Medical target region generation method and device and electronic equipment

Abstract

The invention provides a medical target region generation method, and the method comprises the steps: obtaining target image data which at least represents a plurality of organ images of a target object; according to the target image data, a three-dimensional space distribution diagram is generated, and the three-dimensional space distribution diagram at least represents three-dimensional images of the organs; according to the three-dimensional space distribution diagram, space structure information is obtained, and the space structure information represents the space relation between the different organs; and generating a target region segmentation result according to the space structure information and the target image data.

Description

Technical Field

[0001] The present disclosure relates to the medical field, and more particularly, to a method, apparatus, and electronic device for generating a medical target area. Background Art

[0002] In the field of radiotherapy, the accurate delineation of tumor target areas is the key to the success of treatment planning, directly affecting the effective distribution of radiation dose and the degree of protection of healthy tissues. Currently, target area delineation mainly relies on the gray-scale information or multi-modal features of medical image data, and the tumor area is identified through automatic segmentation techniques. However, due to the complex morphology and blurred boundaries of tumors, and the often insufficient signal differences between tumors and surrounding normal tissues, the segmentation accuracy and consistency of traditional methods in complex anatomical structures are difficult to meet clinical requirements. In addition, existing technologies usually focus on feature extraction from the images themselves, ignoring the relative positions and spatial anatomical relationships between organs at risk and target areas, resulting in insufficient robustness of segmentation results in the face of individual differences or pathological changes. For example, traditional segmentation algorithms may not be able to effectively distinguish the boundaries between tumors and adjacent organs at risk, thereby affecting the optimization of treatment plans. To address the above problems, there is an urgent need for a new method for delineating target areas that can comprehensively utilize the spatial structure information of organs at risk to improve the accuracy and clinical applicability of segmentation. Summary of the Invention

[0003] In view of this, the present disclosure provides a method and an electronic device for generating a medical target area.

[0004] One aspect of the present disclosure provides a method for generating a medical target area, including: obtaining target image data, where the target image data at least represents the images of multiple organs of a target object; generating a three-dimensional spatial distribution map according to the target image data, where the three-dimensional spatial distribution map at least represents the three-dimensional images of multiple organs; obtaining spatial structure information according to the three-dimensional spatial distribution map, where the spatial structure information represents the spatial relationship between different organs; and generating a target area segmentation result according to the spatial structure information and the target image data.

[0005] According to an embodiment of the present disclosure, the organs include organs at risk. Obtaining spatial structure information according to the three-dimensional spatial distribution map includes: determining multiple organs at risk according to the three-dimensional spatial distribution map; and generating at least one piece of spatial structure information according to the spatial information of each organ at risk.

[0006] According to an embodiment of the present disclosure, the spatial structure information includes the spatial distance ratio between each organ at risk.

[0007] According to an embodiment of the present disclosure, generating a target area segmentation result includes: inputting the spatial structure information and the target image data into a target model to generate a target area segmentation result.

[0008] According to an embodiment of the present disclosure, the training process of the target model includes:

[0009] Obtain the sample space structure information, sample image data, and sample target area segmentation result. The sample space structure information represents the spatial relationship between at least some organs in the sample image data; input the sample space structure information and the sample image data into the target model to obtain the output information of the target model; calculate the target loss according to the output information and the sample target area segmentation result; adjust the parameters of the target model according to the target loss; repeat the above operations until the target model converges.

[0010] According to an embodiment of the present disclosure, calculating the target loss includes: generating a target loss function according to the sample space structure information, and the loss function is at least used to calculate the error between the spatial information between each organ in the output information of the target model and the sample space structure information; calculating the target loss according to the output information, the sample target area segmentation result, and the target loss function.

[0011] According to an embodiment of the present disclosure, generating a three-dimensional spatial distribution map according to the target image data includes: obtaining the spatial information of each organ according to the target image data, and the spatial information includes at least one of geometric shape characteristics, contour and boundary information, center point coordinates, and spatial occupancy area; generating a three-dimensional spatial distribution map according to the spatial information.

[0012] According to an embodiment of the present disclosure, the medical target area generation method further includes: obtaining a plurality of initial image data, and the data formats of the initial image data are different, and the initial image data is one of CT image data, magnetic resonance image data, or positron emission tomography data; performing a fusion registration operation on the plurality of initial image data to obtain the target image data.

