A tumor target region extraction system based on magnetic resonance guided radiotherapy

Abstract

The application relates to the field of nuclear magnetic resonance guidance technology and provides a tumor target area extraction system based on magnetic resonance guidance radiotherapy, which comprises a dynamic information acquisition module, an MRI image of a patient in a stable state is acquired before radiotherapy plan execution, and four-dimensional dynamic information of the MRI is formed; a dynamic information positioning module, a tumor target area is outlined from the four-dimensional dynamic information of the MRI, and four-dimensional target area dynamic information of the tumor target area in the stable state is rendered; and an information positioning module, matching point information in different states is extracted from the four-dimensional target area dynamic information.The application does not consider the texture features of the target area boundary, but finds specific marker points from the positions of organs, bones and blood vessels, the positions of the marker points can accurately restore the required radiotherapy boundary, and the required target area range can be quickly and accurately described following the breathing condition of the patient in practice.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of nuclear magnetic resonance guidance, in particular to a tumor target area extraction system based on magnetic resonance guided radiotherapy. BACKGROUND

[0002] The content of this part only provides background information related to the present application, which may not constitute prior art.

[0003] The emergence of the nuclear magnetic accelerator (i.e. magnetic resonance guided linear accelerator, MR-Linac) technology brings great progress to online adaptive radiotherapy. The core of this technology platform is to highly integrate a diagnostic level magnetic resonance imaging (MRI) system and a medical linear accelerator (Linac) in the same rack to form an integrated treatment device. Based on the unique architecture of this nuclear magnetic accelerator, its core advantage can be fully utilized, that is, the powerful "real-time response" capability.

[0004] The nuclear magnetic accelerator (MR-Linac) can monitor the position, shape change of the tumor and the displacement or deformation of the surrounding normal organs (such as bladder filling degree, respiratory motion) in real time before and during the treatment of the patient by using its built-in fast MRI scanning sequence. The system then optimizes the treatment plan online based on the latest anatomical structure information through the registration assisted by the technician. Finally, the linear accelerator integrated in the same device can real-time irradiate the accurately calculated dose to the moving or deformed target area. This closed-loop process aims to ensure that the tumor receives sufficient dose while the critical organs are protected to the maximum extent, thereby significantly improving the accuracy and safety of radiotherapy.

[0005] The nuclear magnetic accelerator (MR-Linac) provides unprecedented real-time image guided capability, but when the physicist adjusts the radiotherapy plan according to physiological activities such as respiratory motion, a fixed size of the radiation target area is generally set, and the radioactive rays are released at the right time according to the patient's respiratory cycle. This radiotherapy method has certain delay effect in practice, and the radiation target area cannot track the displacement and expansion of the tumor. In order to pursue better radiotherapy effect, the irradiation area needs to be expanded, so that the irradiation area cannot focus on the actual tumor target area, affecting the accuracy of radiotherapy. SUMMARY

[0006] Therefore, the purpose of the present application is to provide a tumor target area extraction system based on magnetic resonance guided radiotherapy. The tumor target area extraction system based on magnetic resonance guided radiotherapy disclosed in the present application can achieve.

[0007] The purpose of the present application is achieved by the following technical solutions:

[0008] A tumor target area extraction system based on magnetic resonance guided radiotherapy, comprising:

[0009] a dynamic information acquisition module, acquiring MRI images of a patient in a stable state before radiotherapy plan execution to form MRI four-dimensional dynamic information;

[0010] a dynamic information positioning module, outlining a tumor target region from the MRI four-dimensional dynamic information and rendering four-dimensional target region dynamic information of the tumor target region in the stable state;

[0011] an information positioning module, extracting matching point information in different states from the four-dimensional target region dynamic information;

[0012] a target region tracking module, extracting feature point information of a current state from MRI images during radiotherapy plan execution, matching the feature point information with the matching point information, and generating real-time target region boundaries in the current time based on a positional relationship between the matching point information and the target region boundaries;

[0013] wherein the matching points at least include bone points, organ points and blood vessel points for representing human body positions; and the state is a gas retention amount of a patient's lung.

[0014] The core advantage of the scheme lies in its unique target region boundary determination method: by extracting matching points and feature points with clear anatomical meanings from MRI images (rather than directly relying on blurred image textures), and analyzing the spatial positional relationship between these points to deduce the target region boundaries, the key problem of CTV (clinical target volume) boundary ambiguity in clinical practice is effectively solved, and thus the radiotherapy rays can be tracked along these boundaries by accurately acquiring the patient's breathing cycle during radiotherapy.

[0015] The existing radiotherapy positioning method based on MRI images cannot effectively distinguish voxels with the same spatial coordinates (x, y, z) but changed actual anatomical positions in different breathing phases during patient breathing movement. This dynamic position ambiguity makes it extremely difficult to accurately identify and track the target region boundaries (especially the ambiguous extended region of the clinical target volume CTV) that change with breathing, and it is severely dependent on manual experience and inefficient, resulting in inaccurate dynamic target region outlining and affecting radiotherapy precision.

