Cancer dynamic image analysis method and system based on artificial intelligence

Abstract

The invention provides a cancer dynamic image analysis method and system based on artificial intelligence, which are applied to the technical field of medical image analysis, and are characterized in that a spatial transformation relationship between each image frame and a reference coordinate system is established, and superposition information is managed and converted under the reference coordinate system; and performing dynamic inverse transformation and superposition information according to a target image frame during display, thereby realizing dynamic and accurate spatial alignment presentation of a computer analysis result and the instantaneous position of the focus in each frame of the original image sequence, and realizing dynamic and accurate spatial alignment presentation of the computer analysis result and the instantaneous position of the focus in each frame of the original image sequence. And a doctor is supported to directly carry out accurate interactive operation and information marking on the dynamically aligned view, and meanwhile, the marking information can be kept in accurate space association with the instantaneous position of a focus in a frame when reappearing on any frame of the original sequence.

Description

Technical Field

[0001] The present application relates to the field of medical image analysis technology, and in particular to an artificial intelligence-based cancer dynamic image analysis method and system. Background Art

[0002] Dynamic contrast-enhanced imaging (such as DCE-CT and DCE-MRI) is a key tool in cancer diagnosis and treatment, providing information on lesion hemodynamics and contrast agent dynamics. However, traditional methods rely on the physician's subjective judgment, resulting in low efficiency and inconsistent results. While computer-assisted analysis systems can extract perfusion curves and quantify parameters, they have limitations: their analysis results are difficult to dynamically align with the instantaneous position of the lesion in the original image frame. This requires repeated comparison and adjustment when physicians perform marking and outlining operations on the original image, reducing efficiency and potentially introducing errors.

[0003] Furthermore, when performing interventional procedures or planning treatments, doctors need to refer to the status of lesions at multiple time points in dynamic image sequences. However, existing systems cannot dynamically and accurately align analysis results with the instantaneous position of the lesion in each frame of the original image sequence, and do not support direct interaction between doctors and the dynamic overlay view. This makes it difficult for doctors to accurately mark or plan lesions based on an accurate understanding of their characteristics.

[0004] After marking or planning, doctors need to save this information for subsequent review or reporting. However, if the marking information is only associated with a specific frame or internal coordinate system, body movement can cause displacement between the original image frames. When the marking is reproduced in other frames, it will deviate from the instantaneous position of the lesion, affecting the accuracy and safety of subsequent operations.

[0005] Therefore, in the dynamic enhanced imaging analysis and clinical application of cancer, there is an urgent need for a technical solution to overcome the influence of patient body motion, achieve dynamic and precise spatial alignment of computer analysis results with the instantaneous position of the lesion in each frame of the original image sequence, support doctors to directly interact and mark information on the dynamic alignment view, and ensure that the marked information maintains accurate spatial correlation with the instantaneous position of the lesion when reproduced on any frame of the original sequence.

[0006] In view of the above problems, the existing technology needs to be improved urgently. Summary of the Invention

[0007] In view of the above-mentioned shortcomings of the existing technology, the present application provides an artificial intelligence-based cancer dynamic image analysis method and system, which has the advantage of realizing dynamic and precise spatial alignment of computer analysis results with the instantaneous position of the lesion in each frame of the original image sequence, and supporting doctors to perform precise interactive operations and information marking directly on this dynamically aligned view, while ensuring that these marking information can maintain accurate spatial correlation with the instantaneous position of the lesion in any frame of the original sequence when reproduced on the frame.

[0008] In a first aspect, a method for analyzing cancer dynamic images based on artificial intelligence is provided, comprising the steps of: S1: Acquire a cancer dynamic enhancement image sequence; the cancer dynamic enhancement image sequence comprises a plurality of image frames arranged in chronological order; S2: establishing a spatial transformation relationship between each image frame in the cancer dynamic enhanced imaging sequence and a preset reference coordinate system; S3: Acquire superimposed information related to the lesion, and based on the spatial transformation relationship, convert and store the superimposed information in the reference coordinate system; S4: For any target image frame in the cancer dynamic enhanced image sequence, based on the spatial transformation relationship, inversely transform the superimposed information to the coordinate system of the target image frame, so as to be superimposed and displayed with the target image frame.

[0009] The present application proposes an artificial intelligence-based dynamic cancer image analysis method that can achieve dynamic and precise spatial alignment of computer analysis results with the instantaneous position of the lesion in each frame of the original image sequence, overcoming the influence of patient body movement.

[0010] Furthermore, step S2 includes: S21: acquiring, for at least one image frame in the cancer dynamic enhanced imaging sequence, a spatial distribution of pixel intensities within the image frame; S22: generating a structured data set corresponding to the image frame according to the spatial distribution of pixel intensities; S23: Calculating the spatial transformation relationship between the image frame and the reference coordinate system by spatially aligning the structured data set with a preset reference structured data set.

[0011] This application proposes an artificial intelligence-based cancer dynamic imaging analysis method, which provides a specific method for establishing spatial transformation relationships and improves the accuracy of registration.

[0012] Furthermore, step S22 includes: S221: Identifying a lesion area and a non-lesion area in the image frame; S222: generating a first structured subset and a second structured subset based on the spatial distribution of pixel intensities of the non-lesion area and the spatial distribution of pixel intensities of the lesion area, respectively; S223: Combine the first structured subset and the second structured subset to generate the structured data set, wherein, during the combination process, the contribution weight of the second structured subset to the generation of the structured data set is less than the contribution weight of the first structured subset to the generation of the structured data set.

[0013] This application proposes an artificial intelligence-based cancer dynamic image analysis method, which improves the robustness of structured data sets by weighted processing of lesion areas and non-lesion areas, and is beneficial for subsequent alignment.