[0013] Another aspect of the present disclosure provides a medical target area generation device, including: a first acquisition module for acquiring target image data, and the target image data at least represents the images of multiple organs of a target object; a first generation module for generating a three-dimensional spatial distribution map according to the target image data, and the three-dimensional spatial distribution map at least represents the three-dimensional images of multiple organs; a second acquisition module for obtaining spatial structure information according to the three-dimensional spatial distribution map, and the spatial structure information represents the spatial relationship between different organs; and a second generation module for generating a target area segmentation result according to the spatial structure information and the target image data.

[0014] Another aspect of the present disclosure provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the medical target area generation method of any one of the foregoing embodiments.

[0015] Another aspect of the present disclosure provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the medical target region generation method according to any one of the foregoing embodiments.

[0016] Another aspect of the present disclosure provides a computer program product, including a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, it implements the operations of the medical target region generation method according to any one of the foregoing embodiments.

[0017] According to the embodiments of the present disclosure, the medical target region generation method provided by the present disclosure has at least one of the following beneficial effects: By fusing the spatial structure information between organs and the target image data, the segmentation accuracy and anatomical consistency are significantly improved. First, by introducing the spatial distance ratio between critical organs as structural prior information, the model can reasonably infer the relative position of the target region, reducing the risk of overfitting and missegmentation, especially performing excellently in scenarios where organs are dense and tumor boundaries are blurred. Second, through the spatial structure modeling based on the three-dimensional spatial distribution map, the adaptability of the model to different individual anatomical differences is enhanced, improving the generalization ability of the model, and it can effectively support the automatic recognition of target regions under various body types, ages, or pathological conditions. In addition, the constructed target loss function simultaneously optimizes the image-level segmentation accuracy and the spatial-level structural consistency, strengthening the interpretability and clinical acceptability of the model, and solving the problem that traditional image segmentation methods only focus on pixel accuracy while ignoring the anatomical spatial relationship. By extracting spatial information from the fused and registered target image data, the spatial offset caused by the conduction of registration errors is avoided, further improving the stability of the spatial relationship modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Through the following description of the embodiments of the present disclosure with reference to the drawings, the above and other objects, features, and advantages of the present disclosure will become clearer. In the drawings:

[0019] Figure 1 Schematically shows a flowchart of the medical target region generation method according to an embodiment of the present disclosure;

[0020] Figure 2 Schematically shows a preprocessing flowchart of the medical target region generation method according to an embodiment of the present disclosure;

[0021] Figure 3 Schematically shows a flowchart of generating spatial structure information in the medical target region generation method according to an embodiment of the present disclosure;

[0022] Figure 4 Schematically shows another flowchart in the medical target region generation method according to an embodiment of the present disclosure;

[0023] Figure 5Schematically shows a flowchart of training a target model in a medical target area generation method according to an embodiment of the present disclosure;

[0024] Figure 6 Schematically shows a flowchart of calculating a target loss in the training process of a target model according to an embodiment of the present disclosure;

[0025] Figure 7 Schematically shows a flowchart of generating a three-dimensional spatial distribution map in a medical target area generation method according to an embodiment of the present disclosure;

[0026] Figure 8 Schematically shows a block diagram of a medical target area generation device according to an embodiment of the present disclosure; and

[0027] Figure 9 Schematically shows a block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present disclosure. Detailed implementation manners

[0028] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for the sake of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, obviously, one or more embodiments can also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present disclosure.

[0029] The terms used herein are merely for describing specific embodiments and are not intended to limit the present disclosure. The terms "including", "comprising", etc. used herein indicate the presence of the described features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0030] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0031] In cases where expressions similar to "at least one of A, B, and C, etc." are used, generally, it should be interpreted according to the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include, but not be limited to, a system having only A, only B, only C, having A and B, having A and C, having B and C, and / or having A, B, and C).

[0032] In the embodiments of the present disclosure, in aspects such as the collection, update, analysis, processing, use, transmission, provision, disclosure, storage, etc. of the involved data (e.g., including but not limited to user personal information), it complies with the provisions of relevant laws and regulations, is used for legal purposes, and does not violate public order and good customs. In particular, necessary measures are taken for user personal information to prevent illegal access to user personal information data and to safeguard user personal information security, network security, and national security.

[0033] Embodiments of the present disclosure provide a method for generating a medical target area, including obtaining target image data, where the target image data at least represents multiple organ images of a target object; generating a three-dimensional spatial distribution map based on the target image data, where the three-dimensional spatial distribution map at least represents the three-dimensional images of multiple organs; obtaining spatial structure information based on the three-dimensional spatial distribution map, where the spatial structure information represents the spatial relationship between different organs; and generating a target area segmentation result based on the spatial structure information and the target image data.