[0016] In some possible embodiments, the MRI four-dimensional dynamic information P{x, y, z, v}, wherein x, y and z represent the horizontal coordinate, vertical coordinate and height coordinate of a voxel in the MRI image, v represents the gas retention amount of the patient's lung, and P represents the voxel value;

[0017] The four-dimensional target region dynamic information PE{x, y, z, v, b}, wherein b represents whether it is a target region boundary, b=1 represents that the current voxel is a target region boundary, b=0 represents that the current voxel is not a target region boundary, and PE represents the voxel value.

[0018] The application introduces lung gas retention amount (v) as the fourth dimension into MRI dynamic information (forming P{x, y, z, v}) innovatively, and constructs four-dimensional target area dynamic information (PE{x, y, z, v, b}) containing boundary identification (b), which fundamentally solves the problem of dynamic position ambiguity caused by respiratory motion. The core is that when the four-dimensional dynamic information (x, y, z, v) of two voxels is exactly the same, it can be uniquely determined that they belong to the same anatomical position in the respiratory cycle. This provides a reliable data basis and key discriminant basis for accurately and automatically identifying and tracking the spatial position change of the target area boundary (especially the CTV boundary) in the dynamic respiratory process, and significantly improves the accuracy and efficiency of dynamic target area positioning.

[0019] In some possible embodiments, the target area boundary in the four-dimensional target area dynamic information is manually marked.

[0020] The existing radiotherapy target area positioning method based on feature point matching is mainly limited by two factors: one is that the feature points (such as the extreme points of organs, bones, and blood vessels) relied on matching have single information dimension (usually only contain spatial position) and lack dynamic spatial correlation information with the target area boundary; the other is that the dynamic change rule of the relative position relationship between the feature points and the target area boundary under physiological activities such as respiratory motion is ignored. This leads to the fact that the matching result of the feature points cannot accurately reflect the real displacement of the target area boundary, and especially when the respiratory causes tissue deformation, the positioning accuracy significantly decreases.

[0021] In some possible embodiments, the matching point information acquisition method comprises the following steps:

[0022] Step 1: decompose the four-dimensional target area dynamic information PE into different state MRI image information groups;

[0023] Step 2: for each MRI image information group, select the target area boundary, part of the organ boundary, part of the bone boundary, and part of the blood vessel boundary therefrom;

[0024] Step 3: take the voxels of the target area boundary as the target extraction set, extract a plurality of extreme points from the part of the organ boundary, the part of the bone boundary, and the part of the blood vessel boundary respectively to generate a first extreme point set, a second extreme point set, and a third extreme point set;

[0025] Step 4: take the relative position information of each extreme point in the first extreme point set and each pixel point on the target area boundary as the first relative information;

[0026] Take the relative position information of each extreme point in the second extreme point set and each pixel point on the target area boundary as the second relative information;

[0027] the relative position information between each extreme value point in the third extreme value point set and each pixel point on the target region boundary as third relative information;

[0028] Step 5: taking the first relative information, the second relative information, and the third relative information as matching point information.

[0029] The present application significantly improves the accuracy and robustness of feature point matching by innovatively constructing "matching point information" containing multi-dimensional relative position relationships. The core lies in not only extracting the boundary extreme points of multiple types of anatomical structures (organs, bones, blood vessels), but more importantly, calculating and integrating the dynamic "relative position information" (first, second, and third relative information) between these extreme points and each pixel point on the target region boundary. This multi-dimensional information that combines the feature points themselves and their spatial correlation with the target region can accurately depict and quantify the dynamic change relationship between the human tissues represented by the feature points and the target region boundary during the respiratory process, such as expansion and displacement. Therefore, in practice, it can more accurately describe and predict the target region displacement caused by respiratory motion, providing a reliable basis for high-precision, adaptive dynamic target region positioning.

[0030] Existing radiotherapy target region tracking methods based on fixed feature point spatial position matching are difficult to effectively cope with the non-rigid and random deformation of internal tissues caused by respiratory motion. Relying solely on feature point pixel coordinates for rigid matching cannot quantify and integrate the randomness of tissue changes, resulting in deviations in the mapping relationship established during the respiratory dynamic process and actual anatomical displacement, ultimately leading to insufficient accuracy in target region tracking positioning.

[0031] Further, the target region tracking module comprises:

[0032] a feature point information extraction unit for extracting feature point information in an MRI image during radiotherapy plan execution;

[0033] a matching information unit for matching the feature point information with the matching point information, so that the feature points in the feature point information and the matching points establish a mapping relationship;

[0034] a target region tracking unit for delineating a dynamic target region in the MRI image during radiotherapy plan execution based on the correspondence between the feature points and the matching points;

[0035] wherein the mapping relationship between the feature points and the matching points includes the probability of each feature point corresponding to each matching point; and the target region tracking unit generates at least three different levels of target region boundaries based on the mapping relationship between the feature points and the matching points.

[0036] The present application significantly improves the robustness and accuracy of target region tracking under respiratory dynamics by introducing a probability cloud matching mechanism. The core is to abandon the traditional deterministic point-to-point matching and instead establish a probability mapping relationship between feature points and matching points (i.e. the possibility distribution of each matching point corresponding to each feature point). This probability density-based matching method can effectively model and adapt to the inherent randomness of tissue deformation in the respiratory process. Based on this probability mapping relationship, the system can generate multiple target region boundaries with different confidence levels (at least 3 levels), providing more realistic anatomical uncertainty for dynamic target region range selection in clinical practice, and achieving more accurate and flexible target region tracking under complex physiological changes.