[0014] Furthermore, step S23 includes: S231: Identifying a plurality of structural anchor points in the reference structured dataset; S232: Establishing structural constraint conditions based on the spatial arrangement of the plurality of structural anchor points; S234: When spatially aligning the structured data set corresponding to the image frame with the reference structured data set, performing calculations based on the structural constraint conditions to obtain the spatial transformation relationship between the image frame and the reference coordinate system.

[0015] This application proposes an artificial intelligence-based cancer dynamic imaging analysis method, which further improves the accuracy of spatial alignment by introducing structural constraints.

[0016] Furthermore, step S232 includes: S2321: For each of the plurality of image frames, calculating spatial relationship parameters between the plurality of structure anchor points to obtain a set of spatial relationship parameters corresponding to the plurality of image frames; S2322: Based on the spatial relationship parameter set, determine the allowable variation range of the spatial relationship parameters, and use the allowable variation range as the structural constraint condition.

[0017] Furthermore, step S234 includes: S2341: Determine a search range for a set of transformation parameters of the spatial transformation relationship based on the structural constraint condition; S2342: Perform optimization calculation within the determined search range to obtain the transformation parameters that optimally align the structured data set corresponding to the image frame with the reference structured data set, and determine the transformation parameters as the spatial transformation relationship.

[0018] Furthermore, step S3 includes: S31: Acquire a lesion analysis result related to the lesion as first superimposed information, and acquire user modification information for the first superimposed information; S32: Based on the user modification information, generate second overlay information representing the difference between the first overlay information and the result after the user modification; S33: Based on the spatial transformation relationship, transform the first overlay information and the second overlay information into the reference coordinate system to obtain first overlay transformation information and second overlay transformation information; S34: Store the first overlay transformation information and the second overlay transformation information in the reference coordinate system.

[0019] Furthermore, step S4 includes: S41: receiving a display mode instruction; S42: Processing the converted first overlay information and the converted second overlay information in the reference coordinate system according to the display mode instruction to generate target overlay information to be displayed; S43: Inversely transforming the target overlay information into the coordinate system of the target image frame, so as to overlay and display the target image frame.

[0020] Furthermore, step S42 includes: S421: If the display mode instruction is to display the original analysis result, select the converted first overlay information as the target overlay information; S422: If the display mode instruction is to display the final analysis result, combining the converted first overlay information with the converted second overlay information to generate the target overlay information; S423: If the display mode instruction is to display the modified difference, select the converted second overlay information as the target overlay information.

[0021] In a second aspect, an artificial intelligence-based cancer dynamic imaging analysis system is provided for implementing any of the above methods, the system comprising: Acquisition module: acquires a cancer dynamic enhanced image sequence; the cancer dynamic enhanced image sequence comprises a plurality of image frames arranged in chronological order; Establishing module: establishing a spatial transformation relationship between each image frame in the cancer dynamic enhanced imaging sequence and a preset reference coordinate system; A conversion module: acquiring superimposed information related to the lesion, and based on the spatial transformation relationship, converting and storing the superimposed information in the reference coordinate system; Display module: for any target image frame in the cancer dynamic enhanced image sequence, based on the spatial transformation relationship, inversely transforms the superimposed information into the coordinate system of the target image frame, so as to be superimposed and displayed with the target image frame.

[0022] Beneficial effects: The present application proposes an artificial intelligence-based cancer dynamic image analysis method and system, which establishes a spatial transformation relationship between each image frame and a reference coordinate system, manages and transforms the overlay information in the reference coordinate system, and then dynamically inverse-transforms the overlay information according to the target image frame during display, thereby achieving dynamic and precise spatial alignment of the computer analysis results with the instantaneous position of the lesion in each frame of the original image sequence. It has the advantage of achieving dynamic and precise spatial alignment of the computer analysis results with the instantaneous position of the lesion in each frame of the original image sequence, supporting doctors to directly perform precise interactive operations and information marking on this dynamically aligned view, while ensuring that these marking information maintains accurate spatial association with the instantaneous position of the lesion in any frame of the original sequence when reproduced on that frame. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flowchart of an artificial intelligence-based cancer dynamic imaging analysis method proposed in this application.

[0024] Figure 2 This is a structural diagram of an artificial intelligence-based cancer dynamic imaging analysis system proposed in this application.

[0025] Figure 3 This is an architectural diagram of an artificial intelligence-based cancer dynamic imaging analysis system proposed in this application.

[0026] Description of reference numerals: 201, acquisition module; 202, establishment module; 203, conversion module; 204, display module. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and marked in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work fall within the scope of protection of the present application.

[0028] It should be noted that similar reference numerals and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined or explained in subsequent figures. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and should not be understood as indicating or implying relative importance.

[0029] Please refer to Figure 1 , a cancer dynamic imaging analysis method based on artificial intelligence, the method comprising the steps of: S1: Acquire a cancer dynamic enhancement image sequence; the cancer dynamic enhancement image sequence includes multiple image frames arranged in time sequence; S2: Establishing the spatial transformation relationship between each image frame in the cancer dynamic enhancement imaging sequence and the preset reference coordinate system; S3: Obtaining superimposed information related to the lesion, and based on the spatial transformation relationship, converting and storing the superimposed information in a reference coordinate system; S4: For any target image frame in the cancer dynamic enhanced image sequence, based on the spatial transformation relationship, the superimposed information is inversely transformed to the coordinate system of the target image frame to be superimposed and displayed with the target image frame.

[0030] The acquisition of a dynamic enhanced cancer image sequence refers to a set of medical image frames containing contrast agent enhancement information acquired in chronological order, such as a dynamic enhanced CT or dynamic enhanced MRI sequence, which is used to provide image data of the lesion at different time points.