[0034] Figure 1 A flowchart of the method for generating a medical target area according to an embodiment of the present disclosure is schematically shown.

[0035] As Figure 1 shown, the method for generating a medical target area may at least include operations S110 to S140.

[0036] In operation S110, target image data is obtained, where the target image data at least represents multiple organ images of a target object.

[0037] Specifically, the target image data may be medical image data representing the three-dimensional structure and internal characteristics of multiple organs, with a unified spatial coordinate system, and can be used for collaborative analysis of multiple tissue organs. This data may include a unified representation after fusing multiple different modality images, which not only retains the clarity of the anatomical structure but also has a certain ability to identify tissues and functions. For example, the target image data can simultaneously show the metabolic activity of the tumor region, the spatial contact relationship with surrounding organs, and the soft tissue boundary, thus providing comprehensive support for subsequent three-dimensional structure modeling and target area identification.

[0038] In operation S120, a three-dimensional spatial distribution map is generated based on the target image data, where the three-dimensional spatial distribution map at least represents the three-dimensional images of multiple organs.

[0039] Specifically, through the organ recognition and spatial mapping method, the spatial morphological information of each organ can be extracted from the target image data and organized in a unified three-dimensional coordinate system, thereby constructing a complete three-dimensional spatial distribution map. This distribution map can accurately reflect the geometric structures, boundary contours of multiple organs, and their mutual positional relationships in space. For example, the organs in the image can be processed by layering and spatially reconstructed to form a spatial layout map including parameters such as the position, orientation, and size of the organs, so as to intuitively display the structural coupling state between the organs.

[0040] In operation S130, according to the three-dimensional spatial distribution map, spatial structure information is obtained, and the spatial structure information characterizes the spatial relationship between different organs.

[0041] Specifically, by analyzing information such as the relative positions, arrangement patterns, and boundary contacts of different organs in the three-dimensional spatial distribution map, quantitative data that can reflect the anatomical structural relationship between the organs is extracted. The spatial structure information may include parameters such as distance, azimuth angle, and contact area, which are used as important bases for subsequent target area positioning and discrimination. For example, if the target area is often located near a certain specific critical organ, the relative spatial distribution of this critical organ and other structures can be used to deduce the estimated position of the target area, realizing precise guidance based on the structural information.

[0042] In operation S140, according to the spatial structure information and the target image data, a target area segmentation result is generated.

[0043] Specifically, during the target area segmentation process, the spatial structure information is used as one of the input information, combined with the texture, density, or metabolic characteristics in the target image data, and input into the target area recognition model to assist the model in making a high-precision segmentation determination. This process improves the accuracy of target area contour recognition through the fusion of spatial position constraints and image semantic information, and is particularly suitable for complex regions with blurred boundaries or irregular shapes. For example, in the model inference stage, the target area is restricted within a position range that has a specific spatial relationship with the key organs, thus avoiding mis-segmentation to far-away regions or mis-identification of non-target tissues.

[0044] Figure 2 Schematically shows a preprocessing flowchart of a medical target area generation method according to an embodiment of the present disclosure.

[0045] As Figure 2 shown, on the basis of the foregoing embodiments, the medical target area generation method may include operations S210 to S220.

[0046] In operation S210, a plurality of initial image data are obtained, and the data formats between the respective initial image data are different. The initial image data is one of CT image data, magnetic resonance image data, or positron emission tomography data.

[0047] Specifically, the initial image data can be obtained through different imaging devices and techniques, and each imaging modality provides different types of information. For example, CT image data can provide detailed bone structure information, while MRI image data has strong soft tissue imaging capabilities, and PET image data can reflect cell metabolic activity. Since the formats, spatial resolutions, and data dimensions of these image data may vary, appropriate preprocessing and standardization are required during the fusion process. For example, CT image data is usually stored in the form of grayscale images, MRI data may contain multiple planes or different slice layers, and PET image data contains the distribution of radioactive tracers. These image data need to be uniformly processed in the same coordinate system.

[0048] In operation S220, for multiple initial image data, a fusion registration operation is performed to obtain target image data.