[0037] In existing dynamic tracking technology of radiotherapy target region, in order to improve the matching accuracy, a large number of feature points need to be extracted in the body, which not only significantly increases the computational complexity of feature point extraction and subsequent matching process, but also introduces a large number of non-key or poor stability feature points. These redundant points not only consume valuable computing resources and prolong processing time, but also may interfere with matching accuracy due to the inclusion of noise or variable information, making it difficult to meet the dual demands of efficiency and reliability for real-time adaptive radiotherapy.

[0038] Further, the extraction method of feature point information includes the following steps:

[0039] S1: Extracting the MRI image according to the state to decompose the MRI image information extraction group under different states during radiotherapy plan execution;

[0040] S2: Corresponding the MRI image information extraction group under different states with the MRI image information group under different states;

[0041] S3: Extracting the extreme points near the boundaries of the viscera, part of the skeleton and part of the blood vessels from the MRI image information extraction group;

[0042] S4: Taking the extreme points of the viscera boundary as the first extraction set, the extreme points of the skeleton boundary as the second extraction set, and the extreme points of the blood vessel boundary as the third extraction set;

[0043] S5: Taking the first extraction set, the second extraction set and the third extraction set as the feature point information.

[0044] The present application realizes intelligent screening and simplification of feature points by focusing on specific regions (organs, bones, blood vessels) with stable anatomical structure and clear boundaries to extract feature points (first, second and third extraction sets). The present scheme can significantly reduce the number of feature points required to be extracted while ensuring the representativeness and stability of the feature points. This not only greatly reduces the computational overhead of feature point information storage and processing, improves system efficiency, but more importantly, by excluding a large amount of irrelevant or variable tissue information, it improves the quality and reliability of the feature point set for matching, laying a solid foundation for subsequent efficient and accurate feature point matching, thereby optimizing the performance of the entire dynamic target area tracking process.

[0045] In existing schemes, when determining the position of a dynamic target area, the accurate dynamic target area is generally obtained based on the extracted information, such as texture information. This target area acquisition method requires a large amount of computational resources to learn the difference between normal tissue texture and tumor texture. Moreover, it cannot accurately identify the actual difference between tumor and normal tissue infiltration areas, so it cannot obtain an accurate radiotherapy target area.

[0046] The matching information unit matches the feature point information and the matching point information based on the following scheme:

[0047] Z1: combining the first extraction set with the first relative information, combining the second extraction set with the second relative information, and combining the third extraction set with the third relative information to form three matching groups;

[0048] Z2: for each matching point in the same matching group, finding the feature point with the highest similarity to generate a matching relationship, and in a matching relationship, taking the similarity of the feature point and the matching point as the first matching feature, and taking the number of feature points around the feature point that are not matched with the matching point as the second matching feature. The feature points around the feature point are a preset window of 3 voxel grids.

[0049] Z3: establishing a probability matching model P based on the matching features of the three matching groups;

[0050] P={p1, p2, …, pn} represents the position of the i-th pixel point on the dynamic target area. i …p z}, p i represents the position of the i-th pixel point on the dynamic target area.

[0051] P i =p i (z 1,i , z 2,i , z 3,i , x 1,i , x 2,i , x 3,i );

[0052] z1,i , z 2,i , z 3,i are respectively first matching features of three matching groups, x 1,i , x 2,i , x 3,i are respectively second matching features of three matching groups.

[0053] In the technical scheme provided in the application, when the dynamic target area is calculated, the calculation is not based on texture information, but based on the similarity between the matching points and the feature points, the calculation resource is smaller than the calculation resource consumed by texture calculation, and the reaction efficiency can be faster. Moreover, the tumor target area is obtained according to the relative position relationship between the matching points and the feature points, that is, the target area is determined according to the displacement of the human organ. When the patient inhales air, the lung will expand, and the corresponding lung boundary will also expand. Thus, the displacement of the tumor will adapt to the expansion of the lung, and a relatively accurate tumor target area can be obtained. At the same time, when the position of the dynamic target area is calculated, the dynamic probability representation method is used, which can more accurately capture the activity of the tumor target area in the dynamic time dimension of the patient. In the scheme, in addition to the similarity between the feature points and the matching points, the probability density of the feature points is also introduced, that is, the number of feature points around the feature points is large, and the feature points are not matched with the matching points. It is indicated that the organ boundary appears large deformation due to the patient's breathing, and only scattered feature points can be successfully matched. Therefore, the confidence of such feature points needs to be reduced.

[0054] The extraction method of the feature points and the matching points is as follows:

[0055] S01: Extracting extreme points from the MRI image based on the SIFT algorithm.

[0056] The SIFT algorithm is prior art, and how to extract extreme points from the MRI image will not be described here.

[0057] S02: Setting a standard window, traversing the nearby area of the reference object in the MRI image, when the number of extreme points in the standard window and the extreme point density exceed the preset threshold, the area corresponding to the standard window is taken as an extraction window, and all extreme points in the extraction window are extracted as feature points or matching points.

