometric waveform; Middle figure, corresponding to some breathings constantly, the arcuation face CT of the pulmonary image of synchronization gain under the same scan parameter; Figure below, the knub position and the form of the region of interest of corresponding every width of cloth CT image.110～160 are respectively six different states in the respiratory; The form of corresponding label, CT image anatomical structure, region of interest tumor target has been described respectively among the figure of upper, middle and lower under the different moment states.

8. shown in 23 among Fig. 1: real time imaging f (t _i) with the image sequence in tumor metamorphosis interval, in region of interest, carry out registration one by one; At first adopt based on the method for characteristics of image and set up the registration Algorithm model, and with tumor volume image f ' (t in the region of interest _i) with the image sequence D (k in tumor metamorphosis interval _j-α)～D (k _j+ each image D ' (k in α) _j+ m), and m=-α ... ,+α, registration one by one, then, the some registration results according to calculating obtain motion vector field V ' (t respectively _i, m).

9. shown in 24 among Fig. 1: from registration results, select the best tumor aspect graph of a width of cloth as D (k _j+ m), m ∈ [k _j-α, k _j+ α], represent t _i+ Δ t tumor form constantly; At first, structure energy cost function H (), this function be input as t _I-1The moment and t _iThe motion vector field V ' (t of moment tumor target _I-1, m) and V ' (t _i, m), i ∈ 1 ..., and N}, under motion smoothing and continuity constraint condition, motion continuity on average estimated during this function was output as between dynamic area ${\tilde{E}}_m,m\∈\{-\mathrm{\α},...,+\mathrm{\α}\},$ Then, from registering images, select maximum measure value

Corresponding tumor aspect graph is as D (k _j+ m '), m ' ∈ α ..., and+α }, as best tumor aspect graph picture.

10. t is described in output _i+ Δ t is the image I (k of tumor anatomical structure constantly _j+ m ');

Shown in 25 among Fig. 1: as pairing anatomical structure image, calculate required radiation dose distribution according to tumor aspect graph in the linked database; By D (k _j+ m ') Dui Ying breathing state characteristic key goes out corresponding tumor anatomical structure image I (k _j+ m '), implement the calculating of dose distribution with this.

11. shown in 26 among Fig. 1: beam transmission; The selection of irradiation orientation and dosage is according to analyzing about the criterion calculation method of the tissue amount of being subjected in dosiology in the radiotherapy or the radiological medicine and planning that planned outcome is carried out beam transmission by control system.

Claims (10)

1. a real-time tracking method for dynamic tumor is characterized in that, comprises following content and step:

(1) input: the image sequence I (k) of tumor anatomical structure, breathing state feature set R (k);

(2) set up the state relation data base, comprise: the image sequence I (k) of tumor anatomical structure, tumor aspect graph are as sequence D (k), marker feature collection S (k), breathing state feature set R (k), make that T is length normal period, above-mentioned four kinds of status datas are according to phase sequence k=k for the moment ₁, k ₂... k _TSet up related;

(3) obtain t=t in the treatment _iReal time imaging f (the t of moment dynamic tumor _i) and breathing state feature R (t _i);

(4) by forecast model, the breathing state feature R (t behind the prediction time-delay Δ t _i+ Δ t); And in tumor form image sequence D (k), determine R (t _i+ Δ t) the image sequence D (k in Dui Ying tumor metamorphosis interval _j-α)～D (k _j+ α), k _j∈ 1 ..., K};

(5) differentiate t _iThe time tumor motion and the respiratory movement dependency inscribed;

(6) with real time imaging f (t _i) with the image sequence in tumor metamorphosis interval, in region of interest, carry out registration one by one;

(7) from registration results, select the best tumor aspect graph of a width of cloth as D (k _j+ m), m ∈ [k _j-α, k _j+ α], represent t _i+ Δ t tumor form constantly;

(8) t is described in output _i+ Δ t is the image I (k of tumor anatomical structure constantly _j+ m).

2. real-time tracking method for dynamic tumor as claimed in claim 1 is characterized in that, the concrete steps of setting up linked database in the described step (2) are:

(a) body surface is provided with some labels near patient tumors;

(b) standard is under the breathing cycle: obtain moment k in the breathing cycle ₁, k ₂... k _TThe image sequence I (k) of occlusion body list notation thing and tumor anatomical structure;

(c) obtain corresponding down breathing state feature set R (k) constantly of this cycle, k=k synchronously ₁, k ₂... k _T

(d) according to the tumor anatomical structure image sequence I (k) of this breathing cycle, k=k ₁, k ₂... k _T, calculate each corresponding constantly tumor aspect graph respectively as sequence D (k) by the projection attenuation model of target;

(e) from the tumor aspect graph as the sequence D (k), phase sequence k=k on time ₁, k ₂... k _TExtract body surface label and body internal reference thing respectively by Image Segmentation Model, set up marker feature collection S (k) thus, this feature set has comprised the spatial position data of body surface label and the displacement data of the interior static object of reference of relative body;

(f) according to the time corresponding relation, order is set up the tumor anatomical structure image sequence I (k), tumor aspect graph of this breathing cycle linked database as sequence D (k), marker feature collection S (k), breathing state feature set R (k), and the numerical value by R (k) or S (k) in subsequent applications carries out the retrieval of association status.