[0049] Specifically, in order to obtain the target image data, it is first necessary to perform registration processing on the image data from different sources to make them have a consistent spatial coordinate system. The registration operation can adopt rigid or non-rigid registration algorithms, and adjust the relative positions of different image data according to the features of the images (such as landmark points, boundaries, gray information, etc.) to ensure that they are aligned in the same coordinate system. Image scaling, rotation, and shear transformations may also be involved during the registration process to eliminate the differences caused by different shooting angles and devices. For example, using the feature point-based registration method, align the bone structure in the CT image with the soft tissue structure in the MRI image to ensure the accurate overlap of the two in the same space, and then combine it with the metabolic activity data in the PET image to generate comprehensive target image data. Through this process, the spatial information and functional information in different modalities can be unified, providing accurate data support for subsequent target volume delineation.

[0050] Figure 3 Schematically shows a flowchart of generating spatial structure information in the medical target volume generation method according to an embodiment of the present disclosure.

[0051] According to an embodiment of the present disclosure, the organ may include an organ at risk, such as Figure 3 As shown, based on the foregoing embodiment, S130 may include operations S310 to S320.

[0052] In operation S310, multiple organs at risk are determined according to the three-dimensional spatial distribution map.

[0053] Specifically, an organ at risk refers to a key normal tissue organ that needs to be particularly protected during radiotherapy to avoid high-dose radiation exposure, such as the brainstem, spinal cord, heart, lungs, optic nerve, kidneys, liver, bladder, etc. Its function has an important impact on the patient's physiological activities. Specifically, the system can match the spatial tags of each organ in the three-dimensional spatial distribution map with the medical knowledge base to identify the structures belonging to the category of organs at risk. During the identification process, information such as the morphological characteristics, anatomical location, and label encoding of each organ in the image can be combined to quickly screen out the relevant organs at risk that may affect the target area distribution. For example, in the radiotherapy scenario for chest tumors, organs such as the heart, lungs, and esophagus in the three-dimensional spatial distribution map can be identified as organs at risk; in the head and neck scenario, structures such as the optic nerve, brainstem, and cochlea may be included.

[0054] In operation S320, at least one spatial structure information is generated according to the spatial information of each organ at risk. Specifically, the spatial attribute information of each organ at risk in the three-dimensional spatial distribution map can be extracted, including but not limited to position coordinates, boundary point sets, voxel distributions, spatial occupancy regions, adjacency relationships, etc., and processed through structural modeling or spatial mapping methods to generate the spatial structure information for subsequent target segmentation. This spatial structure information can be used to express the spatial layout between each organ at risk and between it and the potential target area, providing a basis for constructing a spatial relationship model. For example, the relative position vector between each organ at risk and other organs can be calculated based on the center point position of each organ at risk, or a spatial topology map of the organs can be constructed to describe the spatial relationship of "organ A is located below organ B and the distance is less than d". These structural information can be used as an important basis for the subsequent model to judge the approximate position of the target area.

[0055] According to an embodiment of the present disclosure, the spatial structure information includes the spatial distance ratios between various organs at risk. The spatial distance ratios can be used to quantify the relative distance relationships between multiple organs at risk in three-dimensional space, thereby reflecting the proportional distribution characteristics of the organs in the individual anatomical structure. This ratio is usually generated by calculating the distance between two organs at risk and normalizing it with a selected reference distance to produce a proportional factor with relative scale significance. There are various options for calculating the spatial distance ratio, specifically including: Euclidean distance: The most common straight-line distance, applicable to most cases. Manhattan distance: Applicable to cases where the path along the coordinate axes needs to be calculated, such as in certain planes or anatomical regions. Chebyshev distance: Applicable to application scenarios where the maximum distance in any direction needs to be considered, especially in the independent directionality of anatomical structures. Mahalanobis distance: Applicable when considering the variance and covariance of organ spatial data, especially when dealing with anatomical regions of different scales and variances. Weighted distance: In specific cases, the relationships between certain organs may need to be weighted to highlight their importance or functional requirements. Angular distance: When the spatial relationship between organs is not only about distance, the distance can be calculated based on their relative angles.

[0056] For example, let the Euclidean distance between organs at risk A and B be D1, and the Euclidean distance between organs A and C be D2. Then, the ratio R = D1 / D2 can be constructed as the spatial structure information input reflecting the relative positional relationship between the organs. If in the scenario of head and neck tumors, the R value stably reflects the geometric relationship between the optic nerve and the brainstem for a long time, the model can use this ratio to assist in inferring the possible physiological positions of the target area, thereby improving the consistency and accuracy of target area delineation.

[0057] Figure 4 Another flowchart in the medical target area generation method according to an embodiment of the present disclosure is schematically shown.