[0031] Establishing a spatial transformation relationship between each image frame and a preset reference coordinate system refers to calculating the geometric mapping relationship of each image frame in the sequence relative to a fixed spatial reference (reference coordinate system). This can be achieved through image registration technology, such as performing alignment calculations based on image content or specific structural features, which is used to quantify and record the spatial displacement and deformation of each image frame caused by body movement.

[0032] Among them, a frame in the cancer dynamic enhancement imaging sequence can be selected as a reference frame. By using the image space coordinate system of the selected reference frame as the preset reference coordinate system, when other image frames in the sequence are subsequently processed, their spatial transformation relationship relative to this reference frame can be calculated. In practical applications, in order to optimize the subsequent alignment effect, frames with specific characteristics in the sequence can be selected as reference frames, such as the image frame when the contrast agent reaches its peak, or the image frame where the lesion morphology is most clearly displayed.

[0033] Obtaining superimposed information related to lesions refers to spatial information related to lesion analysis results or user interaction operations. This can be a lesion feature map generated by a computer algorithm or a doctor's marking or outlining of areas on the image, which is used to provide additional information to assist clinical decision-making and operations.

[0034] Converting and storing the overlay information in a reference coordinate system means using the established spatial transformation relationship to map the overlay information from its original coordinate system to a preset reference coordinate system and save it in this reference coordinate system. It is used to create a stable spatial expression of the overlay information that is not affected by the movement of the original image frame.

[0035] Inversely transforming the overlay information to the coordinate system of the target image frame refers to mapping the overlay information stored in the reference coordinate system back to the instantaneous coordinate system of the target image frame, utilizing the inverse process of the spatial transformation relationship between the target image frame and the reference coordinate system, when any image frame in the sequence (the target image frame) needs to be displayed. This is used to spatially align the overlay information with the lesion location in the currently displayed image frame. Overlay display with the target image frame refers to fusing the inversely transformed overlay information with the original image of the target image frame for presentation, providing the user with a visual representation of the dynamic alignment of the overlay information with the original image.

[0036] Specifically, the method first acquires a dynamic contrast-enhanced cancer imaging sequence consisting of multiple image frames arranged in chronological order. Due to patient motion, these image frames experience spatial displacement. To overcome this, the method establishes a spatial transformation relationship between each image frame in the sequence and a pre-defined reference coordinate system. This relationship quantifies the geometric differences of each frame relative to the reference system. Subsequently, overlay information related to the lesion is obtained, such as computer analysis results or user markings.

[0037] Based on the established spatial transformation relationship, this overlay information is converted from its original coordinate system to a reference coordinate system for storage. In this way, all overlay information is unified into a stable spatial reference. When it is necessary to display any target image frame in the sequence, the method uses the inverse transformation of the spatial transformation relationship between the target image frame and the reference coordinate system to map the overlay information stored in the reference coordinate system back to the coordinate system of the target image frame.

[0038] Because the inverse transform takes into account the instantaneous position of the target image frame, the converted overlay information can be spatially aligned with the instantaneous position of the lesion in the target image frame. Finally, the inverse transformed overlay information is superimposed on the target image frame and displayed, so that when viewing a dynamic sequence, the overlay information can dynamically follow the movement of the lesion and maintain alignment.

[0039] Through the above solution, this application solves the problem of misalignment between the superimposed information and the instantaneous position of the lesion in the original image frame due to patient movement in a cancer dynamic enhanced image sequence. By establishing a unified reference coordinate system and performing dynamic spatial transformation, the superimposed information is spatially aligned on each frame of the original dynamic image sequence. This supports user interaction and information tagging on the dynamic image, and ensures that this information remains spatially associated with the lesion location when reproduced on any frame in the sequence, improving the accuracy of clinical analysis, planning, and operation.

[0040] Furthermore, step S2 includes: S21: acquiring, for at least one image frame in a cancer dynamic enhancement imaging sequence, a spatial distribution of pixel intensities within the image frame; S22: generating a structured data set corresponding to the image frame according to the spatial distribution of pixel intensities; S23: Calculate the spatial transformation relationship between the image frame and the reference coordinate system by spatially aligning the structured dataset with a preset reference structured dataset.

[0041] The pixel intensity spatial distribution refers to the set of grayscale values or color values of each pixel in the image frame and its position information in the image coordinate system, which can be obtained by directly reading the pixel data in the original image file.

[0042] A structured dataset refers to an abstract representation extracted or constructed from the spatial distribution of raw pixel intensities, which captures the key information or structural features in the image. For example, it can be based on image feature points (such as SIFT, SURF features), edge information, regional attributes, or a compact representation obtained through dimensionality reduction, feature encoding and other techniques.

[0043] Spatial alignment refers to the process of calculating the geometric transformation parameters required to transform the spatial position of one dataset to the spatial position of another dataset by comparing the similarity or correspondence between two datasets (such as a structured dataset and a reference structured dataset). It can be achieved by using a feature matching and iterative closest point (ICP) algorithm.

[0044] A preset reference structured dataset refers to a structured data representation associated with a preset reference coordinate system. It may be a structured dataset from a selected reference frame in a sequence, or a structured dataset of an independently constructed template representing the anatomical structure.

[0045] A spatial transformation relationship refers to a mathematical model that describes how a point in one coordinate system is mapped to a point in another coordinate system. For example, it can be a set of parameters such as translation, rotation, scaling, affine transformation, or non-rigid deformation.

[0046] In a specific embodiment, when establishing the spatial transformation relationship between an image frame and a reference coordinate system, pixel intensity data can be first read for each frame in the dynamically enhanced image sequence. A feature extraction algorithm, such as the SIFT or SURF algorithm, can then be used to detect and describe key points within the image frame from the spatial distribution of pixel intensities. The coordinates and descriptors of these key points are then used as a structured dataset corresponding to the image frame.