3. real-time tracking method for dynamic tumor as claimed in claim 1 is characterized in that, obtains the real time imaging of tumor in the treatment in the described step (3), and concrete steps are:

(a) utilize two cover imaging systems to set up the projection imaging mode that space optical path intersects, every cover imaging system is made of source and detecting plate;

(b) above-mentioned light path intersection region has covered body surface label, tumor motion district, equipment isocenter point;

(c) t _iConstantly, i ∈ 1 ..., N} can be by the tumor projected image f of two cover systems collection ₁(t _i) and f ₂(t _i) calculate tumour 3 D real time imaging f (t _i).

4. real-time tracking method for dynamic tumor as claimed in claim 1 is characterized in that, also comprises (the t to breathing state feature set R in the described step (3) _i), i ∈ 1 ..., and N} carries out the differentiation of state, and concrete steps are:

(a) obtain t=t in the treatment _iReal time imaging f (the t of moment dynamic tumor _i) and breathing state feature R (t _i)

(b) with breathing state feature R (t _i) the input state discrimination model;

(c) according to speed and the half period of direction or first derivative zero crossing or time varying signal and the scope in cycle of breathing time varying signal, judge t _iBreathing state constantly,

If breathing state is normal, execution in step (d) then;

Otherwise, enter next t=t constantly _I+1, and execution in step (a);

(d) the normal breathing state feature R (t of output _i).

5. real-time tracking method for dynamic tumor as claimed in claim 1 is characterized in that, predicts the breathing state feature after delaying time in the described step (4), determines the image sequence in tumor metamorphosis interval, and concrete steps are:

(a) with t _iBreathing state feature R (t constantly _i) input breathing forecast model, the breathing state feature R (t behind the prediction time-delay Δ t _i+ Δ t);

(b) by R (t _i+ Δ t) among the retrieving reference data base corresponding tumor aspect graph as D (k _j), j ∈ 1 ... T}, and select suitable irrelevance α, thus generate t _iImage sequence D (the k in+Δ t tumor metamorphosis interval constantly _j-α)～D (k _j+ α).

6. real-time tracking method for dynamic tumor as claimed in claim 1 is characterized in that, the concrete steps of differentiating tumor motion and respiratory movement dependency in the described step (5) are:

(a) set up adaptation function F ₁(), input item are real time imaging f (t _i) marker feature S (t _i) and the tumor aspect graph as the feature set S (k) of the label in the sequence D (k), output item is the similarity measure ρ (S (t of label _i), S (k));

(b) utilize F ₁() carries out marker feature S (t _i) with the quick coupling of marker feature collection S (k); Select maximum similarity measure ρ _Max(k _j), the tumor aspect graph of 1≤j≤T correspondence is as D _Max(k _j) as real time imaging f (t _i) corresponding neoplastic state;

(c) set up similarity function F ₂(), input comprises the f ' (t of region of interest _i) and D ' _Max(k _j) gradation data of image, under least mean-square error constraints, be output as tumor form deviation δ (t in the region of interest _i);

If δ (t _i) meet the requirements, execution in step (7) then,

Otherwise execution in step (3) enters next t=t constantly _I+1Calculating.

7. real-time tracking method for dynamic tumor as claimed in claim 1 is characterized in that, selects the best tumor aspect graph of width of cloth picture in the described step (7) from registration results, and concrete steps are:

(a) employing is set up the registration Algorithm model based on the method for characteristics of image, and with tumor volume image f ' (t in the region of interest _i) with the image sequence D (k in tumor metamorphosis interval _j-α)～D (k _j+ each image D ' (k in α) _j+ m), and m=-α ... ,+α, registration one by one;

(b) according to the some registration results that calculate, obtain motion vector field V ' (t respectively _i, m), i ∈ 1 ..., N};

(c) structure energy cost function H (), this function be input as t _I-1The moment and t _iThe motion vector field V ' (t of moment tumor target _I-1, m) and V ' (t _i, m), i ∈ 1 ..., and N}, under motion smoothing and continuity constraint condition, motion continuity on average estimated during this function was output as between dynamic area

M ∈ α ... ,+α };

(d) from registering images, select maximum measure value

Corresponding tumor aspect graph is as D (k _j+ m '), m ' ∈ α ..., and+α }, as best tumor aspect graph picture.

8. as content as described in claim 1 or the claim 2, wherein, the image sequence I (k) of tumor anatomical structure can be the medical image of the clinical multiple modalities of obtaining, and comprising: CT, MRI, PET etc.; About tumor aspect graph picture, mainly refer to calculate the digital reconstruction radioscopic image of tumor according to the projection attenuation model from the imaging geometry principle of mathematics simulation CT; Further, under the three-dimensional reconstruction algorithm, obtain the tumor aspect graph as sequence D (k), (k=k ₁, k ₂... k _T).

9. real-time tracking method for dynamic tumor as claimed in claim 7 is characterized in that, t in the described content _iThe time f ' (t that inscribes _i), V ' (t _i) be respectively f (t _i), V (t _i) the region of interest expression way; Same D ' (k _j) be respectively D (k _j) region of interest; Dynamic tumor target coverage in the described region of interest maximum region of tumor target travel in all kinds of images, have unified bulk, can be two dimension or three-dimensional data form under in all kinds of calculation procedures and implementation.

10. real-time tracking method for dynamic tumor as claimed in claim 1 is characterized in that, the obtaining means of described breathing state feature R (t) comprises:

(a) displacement signal by the static relatively object of reference of measuring body list notation thing changes, and extracts the breathing state feature;

(b) under the breathing condition, obtain breath signal, extract the breathing state feature by respiration measurement equipment.

Images