[0058] As Figure 4 shown, based on the foregoing embodiment, S140 may include operation S410.

[0059] In operation S410, the spatial structure information and the target image data are input into the target model to generate a target area segmentation result.

[0060] Specifically, the target model can be a deep learning network, usually a convolutional neural network (CNN) or other models suitable for spatial information processing (such as a graph neural network or a Transformer model). This model receives two inputs: spatial structure information and target image data. The spatial structure information is quantitative data describing the spatial relationships between various organs, while the target image data provides the specific image features of the organs and tumors. By fusing these two types of information, the model can accurately identify and segment the target area in the image data and distinguish the boundaries from normal tissues and organs at risk.

[0061] For example, the input of the model can be the three-dimensional reconstructed image in the target image data and the planar images at various angles. Combining the relative positional relationship between organs (such as the spatial distance ratio), through adaptive learning, the network can identify the tumor region and avoid the regions in contact with the critical organs, thereby generating a high-quality target segmentation result.

[0062] Figure 5 Schematically shows the training flow chart of the target model in the medical target area generation method according to an embodiment of the present disclosure.

[0063] As Figure 5 shown, on the basis of the foregoing embodiment, the training process of the target model may include operations S510 to S540.

[0064] In operation S510, sample space structure information, sample image data, and sample target segmentation results are obtained. The sample space structure information characterizes the spatial relationship between at least some organs in the sample image data.

[0065] The sample image data may be labeled historical medical image data, having high image quality and anatomical structure clarity. The sources include imaging modalities such as CT, MRI, or PET, and have been preprocessed and fused through a standard processing flow.

[0066] The sample space structure information can be obtained by modeling the spatial relationship of the known organ regions in the sample image data, including structural parameters such as the spatial distance, relative positional relationship, angular relationship, or scale factor between organs, so as to reflect individual anatomical characteristics.

[0067] The sample target segmentation result is a manually delineated or automatically segmented result verified by an authority by a clinician, which can be used as a training supervision signal to compare the segmentation result output by the model and guide the learning direction.

[0068] Specifically, a batch of patient sample data sets that meet the requirements can be extracted from the database. Each sample simultaneously includes image data, organ structure spatial encoding, and standard target area labels, and is processed by normalization, interpolation, or spatial resampling as needed to unify the size and format of the input model. For example, a certain training sample includes a chest MRI image, the extracted spatial ratio relationship between the heart and the lungs, and the contour of the lung tumor delineated by a doctor, and these three will be used together as training data for model learning.

[0069] In operation S520, the sample space structure information and sample image data are input into the target model to obtain the output information of the target model. Specifically, the model can adopt a dual-input structure. The spatial structure information and the image data are respectively subjected to feature extraction through different encoding paths and then fused. After interaction in the feature fusion layer, an output tensor for target area prediction is generated. This output information is usually a segmentation probability map or mask, indicating the possibility that each pixel or voxel in the image belongs to the target area. For example, the spatial structure information can be input into a fully connected network for vector quantization encoding, and the image data can be input into a convolutional network to extract image features. The two are fused in the subsequent network, and finally a three-dimensional prediction result consistent with the size of the original image is generated.

[0070] In operation S530, according to the output information and the sample target area segmentation result, the target loss is calculated. The target loss characterizes the error between the output of the target model and the sample target area segmentation result. Specifically, by comparing the predicted segmentation output by the model with the true label of the sample (the sample target area segmentation result), the segmentation accuracy of the model on this sample is evaluated currently. The larger the loss value, the greater the prediction error of the model, and further adjustment is required. For example, the cross-entropy loss, Dice loss, or other common image segmentation error calculation methods can be used to perform pixel-by-pixel comparison between the predicted mask and the true mask to obtain the error index of the current training round.

[0071] In operation S540, according to the target loss, the parameters of the target model are adjusted. Specifically, the gradient of the loss function with respect to the model parameters is calculated through the backpropagation algorithm, and the model parameters are updated based on the gradient descent method (such as Adam, SGD, etc.) to make them closer to the true segmentation result. For example, if the edge of the target area predicted by the model is offset too much, the error calculated by the loss function will be fed back to the convolutional layer parameters of the model, making the next round of training pay more attention to the accuracy of the edge area.