[0058] The SIFT algorithm is used in the scheme to increase the accuracy of the extraction of the feature points and the matching points, and to reduce the matching difficulty of the feature points and the matching points.

[0059] Further, the target area tracking unit generates a dynamic target area based on the following steps:

[0060] T1: Obtain a probability matching model P, P={p1, p2, …, p i …p z}, pi represents the position of the i-th pixel point on the dynamic target area; P i = p i (z 1,i , z 2,i , z 3,i , x 1,i , x 2,i , x 3,i );

[0061] T2: based on the probability matching model P, the matching points corresponding to the feature points are analyzed, and according to the relative position information of the matching points and the pixel points on the target area boundary, the position y i of each pixel point on the dynamic target area is obtained.

[0062] wherein, , wherein (x1, y1) i represents the position of the pixel point y i determined by the first matching group, q1 represents the probability of the position of the pixel point y i determined by the first matching group; (x2, y2) i represents the position of the pixel point y i determined by the second matching group, q2 represents the probability of the position of the pixel point y i determined by the second matching group; (x3, y3) i represents the position of the pixel point y i determined by the third matching group, q3 represents the probability of the position of the pixel point y i determined by the third matching group. , , ;

[0063] T3: analyzing the discrete variable of the position y i of each pixel point on the dynamic target area to generate three dynamic target areas of different ranges.

[0064] In the technical solution used in the present application, all possible positions of the pixel point y i are collected together to form a dynamic target area range, and whether each point in the range is a target area boundary is described by probability, so the determination of the dynamic target area boundary is described by probability state, and the accuracy is higher when screening the specific dynamic target area boundary.

[0065] Further, T3 includes the following steps:

[0066] T31: obtaining the discrete variable y i to obtain the range point set N of the dynamic target area;

[0067] T32: setting a minimum probability threshold in advance, and deleting the pixel point yi obtain a dynamic target region distribution scatter plot;

[0068] T33: taking the outermost boundary of the dynamic target region distribution scatter plot as a high-level target region boundary, taking the innermost side of the dynamic target region distribution scatter plot as a low-level target region boundary, and taking a path with equal distance from the high-level target region boundary and the low-level target region boundary as an intermediate-level target region boundary, the high-level target region boundary, the low-level target region boundary and the intermediate-level target region boundary being three dynamic target regions of different ranges.

[0069] In the technical scheme provided in the application, the selection error of the dynamic target region can be reduced by deleting the pixel points with the deletion probability less than the probability threshold. Meanwhile, the three levels of target region boundaries selected are relatively more flexible, and a physicist can select a suitable target region boundary to track the tumor of a patient according to a treatment tendency.

[0070] The technical scheme of the embodiment of the application has at least the following advantages and beneficial effects:

[0071] (1) Realize accurate positioning of respiratory phase and eliminate dynamic ambiguity of position: by innovatively introducing the lung gas retention amount (v) as the fourth dimension into the MRI dynamic information (forming P{x, y, z, v} and PE{x, y, z, v, b}), the scheme fundamentally solves the dynamic ambiguity problem of "different anatomical positions at the same spatial coordinates" caused by respiratory motion.

[0072] (2) Enhance the robustness of matching of deformed tissues and improve the accuracy of target region tracking: the scheme breaks through the limitation of traditional single dependence on the similarity of feature points and innovatively integrates the spatial relationship information (including direction difference and distance ratio) between the feature points and stable anatomical references (such as bones and blood vessels). The probability model constructed by this multi-dimensional matching strategy (similarity + direction difference + distance ratio) can effectively depict the dynamic displacement relationship between tissues under non-rigid deformation caused by respiration, significantly improving the discrimination and anti-interference ability of feature point matching. The effect is that even when the tissues are randomly deformed, the mapping relationship of the target region boundary can be stably and accurately established, greatly improving the reliability of dynamic target region tracking. BRIEF DESCRIPTION OF DRAWINGS

[0073] Figure 1 FIG. 1 is a structural schematic diagram of a tumor target region extraction system based on magnetic resonance guided radiotherapy. DETAILED DESCRIPTION

[0074] For the purposes of the present application, the technical solutions and advantages thereof will be more clearly understood from the following detailed description of the embodiments, taken in conjunction with the accompanying drawings. The same reference numbers in different drawings represent the same or similar elements. It should be noted that the described embodiments are part of the present application, but not all embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0075] Compared with the embodiments shown in the drawings, the feasible implementation solutions within the protection scope of the present application can have fewer components, other components not shown in the drawings, different components, differently arranged components, or differently connected components, etc. In addition, two or more components in the drawings can be implemented in a single component, or a single component shown in the drawings can be implemented as a plurality of separate components.

[0076] Unless otherwise defined, technical terms or scientific terms used herein should be understood as having the common meaning in the field of the present application to those having ordinary skill in the art. The terms "first", "second", and similar terms used in the specification and claims of the present application do not necessarily indicate any order, number, or importance, but are only used to distinguish different components. Similarly, "one" or "a" and similar terms do not necessarily indicate a quantity limitation. "Up", "down", and the like are only used to indicate relative positional relationships, which may change accordingly when the absolute position of the described object changes.