[0047] At the same time, the preset reference structured dataset can be obtained by performing the same feature extraction process on a reference frame in the sequence. Then, by matching the correspondence between the structured dataset (keypoint set) of the current image frame and the reference structured dataset (reference keypoint set), and using robust algorithms such as RANSAC to eliminate mismatched points, the affine transformation matrix required to transform the current image frame coordinate system to the reference coordinate system is calculated based on the successfully matched keypoint pairs. This matrix is the desired spatial transformation relationship.

[0048] Furthermore, step S22 includes: S221: Identifying a lesion area and a non-lesion area in the image frame; S222: generating a first structured subset and a second structured subset based on the spatial distribution of pixel intensities in the non-lesion area and the spatial distribution of pixel intensities in the lesion area, respectively; S223: Combining the first structured subset with the second structured subset to generate a structured data set, wherein, during the combination process, the contribution weight of the second structured subset to the generation of the structured data set is less than the contribution weight of the first structured subset to the generation of the structured data set.

[0049] Step S221 involves dividing the pixels in the image frame into portions belonging to the lesion and portions not belonging to the lesion through image processing or analysis techniques. This can be achieved using a variety of existing technologies, such as pixel intensity threshold segmentation, region growing-based segmentation, and machine learning model-based image segmentation (e.g., semantic segmentation using convolutional neural networks). A detailed description of image segmentation techniques is not provided here.

[0050] The non-lesion area corresponds to the portion of the image frame other than the lesion area, and usually includes surrounding normal tissues, bone structures or other relatively stable anatomical landmarks.

[0051] Step S222 refers to extracting and organizing structured information from these two different regions. The spatial distribution of pixel intensity includes the grayscale value or color value of the pixel and its position information in the image. Based on this information, structured subsets can be generated by extracting feature points (such as corner points, spots), edge features, texture features, or shape descriptors in each region, and organizing these extracted features and their spatial coordinates or mutual relationships into a data structure. The first structured subset represents the structural features of the non-lesion area, and the second structured subset represents the structural features of the lesion area.

[0052] In step S223, during the combination process, the contribution weight of the second structured subset to the generation of the structured data set is less than the contribution weight of the first structured subset to the generation of the structured data set, which means that the structured information extracted from the lesion area and the non-lesion area is integrated to form a data set for overall spatial alignment. The combination method can be to merge the data of the two subsets, such as merging two feature point sets or splicing two feature vectors. During the merging or integration process, by assigning a lower weight to the second structured subset (corresponding to the lesion area), its influence in the final structured data set can be reduced. The weight can be applied by scaling the data of the second structured subset before merging, or by assigning a lower weight to the feature matching or error term derived from the second structured subset when performing spatial alignment calculations based on the structured data set later.

[0053] As a preferred embodiment, when identifying lesion areas and non-lesion areas within an image frame, a deep learning-based segmentation model can be used. This model has been trained with a large amount of cancer imaging data and can automatically and accurately identify the pixel set in the lesion area. The non-lesion area is determined to be the remaining pixels in the image frame that do not belong to the lesion area.

[0054] When generating the first and second structured subsets, the Scale Invariant Feature Transform (SIFT) algorithm or the Speeded Up Robust Features (SURF) algorithm can be applied to the non-lesion regions to extract feature points and their descriptors, forming the first structured subset. For the lesion regions, their contour information, center point location, and texture features calculated based on the gray-level co-occurrence matrix can be extracted to form the second structured subset. When combining the first and second structured subsets, the feature descriptors in the first structured subset can be concatenated with the feature descriptors in the second structured subset to form a joint feature vector that represents the structured dataset.

[0055] In subsequent spatial alignment calculations, the feature matching error term from the second structured subset can be multiplied by a weight coefficient less than 1, such as 0.5, and the feature matching error term from the first structured subset can be multiplied by a weight coefficient of 1. The weighted error terms are then accumulated to calculate the overall alignment error and perform optimization.

[0056] Furthermore, step S23 includes: S231: identifying a plurality of structural anchors in a reference structured dataset; S232: Establishing structural constraints based on the spatial arrangement of multiple structural anchor points; S234: When spatially aligning the structured dataset corresponding to the image frame with the reference structured dataset, calculation is performed based on the structural constraint conditions to obtain a spatial transformation relationship between the image frame and the reference coordinate system.

[0057] The structural anchors refer to key points or regions with stable anatomical significance in the reference structured dataset, which can be implemented by using key points or regions identified based on anatomical knowledge or machine learning methods.

[0058] Structural constraints are restrictions established based on the spatial arrangement of multiple structural anchor points. These can be implemented, for example, by limiting the range of relative distances or angles between anchor points. Calculations based on structural constraints involve incorporating these constraints into optimization objectives or as a post-processing step during spatial alignment calculations. This can be achieved, for example, using constrained optimization-based alignment algorithms or post-processing filtering methods.

[0059] In one embodiment, the structural anchors can be pre-set as specific anatomical landmarks, such as identifying specific vascular branch points or organ boundary points in abdominal images for easy identification. Establishing structural constraints can be done by calculating spatial relationship parameters between these identified landmarks, such as Euclidean distance and relative angle, and statistically analyzing the distribution of these parameters based on a data set containing a large amount of image data to determine an allowable range of variation as the structural constraint. Calculation based on structural constraints can be performed in an image registration algorithm based on feature point matching, such as using an iterative optimization algorithm, adding a term to the optimization objective function. This term penalizes the calculated spatial transformation relationship that causes the spatial relationship parameters between the structural anchor points to exceed the preset allowable range of variation, thereby guiding the algorithm to find the optimal transformation that satisfies the structural constraints.

[0060] Furthermore, step S232 includes: S2321: For each of the multiple image frames, calculating spatial relationship parameters between the multiple structure anchor points to obtain a set of spatial relationship parameters corresponding to the multiple image frames; S2322: Based on the spatial relationship parameter set, determine the allowable variation range of the spatial relationship parameters, and use the allowable variation range as a structural constraint condition.