[0077] The present application discloses a tumor target area extraction system based on magnetic resonance guided radiotherapy, comprising: a dynamic information acquisition module, acquiring MRI images of a patient in a stable state before radiotherapy plan execution, to constitute MRI four-dimensional dynamic information; a dynamic information positioning module, outlining a tumor target area from the MRI four-dimensional dynamic information, and rendering four-dimensional target area dynamic information of the tumor target area in the stable state; an information positioning module, extracting matching point information in different states from the four-dimensional target area dynamic information; a target area tracking module, extracting feature point information of the current state from the MRI images during radiotherapy plan execution, matching the feature point information with the matching point information, generating real-time target area boundaries in the current time based on the positional relationship between the matching point information and the target area boundaries; wherein the matching points at least include skeletal points, organ points, and blood vessel points for representing human body positions; and the state is the gas retention amount of the patient's lung.

[0078] The MRI four-dimensional dynamic information P{x, y, z, v}, wherein x, y, and z represent the horizontal coordinate, vertical coordinate, and height coordinate of the voxel in the MRI image, v represents the gas retention amount of the patient's lung, and P represents the voxel value;

[0079] Four-dimensional target region dynamic information PE{x, y, z, v, b}, wherein b represents whether it is a target region boundary, b=1 represents that the current voxel is a target region boundary, and b=0 represents that the current voxel is not a target region boundary. The target region boundary in the four-dimensional target region dynamic information is manually marked.

[0080] The dynamic information acquisition module first acquires multiple frames of MRI images of the patient in a stable breathing period before radiotherapy plan execution, integrates the lung gas retention amount (v) as the fourth dimension, and generates time and space correlated MRI four-dimensional dynamic information P{x, y, z, v};

[0081] The dynamic information positioning module receives the four-dimensional information, and a physician draws a tumor target region contour thereon and renders four-dimensional target region dynamic information PE{x, y, z, v, b} fused with a boundary identifier (b). The information positioning module analyzes the PE information, divides the breathing state according to the gas retention amount (v), extracts matching point information (including three types of anatomical markers of bone points, organ points, and blood vessel points and the spatial relationship thereof with the target region) from the target region data of each state. The target region tracking module acquires a current MRI image in real time during radiotherapy execution: a feature point information extraction unit identifies bone / organ / blood vessel feature points from the current image; a matching information unit performs probabilistic mapping matching of the feature points and the pre-stored matching point information; and a target region tracking unit reversely calculates and generates a real-time target region boundary according to the positional relationship between the matching points and the target region boundary.

[0082] Therefore, the most critical part is how to extract matching points and feature points and make one-to-one mapping of the matching points and the feature points. Specifically:

[0083] The matching point information acquisition method includes the following steps:

[0084] Step 1: decompose the four-dimensional target region dynamic information PE into MRI image information groups in different states.

[0085] Discretize the four-dimensional target region dynamic information PE{x, y, z, v, b} according to the values of the lung gas retention amount (v) to classify the states. Specifically, PE represents a voxel value:

[0086] Set the breathing state division threshold (for example, v≤30% is the end-expiratory state, 30%<v≤70% is the mid-breathing state, and v>70% is the end-inspiratory state). The measurement scheme of the lung gas retention amount is a prior art, and the lung gas retention amount can be calculated according to the volume difference between the inhaled gas amount and the exhaled gas amount of the patient.

[0087] Classify all voxel data in PE that satisfy the same state threshold condition to generate an MRI image information group corresponding to the state. Each group of data contains the spatial coordinates (x, y, z) and the boundary identifier (b) of all voxels in the state.

[0088] Step 2: For each MRI image information set, the target region boundary, the partial organ boundary, the partial bone boundary and the partial blood vessel boundary are marked out; the target region boundary is the tumor target region boundary.

[0089] The automatic boundary extraction is performed on each state MRI image information set:

[0090] Target region boundary: the continuous voxel set of b=1 is extracted;

[0091] Organ boundary: including kidney, liver;

[0092] Bone boundary: including rib, spine;

[0093] Blood vessel boundary: including main pulmonary artery, vena cava.

[0094] The organ boundary, the bone boundary and the blood vessel boundary are automatically extracted by the MRI system, and the specific manner is not described here.

[0095] Step 3: The voxels of the target region boundary are taken as the target extraction set, and a plurality of extreme points are extracted from the partial organ boundary, the partial bone boundary and the partial blood vessel boundary respectively to generate a first extreme point set, a second extreme point set and a third extreme point set.

[0096] Step 4: The relative position information of each extreme point in the first extreme point set and each pixel point on the target region boundary is taken as the first relative information; the relative position information of each extreme point in the second extreme point set and each pixel point on the target region boundary is taken as the second relative information; the relative position information of each extreme point in the third extreme point set and each pixel point on the target region boundary is taken as the third relative information.

[0097] Step 5: The first relative information, the second relative information and the third relative information are taken as the matching point information.

[0098] Further, the target region tracking module comprises:

[0099] A feature point information extraction unit is configured to extract feature point information in the MRI image during the radiotherapy plan execution;

[0100] A matching information unit is configured to match the feature point information and the matching point information, so that the feature points in the feature point information and the matching points establish a mapping relationship;

[0101] A target region tracking unit is configured to delineate a dynamic target region in the MRI image during the radiotherapy plan execution based on the corresponding relationship between the feature points and the matching points.

[0102] The mapping relationship between the feature points and the matching points includes probabilities of mutual correspondence between the feature points and the matching points; and the target region tracking unit generates at least three target region boundaries of different levels based on the mapping relationship between the feature points and the matching points.

[0103] The target region tracking module, when the radiotherapy plan is executed, first extracts feature point information of key anatomical structures from real-time collected MRI images through the feature point information extraction unit; then the matching information unit probabilistically matches the extracted feature point information with a pre-set matching point information library to establish a mapping relationship between the feature points and the matching points (the relationship represents the possibility of the feature points belonging to a specific matching point in the form of a probability weight); and finally, the target region tracking unit dynamically generates at least three target region contours of different confidence levels (such as a high-confidence core target region, a medium-confidence extended target region, and a low-confidence risk target region) in the MRI images based on the probability mapping relationship, thereby realizing multi-level accurate tracking and adaptive delineation of the moving target region.

[0104] The extraction method of the feature point information includes the following steps:

[0105] S1: Extracting MRI images during radiotherapy plan execution into MRI image information extraction groups in different states.

[0106] The feature point information extraction first performs a state decomposition operation: the continuous MRI image stream collected in real time during the radiotherapy plan execution process is divided into multiple independent MRI image information extraction groups according to the organ motion state (such as end of inspiration, mid-exhalation, and resting period). Each group corresponds to a spatial anatomical structure snapshot under a specific physiological state.

[0107] S2: Corresponding the MRI image information extraction groups in different states to the MRI image information groups in different states.

[0108] Establishing state synchronization mapping: performing spatiotemporal registration between the MRI image information extraction groups (i.e., the decomposed real-time images) generated in S1 and the pre-stored reference state MRI image information group (the same state image of the same patient collected during the radiotherapy plan formulation) to ensure that the two groups of data are strictly aligned under the same motion state. This step realizes pixel-level spatial correspondence through rigid / non-rigid transformation matrix, eliminating the interference of body position difference.

[0109] S3: Extracting extreme points near the boundaries of the viscera, part of the boundaries of the skeleton, and part of the boundaries of the blood vessels from the MRI image information extraction groups;

[0110] S4: Taking the extreme points of the boundaries of the viscera as a first extraction set, taking the extreme points of the boundaries of the skeleton as a second extraction set, and taking the extreme points of the boundaries of the blood vessels as a third extraction set;

[0111] The first extraction set: the organ boundary extreme point set, representing the target organ deformation characteristics;

[0112] The second extraction set: the skeleton boundary extreme point set, providing a rigid position reference;

[0113] The third extraction set: the blood vessel boundary extreme point set, reflecting the micro-displacement conduction path.

[0114] S5: taking the first extraction set, the second extraction set and the third extraction set as the feature point information.

[0115] Generating fusion feature point information: combining the first (organ), second (skeleton) and third (blood vessel) extraction sets into a unified spatial feature point data set. The data set contains the three-dimensional coordinates of each point, the type of the anatomical structure to which the point belongs and the boundary curvature value, serving as the input feature source of the target region tracking module.

[0116] The above gives the matching method after the extraction of the feature points and the matching points. The application provides an extraction method of the matching points and the feature points according to actual needs. Specifically:

[0117] The extraction method of the feature points and the matching points is as follows:

[0118] S01: extracting extreme points from the MRI image based on the SIFT algorithm.

[0119] The SIFT algorithm is a prior art, and how to extract extreme points from the MRI image will not be described here.

[0120] S02: setting a standard window, traversing the nearby area of the reference object in the MRI image, and when the number of extreme points and the density of extreme points in the standard window exceed a preset threshold, taking the area corresponding to the standard window as an extraction window and the extreme points in the extraction window as the matching points or the feature points.

[0121] The matching information unit matches the feature point information and the matching point information based on the following scheme:

[0122] Z1: combining the first extraction set with the first relative information, combining the second extraction set with the second relative information and combining the third extraction set with the third relative information to form three matching groups;

[0123] Z2: for each matching point in the same matching group, finding the feature point with the highest similarity in turn to generate a matching relationship, and taking the similarity of the feature point and the matching point as the first matching feature and the number of feature points around the feature point that are not matched with the matching point as the second matching feature in a matching relationship. The feature points around the feature point are a preset window of three voxel grids.

[0124] Z3: establishing a probability matching model P based on the matching features of the three matching groups.

[0125] P={p1, p2, …, p i …p z},p i represents the position of the i-th pixel point on the dynamic target area;

[0126] P i =p i (z 1,i , z 2,i , z 3,i , x 1,i , x 2,i , x 3,i );

[0127] z 1,i , z 2,i , z 3,i are the first matching features of the three matching groups, respectively, and x 1,i , x 2,i , x 3,i are the second matching features of the three matching groups, respectively.