[0061] Among them, the spatial relationship parameters between multiple structural anchor points refer to the numerical values or vectors used to quantify the relative positional relationship between multiple structural anchor points. It can be achieved by calculating the Euclidean distance between any two structural anchor points, or the side length ratio, angle and other parameters of the geometric shape (such as triangle, quadrilateral) composed of multiple anchor points.

[0062] Among them, the spatial relationship parameter set refers to a set of spatial relationship parameters between multiple structural anchor points calculated for each frame in multiple image frames, which can be achieved by organizing and storing the spatial relationship parameters calculated for each frame in time sequence or frame index.

[0063] Among them, the allowable variation range refers to the numerical interval or threshold determined based on the spatial relationship parameter set, within which the spatial relationship parameters are allowed to fluctuate. It can be achieved by using the average value in the statistical spatial relationship parameter set and the standard deviation based on the average value plus or minus a set multiple.

[0064] Among them, structural constraints refer to the conditions used to limit or guide the search direction or results of transformation parameters during the spatial alignment calculation process so that they meet the relative position requirements of specific structures. They can be achieved by using the allowable variation range of the determined spatial relationship parameters as a restriction on the spatial relationship parameters of the structural anchor point after alignment.

[0065] In an exemplary embodiment, a method for establishing structural constraints can be implemented as follows: First, multiple structural anchor points are identified in a preset reference structured dataset. These anchor points can correspond to relatively stable anatomical structures in a dynamic contrast-enhanced cancer imaging sequence, such as specific vertebral edges, rib intersections, or major vascular branching points.

[0066] Then, for multiple image frames in a cancer dynamic enhancement imaging sequence, such as all frames in the sequence or a representative number of frames, the following operations are performed for each frame: The location corresponding to the reference structural anchor point is identified in the current image frame. Next, the Euclidean distance between any two structural anchor points is calculated. This process is repeated until the distances between all predetermined pairs of structural anchor points in the current frame have been calculated. The above calculations are performed for each frame in the sequence, thereby obtaining a set of spatial relationship parameters corresponding to the multiple image frames. For example, if the distances between N pairs of anchor points are calculated and M frames are analyzed, M sets of distance vectors will be obtained, constituting the set of spatial relationship parameters. Next, based on the obtained set of spatial relationship parameters, the allowable variation range of the spatial relationship parameters is determined. For example, for each pair of structural anchor points, the distance values across all analyzed frames are counted, and the mean and standard deviation of these distance values are calculated. The allowable variation range of the distance between the pair of anchor points can be set to the interval between the mean plus or minus two standard deviations. This process is repeated for all pairs of anchor points to obtain a set of allowable variation ranges for all spatial relationship parameters.

[0067] Finally, the allowable variation range of the determined spatial relationship parameters is used as a structural constraint. This allowable variation range can be incorporated as a constraint when subsequently performing spatial alignment calculations on the structured dataset corresponding to the image frame and a reference structured dataset, for example using an iterative optimization algorithm. For example, a term can be added to the optimization objective function to penalize situations where the spatial relationship parameters between the aligned structural anchor points exceed their allowable variation range. By minimizing the objective function including this constraint term, a spatial transformation relationship that satisfies the dynamic structural constraints can be calculated.

[0068] Furthermore, step S234 includes: S2341: Determine a search range of a set of transformation parameters of a spatial transformation relationship based on the structural constraint condition; S2342: Perform optimization calculation within the determined search range to obtain transformation parameters that optimize the alignment of the structured data set corresponding to the image frame with the reference structured data set, and determine the transformation parameters as a spatial transformation relationship.

[0069] Among them, a set of transformation parameters of the spatial transformation relationship refers to a set of mathematical parameters that define the spatial mapping relationship between the image frame coordinate system and the reference coordinate system, which may include translation parameters (displacement along the X, Y, and Z axes), rotation parameters (rotation angles around the X, Y, and Z axes), scaling parameters (scaling factors along each axis), and possible shearing parameters, etc.

[0070] The search range refers to the boundary or area of the parameter value space that is limited when optimizing spatial transformation parameters.

[0071] Optimization calculation refers to the process of searching for the optimal value of the objective function within a given parameter space through iteration, which can be implemented using the gradient descent method. Optimizing the alignment of the structured dataset corresponding to the image frame with the reference structured dataset means adjusting the spatial transformation parameters to achieve the best similarity or match between the transformed image frame structured dataset and the reference structured dataset. This can be achieved by calculating a similarity metric (such as mutual information, correlation coefficient, mean square error) or a difference metric (such as the average or maximum value of point-to-point distance or point-to-surface distance) between the two datasets and optimizing it.

[0072] In a specific embodiment, it is assumed that the structured data set should be , the reference structured dataset is , the spatial transformation relationship is composed of a set of transformation parameters Indicates that these parameters include translation parameter T, rotation parameter R, scaling parameter S, etc.

[0073] The purpose of optimization calculation is to find a set of transformation parameters , so that the transformed structured dataset With reference structured dataset The similarity or matching degree of the two is optimized. This can usually be defined using a similarity measure or a difference measure to define the objective function. : For example: mean square error function: , where N is the number of feature points in the structured dataset, is the i-th feature point in the image frame structured dataset; is the i-th feature point in the reference structured dataset; The transformation parameters Define the transformation function.

[0074] Or the mutual information function: ,in, express and The mutual information between them is greater, indicating and The higher the similarity.

[0075] Or the correlation coefficient function: ,in, express and The correlation coefficient between them is closer to 1. and The higher the similarity.

[0076] Since it is also necessary to ensure that the spatial relationship parameters between the transformed structural anchor points meet the allowable variation range. Assuming that the spatial relationship parameter between the structural anchor points is P, its allowable variation range is , then the constraints can be expressed as: Indicates the minimum allowable value of the spatial relationship parameter; Indicates the maximum allowable value of the spatial relationship parameter.