[0128] The target area tracking unit generates the dynamic target area based on the following steps:

[0129] T1: Obtain the probability matching model P, P={p1, p2, …, p i …p z},p i represents the position of the i-th pixel point on the dynamic target area; P i =p i (z 1,i , z 2,i , z 3,i , x 1,i , x 2,i , x 3,i );

[0130] T2: Based on the probability matching model P, analyze the matching points corresponding to the feature points, and according to the relative position information of the matching points and the pixel points on the target area boundary, obtain the discrete variable of the position y i of each pixel point on the dynamic target area;

[0131] wherein, wherein, (x1, y1) i represents the position of the pixel point y i determined by the first matching group, q1 represents the probability of the position of the pixel point y i determined by the first matching group; (x2, y2) i represents the position of the pixel point y i determined by the second matching group, q2 represents the probability of the position of the pixel point y ithe position of the pixel point y i represents the position of the pixel point y i determined by the third matching group, q3 represents the probability of the position of the pixel point y i determined by the third matching group; , , ;

[0132] T3: analyzing the position y i of each pixel point on the dynamic target area to generate three dynamic target areas of different ranges.

[0133] T3 includes the following steps:

[0134] T31: obtaining the discrete variable y i to obtain a range point set N of the dynamic target area;

[0135] T32: setting a minimum probability threshold in advance, deleting the pixel point y i whose position probability is less than the probability threshold in the range point set N to obtain a dynamic target area distribution scatter plot;

[0136] T33: taking the outermost boundary of the dynamic target area distribution scatter plot as a high-level target area boundary, taking the innermost boundary of the dynamic target area distribution scatter plot as a low-level target area boundary, and taking the path with equal distance from the high-level target area boundary and the low-level target area boundary as an intermediate-level target area boundary. The high-level target area boundary, the low-level target area boundary, and the intermediate-level target area boundary are three dynamic target areas of different ranges. The boundary ranges of the high-level target area boundary, the intermediate-level target area boundary, and the low-level target area boundary gradually decrease.

[0137] Further, the high-level target area boundary is determined as follows:

[0138] T331: setting a minimum selection number in advance;

[0139] T332: generating a selected contour that is proportionally enlarged from the tumor target area, and the selected contour wraps all points in the dynamic target area distribution scatter plot;

[0140] T333: gradually reducing the selected contour, and when the number of pixel points y i passed by the selected contour exceeds the minimum selection number, the selected contour is the high-level target area boundary.

[0141] The generation method of the low-level target area boundary is the same as that of the high-level target area boundary, except that the expansion direction of the selected target area is opposite.

[0142] The above merely provides preferred embodiments of the present application, and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the principles and technical scope of the present application shall fall into the scope of the present application.

Claims

1. A tumor target volume extraction system based on magnetic resonance guided radiotherapy, characterized by, The application relates to a dynamic information acquisition module, which acquires MRI images of a patient in a stable state before radiotherapy plan execution to form MRI four-dimensional dynamic information; a dynamic information positioning module, which outlines a tumor target area from the MRI four-dimensional dynamic information and renders four-dimensional target area dynamic information of the tumor target area in the stable state; an information positioning module, which extracts matching point information in different states from the four-dimensional target area dynamic information; and a target area tracking module, which extracts feature point information of a current state from MRI images during radiotherapy plan execution, matches the feature point information with the matching point information, generates real-time target area boundaries at the current time based on the position relationship between the matching point information and the target area boundaries. The matching points at least include skeleton points, organ points and blood vessel points for representing human body positions; and the state is the gas retention amount of the patient's lung. The MRI four-dimensional dynamic information P{x, y, z, v} is shown in the figure, wherein x, y and z represent the horizontal coordinate, vertical coordinate and height coordinate of a voxel in the MRI image, v represents the gas retention amount of the patient's lung, and P represents the voxel value. The four-dimensional target area dynamic information PE{x, y, z, v, b} is shown in the figure, wherein b represents whether it is a target area boundary, b=1 represents that the current voxel is a target area boundary, b=0 represents that the current voxel is not a target area boundary, and PE represents the voxel value. The target area boundary in the four-dimensional target area dynamic information is manually marked. The matching point information acquisition method comprises the following steps:

2. The MR-guided radiotherapy based tumor target volume extraction system of claim 1, wherein, Step 1: decomposing the four-dimensional target area dynamic information PE into MRI image information groups in different states; Step 2: selecting the target area boundary, part of the organ boundary, part of the skeleton boundary and part of the blood vessel boundary from each MRI image information group; 3. The MR-guided radiotherapy based tumor target volume extraction system of claim 1, wherein, Step 3: taking the voxels of the target area boundary as a target extraction set, extracting a plurality of extreme points from part of the organ boundary, part of the skeleton boundary and part of the blood vessel boundary to generate a first extreme point set, a second extreme point set and a third extreme point set; 4. The MR-guided radiotherapy based tumor target volume extraction system of claim 3, wherein, Step 4: taking the relative position information of each extreme point in the first extreme point set and the pixel points on the target area boundary as first relative information; taking the relative position information of each extreme point in the second extreme point set and the pixel points on the target area boundary as second relative information; taking the relative position information of each extreme point in the third extreme point set and the pixel points on the target area boundary as third relative information; Step 5: taking the first relative information, the second relative information and the third relative information as the matching point information. The target area tracking module comprises: a feature point information extraction unit for extracting feature point information in the MRI image during radiotherapy plan execution; a matching information unit for matching the feature point information with the matching point information to establish a mapping relationship between the feature points in the feature point information and the matching points; a target area tracking unit for outlining a dynamic target area in the MRI image during radiotherapy plan execution based on the corresponding relationship between the feature points and the matching points; 5. The MR-guided radiotherapy based tumor target volume extraction system of claim 4, wherein, wherein the mapping relationship between the feature points and the matching points comprises the probability of the mutual correspondence between the feature points and the matching points; and the target area tracking unit generates at least three target area boundaries of different levels based on the mapping relationship between the feature points and the matching points. ​ ​ ​ ​ 6. The MR-guided radiotherapy based tumor target volume extraction system of claim 5, wherein, The feature point information extraction method comprises the following steps: S1: extracting the MRI image information extraction group in different states according to the state decomposition of the MRI image in the radiotherapy plan execution; S2: corresponding the MRI image information extraction group in different states with the MRI image information group in different states; S3: extracting the extreme points near the boundaries of the organs, part of the bone boundaries and part of the blood vessel boundaries from the MRI image information extraction group; S4: taking the extreme points of the organ boundaries as the first extraction set, taking the extreme points of the bone boundaries as the second extraction set, and taking the extreme points of the blood vessel boundaries as the third extraction set; S5: taking the first extraction set, the second extraction set and the third extraction set as the feature point information.

7. The tumor target area extraction system based on magnetic resonance guided radiotherapy according to claim 6, wherein The matching information unit matches the feature point information and the matching point information based on the following scheme: Z1: combining the first extraction set with the first relative information, combining the second extraction set with the second relative information, and combining the third extraction set with the third relative information to form three matching groups; Z2: finding the highest similarity feature point for each matching point in the same matching group to generate a matching relationship, and taking the similarity of the feature point and the matching point as the first matching feature and the number of feature points around the feature point that are not matched with the matching point as the second matching feature in a matching relationship, wherein the feature point around the feature point is a preset 3 voxel grid window; Z3: establishing a probability matching model P based on the matching features of the three matching groups; P={p1, p2, …, p i …p z}, p i represents the position of the i-th pixel point on the dynamic target area; P i = p i (z 1,i , z 2,i , z 3,i , x 1,i , x 2,i , x 3,i ); z 1,i , z 2,i , z 3,i are first matching features of the three matching groups, respectively 1,i , x 2,i , x 3,i are second matching features of the three matching groups, respectively 8. The tumor target area extraction system based on magnetic resonance guided radiotherapy according to claim 5, wherein The extraction method of the feature points and the matching points is as follows: S01: extracting the extreme points from the MRI image based on the SIFT algorithm; S02: setting a standard window, traversing the nearby area of the reference object in the MRI image, and taking the region corresponding to the standard window as an extraction window when the number of extreme points and the extreme point density in the standard window exceed the preset threshold, and extracting all the extreme points in the extraction window as the feature points or the matching points.

9. The tumor target area extraction system based on magnetic resonance guided radiotherapy according to claim 7, wherein The target area tracking unit generates the dynamic target area based on the following steps: T1: Obtain a probability matching model P, P = {p1, p2, …, p i … z}, p i represents the position of the i-th pixel point on the dynamic target area; P i = p i (z 1,i , z 2,i , z 3,i , x 1,i , x 2,i , x 3,i ); T2: Based on the probability matching model P, the matching points corresponding to the feature points are analyzed, and according to the relative position information of the matching points and each pixel point on the target region boundary, the position y of each pixel point on the dynamic target region is obtained i discrete variables; wherein, wherein, (x1, y1) i represents the position of the pixel point y i determined by the first matching group, and q1 represents the probability of the position of the pixel point y i determined by the first matching group; (x2, y2) i represents the position of the pixel point y i determined by the second matching group, and q2 represents the probability of the position of the pixel point y i determined by the second matching group; (x3, y3) i represents the position of the pixel point y i determined by the third matching group, and q3 represents the probability of the position of the pixel point y i determined by the third matching group; , , ; T3: resolve the position y of each pixel point on the dynamic target area i The discrete variables of generate three different ranges of dynamic target areas.

10. The tumor target area extraction system based on magnetic resonance guided radiotherapy according to claim 9, wherein T3 comprises the following steps: T31 : obtaining y i a set of range points N of the dynamic target volume from the discrete variables T32: Pre-set a minimum probability threshold to delete pixels y within the range point set N. i The pixels whose positions have a probability less than a probability threshold are used to obtain a scatter plot of the dynamic target area distribution. T33: taking the outermost boundary of the dynamic target area distribution scatter plot as a high-level target area boundary, taking the innermost boundary of the dynamic target area distribution scatter plot as a low-level target area boundary, taking the path with equal distance from the high-level target area boundary and the low-level target area boundary as an intermediate-level target area boundary, and taking the high-level target area boundary, the low-level target area boundary and the intermediate-level target area boundary as three dynamic target areas of different ranges.

Images