[0077] Under this constraint, an optimization calculation is performed, wherein the optimization calculation can adopt a gradient descent algorithm: , where α is the learning rate, is the objective function at the current parameter The gradient at .

[0078] Or use the penalty function method to deal with the constraints, and the objective function can be modified as follows: .

[0079] Here, λ is the penalty coefficient, which is used to control the strictness of the constraint. is the original objective function, is the optimized objective function. In the transformation parameters Next, the transformed structured dataset corresponding to the image frame The spatial relationship parameters between the structural anchor points.

[0080] and Respectively represent the minimum and maximum values of the allowable variation range of the spatial relationship parameters between structural anchor points. j is an index variable used to traverse the spatial relationship parameters between all structural anchor points that need to be considered. The max function is used to ensure that the penalty is only calculated when the constraint is violated (that is, when the spatial relationship parameter is outside its allowed range of variation). If the spatial relationship parameter is within its allowed range of variation, the corresponding penalty is 0.

[0081] Through the above optimization process, we can finally find a set of optimal transformation parameters , so that the structured dataset corresponding to the image frame is optimally aligned with the reference structured dataset and satisfies the structural constraints.

[0082] Furthermore, step S3 includes: S31: Acquire a lesion analysis result related to the lesion as first superimposed information, and acquire user modification information for the first superimposed information; S32: generating second overlay information representing the difference between the first overlay information and the result after the user modification based on the user modification information; S33: Based on the spatial transformation relationship, transform the first overlay information and the second overlay information into a reference coordinate system to obtain first overlay transformation information and second overlay transformation information; S34: Store the first overlay transformation information and the second overlay transformation information in a reference coordinate system.

[0083] Among them, the lesion analysis results refer to the structured or unstructured data related to the lesion automatically generated by the computer-aided analysis system after processing the cancer dynamic enhancement image sequence, which can be achieved by using a segmentation mask of the lesion area.

[0084] User modification information refers to the adjustment, supplementation or marking of the lesion analysis result by the user through interactive operations after reviewing the result, which can be achieved by editing the boundary data of the original segmentation mask.

[0085] The first overlay information refers to the directly acquired lesion analysis results. As the foundation of the overlay information, it can be represented using a data structure associated with the original image frame coordinate system. The second overlay information is generated by calculating the difference between the user-modified information and the first overlay information. It can be implemented using a mask representing the changed area.

[0086] A spatial transformation relationship is a mathematical model that maps points or regions in the original image frame coordinate system to a reference coordinate system. This can be represented by an affine transformation. The reference coordinate system is a stable spatial coordinate system that is independent of the instantaneous position of each frame in the original dynamic image sequence. This can be implemented using a preset fixed coordinate system or a stable average coordinate system calculated based on the image sequence.

[0087] The first overlay conversion information refers to data obtained by converting the first overlay information into a reference coordinate system based on a spatial transformation relationship, and can be represented by a data structure representing the original analysis results in the reference coordinate system. The second overlay conversion information refers to data obtained by converting the second overlay information into a reference coordinate system based on a spatial transformation relationship, and can be represented by a data structure representing user-modified differences in the reference coordinate system.

[0088] The solution of this application achieves effective differentiation and management of original analysis results and user-modified information by refining the overlay information related to lesions. For example, in a specific implementation, a computer-aided analysis system can first process a cancer dynamic enhancement image sequence, automatically identify and delineate the lesion area, and generate two-dimensional mask data of the lesion area as the first overlay information.

[0089] After reviewing the mask, the doctor finds that the boundary outlined by the system is not accurate enough, so he manually corrects the boundary and may mark a point of interest inside the lesion. These manually corrected boundary data and the coordinates of the marked points constitute the user modification information. Based on the original mask and the user-corrected mask, the system can calculate the difference between the two, for example, generating a difference mask representing the newly added and deleted areas, or recording vector data of the boundary changes and the coordinates of the newly added marked points, which constitute the second overlay information.

[0090] Subsequently, using a pre-calculated spatial transformation matrix that maps the original image frame coordinate system to the reference coordinate system, the original lesion mask data (first overlay information) and the difference mask / marker point data (second overlay information) are transformed, respectively, to obtain the first overlay transformed mask and second overlay transformed mask / marker point data in the reference coordinate system. Finally, these two sets of transformed data are stored as independent data layers in a database or file associated with the reference coordinate system.

[0091] Through the above scheme, the superimposed information related to the lesion is divided into the original analysis results and the user-modified information, and a second superimposed information representing the difference is generated respectively. Then, these two parts of information are independently converted and stored in a stable reference coordinate system. This allows the original analysis results and the user-modified parts to be independently identified, managed and accessed. Later, when displaying, you can flexibly choose to display only the original analysis results, only the user-modified parts, or combine the two to display the final results according to the user's needs. This flexibility improves the efficiency of doctors in reviewing and utilizing the analysis results, and supports doctors to perform precise interactive operations and information marking on the original images, thereby improving the practicality and accuracy of cancer dynamic image analysis in clinical applications.

[0092] Furthermore, step S4 includes: S41: receiving a display mode instruction; S42: Processing the converted first overlay information and the converted second overlay information in a reference coordinate system according to the display mode instruction to generate target overlay information to be displayed; S43: Inversely transforming the target overlay information into the coordinate system of the target image frame, so as to be overlaid with the target image frame for display.

[0093] Among them, display mode instructions refer to control signals or data used to instruct the system how to process and display different types of overlay information, which can be implemented by selection items in the user interface (such as drop-down menus, buttons or check boxes), system configuration parameters or commands received through external interfaces.

[0094] Processing refers to the operations performed on the first overlay information and the second overlay information stored in the reference coordinate system according to the received display mode instructions. These operations may include but are not limited to selection, combination, calculation, logical operation or filtering to generate the final overlay content that meets the user's display intention. It can be implemented using software algorithms, data processing modules or dedicated hardware logic circuits.

[0095] Target overlay information refers to data that is processed and ultimately used for overlay display with the target image frame. It represents the overlay content that the user expects to see in the current display mode, and can be implemented using image data, vector graphics data, text information, or a combination thereof.

[0096] In one exemplary embodiment, the system receives a display mode instruction selected by a user via a user interface. Assume the user selects "Display Final Results" mode. Based on this instruction, the system combines the stored first overlay information (e.g., automatically identified lesion outlines and perfusion maps) and the second overlay information (e.g., user-defined corrections to the outlines and markings of specific regions on the perfusion maps) in a reference coordinate system.

[0097] This combined processing can be to replace the original outline with the outline corrected by the user and superimpose the user mark on the original perfusion map, thereby generating a target overlay information representing the lesion information finally approved by the user. Subsequently, the system converts this target overlay information generated in the reference coordinate system to the coordinate system of the target image frame currently being viewed through an inverse transformation. For example, if the image currently being viewed is a specific time point in a dynamic sequence, the system will use the spatial transformation relationship between the image at that time point and the reference coordinate system to accurately convert the target overlay information to the pixel coordinate system of the image. Finally, the converted target overlay information (including the corrected outline and the perfusion map marked by the user) is superimposed and displayed with the target image frame, so that when the user views the image at that time point, he or she can see the analysis results finally confirmed by the user that are precisely aligned with the instantaneous position of the lesion in the frame.

[0098] Furthermore, step S42 includes: S421: If the display mode instruction is to display the original analysis result, select the converted first overlay information as the target overlay information; S422: If the display mode instruction is to display the final analysis result, combining the converted first overlay information and the converted second overlay information to generate target overlay information; S423: If the display mode instruction is to display the modified difference, select the converted second overlay information as the target overlay information.

[0099] This solution provides a flexible method for generating target overlay information to be displayed based on different display mode instructions. For example, in one specific embodiment, the display mode instructions can be triggered by the user clicking different buttons on the graphical user interface. When the user clicks the "Show Original Results" button, the system receives the corresponding display mode instruction and executes step S421 to directly designate the stored converted first overlay information (e.g., a two-dimensional or three-dimensional data structure representing the original lesion area and perfusion characteristics) as the target overlay information. When the user clicks the "Show Final Results" button, the system receives the corresponding display mode instruction and executes step S422 to combine the converted first overlay information with the converted second overlay information (e.g., a data structure representing areas added, deleted, or modified by the user), for example, through a logical OR operation or a rule-based merging algorithm, to generate target overlay information representing the final lesion analysis status. When the user clicks the "Show Modified Differences" button, the system receives the corresponding display mode instruction and executes step S423 to directly designate the stored converted second overlay information as the target overlay information. The generated target overlay information is then used for subsequent inverse transformation and overlay display.

[0100] Please refer to Figure 2 、 Figure 3 , an artificial intelligence-based cancer dynamic imaging analysis system, used to implement any of the above methods, the system comprising: Acquisition module 201: Acquisition of a cancer dynamic enhancement image sequence; the cancer dynamic enhancement image sequence comprises a plurality of image frames arranged in chronological order; Establishing module 202: establishing a spatial transformation relationship between each image frame in the cancer dynamic enhancement image sequence and a preset reference coordinate system; Conversion module 203: acquires superimposed information related to the lesion, and based on the spatial transformation relationship, converts and stores the superimposed information in a reference coordinate system; Display module 204: for any target image frame in the cancer dynamic enhanced image sequence, based on the spatial transformation relationship, inversely transforms the superimposed information to the coordinate system of the target image frame, so as to be superimposed with the target image frame for display.

[0101] Among them, each module in the system refers to a component or unit with a specific function that can work independently or in coordination with other parts, and can be implemented using software code, hardware circuit, firmware or a combination thereof.

[0102] The acquisition module 201 is a unit responsible for receiving or reading a cancer dynamic enhancement image sequence, and can be implemented using a file reading interface, a network communication interface, or a database access interface.

[0103] The establishment module 202 is a unit responsible for calculating the spatial transformation relationship between each image frame in the image sequence and the reference coordinate system, which can be implemented by an image registration algorithm processing unit or a machine learning model inference unit.

[0104] The conversion module 203 is a unit responsible for converting the overlay information between different coordinate systems, and can be implemented by a coordinate transformation calculation unit or a data mapping processing unit.

[0105] The display module 204 is a unit responsible for presenting the processed image and overlay information to the user, and can be implemented by a graphics rendering engine or a user interface generation unit.

[0106] The system decomposes the functions of the above method into an acquisition module 201 , a creation module 202 , a conversion module 203 and a display module 204 , so that the method can be effectively executed.

[0107] Specifically, the acquisition module 201 is responsible for receiving the original cancer dynamic enhancement image sequence, which is the basic data for all subsequent processing.

[0108] The establishment module 202 receives the image sequence provided by the acquisition module and calculates the spatial transformation relationship between each frame image in the sequence and a preset, stable reference coordinate system. This relationship quantifies the body movement between frames and provides a spatial reference for subsequent overlay information management.

[0109] The conversion module 203 receives overlay information related to the lesion (e.g., system analysis results or user markers) and, using the spatial transformation relationship calculated by the establishment module 202, converts this overlay information from its original coordinate system to a stable reference coordinate system for storage. Storing the overlay information uniformly in the reference coordinate system eliminates its position from being dependent on specific frames in the original image sequence that may have shifted, ensuring the spatial stability of the overlay information.

[0110] When the user selects any target image frame for viewing, the display module 204 uses the spatial transformation relationship between the target image frame and the reference coordinate system calculated by the establishment module 202 to inversely transform the overlay information stored in the reference coordinate system back to the coordinate system of the target image frame, and then overlays and displays it with the target image frame. This inverse transformation ensures that the overlay information can be accurately mapped to the currently displayed original image frame and maintains spatial alignment with the instantaneous location of the lesion in that frame.

[0111] Through this modular system architecture, the system effectively implements the aforementioned method steps, overcoming the overlay deviation caused by body motion. This allows the computer analysis results and user markings to be dynamically and accurately aligned with the instantaneous location of the lesion in each frame of the original image sequence. The combination of this system architecture and the aforementioned method steps enables the method to be applied stably and reliably in clinical practice, resolving the technical issue of misalignment between overlay information and original images in the presence of patient motion.

[0112] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0113] The foregoing is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. Persons skilled in the art will readily appreciate that the present application may be modified and altered in various ways. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A cancer dynamic image analysis method based on artificial intelligence, characterized in that: The method comprises the steps of: S1: Acquire a cancer dynamic enhancement image sequence; the cancer dynamic enhancement image sequence comprises a plurality of image frames arranged in chronological order; S2: establishing a spatial transformation relationship between each image frame in the cancer dynamic enhanced imaging sequence and a preset reference coordinate system; S3: Acquire superimposed information related to the lesion, and based on the spatial transformation relationship, convert and store the superimposed information in the reference coordinate system; S4: For any target image frame in the cancer dynamic enhanced image sequence, based on the spatial transformation relationship, inversely transform the superimposed information to the coordinate system of the target image frame, so as to be superimposed and displayed with the target image frame.

2. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 1, characterized in that: Step S2 includes: S21: acquiring, for at least one image frame in the cancer dynamic enhanced imaging sequence, a spatial distribution of pixel intensities within the image frame; S22: generating a structured data set corresponding to the image frame according to the spatial distribution of pixel intensities; S23: Calculating the spatial transformation relationship between the image frame and the reference coordinate system by spatially aligning the structured data set with a preset reference structured data set.

3. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 2, characterized in that: Step S22 includes: S221: Identifying a lesion area and a non-lesion area in the image frame; S222: generating a first structured subset and a second structured subset based on the spatial distribution of pixel intensities of the non-lesion area and the spatial distribution of pixel intensities of the lesion area, respectively; S223: Combine the first structured subset and the second structured subset to generate the structured data set, wherein, during the combination process, the contribution weight of the second structured subset to the generation of the structured data set is less than the contribution weight of the first structured subset to the generation of the structured data set.

4. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 2, wherein: Step S23 includes: S231: Identifying a plurality of structural anchor points in the reference structured dataset; S232: Establishing structural constraint conditions based on the spatial arrangement of the plurality of structural anchor points; S234: When spatially aligning the structured data set corresponding to the image frame with the reference structured data set, performing calculations based on the structural constraint conditions to obtain the spatial transformation relationship between the image frame and the reference coordinate system.

5. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 4, characterized in that: Step S232 includes: S2321: For each of the plurality of image frames, calculating spatial relationship parameters between the plurality of structure anchor points to obtain a set of spatial relationship parameters corresponding to the plurality of image frames; S2322: Based on the spatial relationship parameter set, determine the allowable variation range of the spatial relationship parameters, and use the allowable variation range as the structural constraint condition.

6. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 4, characterized in that: Step S234 includes: S2341: Determine a search range for a set of transformation parameters of the spatial transformation relationship based on the structural constraint condition; S2342: Perform optimization calculation within the determined search range to obtain the transformation parameters that optimally align the structured data set corresponding to the image frame with the reference structured data set, and determine the transformation parameters as the spatial transformation relationship.

7. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 1, characterized in that: Step S3 includes: S31: Acquire a lesion analysis result related to the lesion as first superimposed information, and acquire user modification information for the first superimposed information; S32: Based on the user modification information, generate second overlay information representing the difference between the first overlay information and the result after the user modification; S33: Based on the spatial transformation relationship, transform the first overlay information and the second overlay information into the reference coordinate system to obtain first overlay transformation information and second overlay transformation information; S34: Store the first overlay transformation information and the second overlay transformation information in the reference coordinate system.

8. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 7, characterized in that: Step S4 includes: S41: receiving a display mode instruction; S42: Processing the converted first overlay information and the converted second overlay information in the reference coordinate system according to the display mode instruction to generate target overlay information to be displayed; S43: Inversely transforming the target overlay information into the coordinate system of the target image frame, so as to overlay and display the target image frame.

9. The method for analyzing cancer dynamic images based on artificial intelligence according to claim 8, characterized in that: Step S42 includes: S421: If the display mode instruction is to display the original analysis result, select the converted first overlay information as the target overlay information; S422: If the display mode instruction is to display the final analysis result, combining the converted first overlay information with the converted second overlay information to generate the target overlay information; S423: If the display mode instruction is to display the modified difference, select the converted second overlay information as the target overlay information.

10. A cancer dynamic image analysis system based on artificial intelligence, characterized in that: For implementing the method according to any one of claims 1 to 9, the system comprises: Acquisition module: acquires a cancer dynamic enhanced image sequence; the cancer dynamic enhanced image sequence comprises a plurality of image frames arranged in chronological order; Establishing module: establishing a spatial transformation relationship between each image frame in the cancer dynamic enhanced imaging sequence and a preset reference coordinate system; A conversion module: acquiring superimposed information related to the lesion, and based on the spatial transformation relationship, converting and storing the superimposed information in the reference coordinate system; Display module: for any target image frame in the cancer dynamic enhanced image sequence, based on the spatial transformation relationship, inversely transforms the superimposed information into the coordinate system of the target image frame, so as to be superimposed and displayed with the target image frame.

Images