[0072] Repeat operations S510 - S540 until the target model converges. The training process will continuously iterate the above steps until the segmentation performance of the target model on the validation set is stable and the change of the loss function tends to be gentle, or the preset number of training rounds, time, or performance threshold is reached. For example, it can be set that when the change of the Dice coefficient of the validation set is less than 0.01 in several consecutive rounds, it is considered that the model converges, the training is stopped, and the final model is saved for the inference stage.

[0073] Figure 6 Schematically shows a flowchart of calculating the target loss in the training process of the target model according to an embodiment of the present disclosure.

[0074] As Figure 6 shown, on the basis of the foregoing embodiment, S530 may include operations S610 - S620.

[0075] In operation S610, a target loss function is generated according to the sample space structure information. The loss function is at least used to calculate the error between the spatial information among various organs in the output information of the target model and the sample space structure information.

[0076] In operation S620, the target loss is calculated according to the output information, the sample target region segmentation result, and the target loss function.

[0077] Specifically, the target loss function can be designed as a polynomial structure, which includes both the traditional image segmentation loss part (such as DiceLoss or cross-entropy Loss) and the geometric loss term used to constrain the spatial structure consistency in the model output. Among them, the spatial structure error term is based on the spatial relationship between organs inferred from the model output result and the true relationship in the sample space structure information, and evaluates its deviation degree. This error can be constructed based on the distance between center points, the included angle of direction vectors, or the ratio difference.

[0078] For example, let the predicted Euclidean distance between organ A and organ B be , and the predicted Euclidean distance between organ A and organ C be , then the predicted spatial distance ratio is:

[0079]

[0080] And the true ratio provided in the sample space structure information is:

[0081]

[0082] Among them, is the true Euclidean distance between organ A and organ B, is the true Euclidean distance between organ A and organ C.

[0083] Then the spatial structure loss term can be defined as the difference between the two:

[0084]

[0085] Further generalized to the ratios between multiple pairs of organs, the loss term L structure can be defined as the average of all ratio differences:

[0086]

[0087] Among them, , are the predicted values of the distances between the two pairs of organs involved in the numerator and denominator in the i-th ratio respectively, , are the true values of the distances between the two pairs of organs involved in the numerator and denominator of the i-th comparison ratio, respectively, and N is the total number of samples.

[0088] Under this premise, the loss function can include two parts: one is the measurement of segmentation accuracy (such as Dice loss), and the other is the structural constraint loss based on the spatial distance ratio. The two together serve as the optimization objective in model training. Among them, the structural loss part is used to ensure that the spatial ratio between the target area output by the model and the key organs is maintained reasonably, simulating the thinking mode of clinicians referring to the anatomical structure ratio when delineating.

[0089] For example, the final target loss function can be expressed as:

[0090]

[0091] Among them, L segmentation represents the target area delineation accuracy loss (such as Dice loss), which is used to measure the similarity between the predicted mask and the true mask. α and β are the corresponding weights respectively, controlling the priority of different loss terms during training. α controls the weight of the segmentation loss term (such as Dice loss or cross-entropy loss) in the total loss. The segmentation loss term is usually used to evaluate the accuracy of the model's prediction at the pixel level, that is, the overlapping degree between the predicted target area (such as a tumor) and the true label. A larger α value means that the model will pay more attention to optimizing the segmentation accuracy. β controls the weight of the spatial structure loss term (such as the loss based on the spatial distance ratio) in the total loss. The spatial structure loss is used to evaluate the consistency between the spatial relationship between the organs output by the model and the true structure. The larger the β value, the more the model will attach importance to maintaining the consistency of the spatial relationship, especially in the structural features such as the relative position and distance between organs.

[0092] Among them, L segmentation can be the Dice loss:

[0093]

[0094] Among them, P is the mask predicted by the model, that is, the output generated by the model, representing the probability value or label (usually a binary representation of 0 or 1) of each pixel or voxel in the image belonging to the target area (such as a tumor, target area, etc.). G is the true label mask, usually obtained by manual annotation or other reliable methods for the "true" area annotation, representing the true situation of each pixel or voxel in the image belonging to the target area.

[0095] It should be noted that the above loss function is only an example. Those skilled in the art can design and select other loss functions that conform to this solution according to the solution provided by the present disclosure, which should be included in the scope protected by the present disclosure.

[0096] Figure 7A flowchart showing the generation of a three-dimensional spatial distribution map in the medical target area generation method according to an embodiment of the present disclosure is schematically shown.

[0097] As Figure 7 shown, based on the foregoing embodiment, S120 may include operations S710 to S720.

[0098] In operation S710, according to the target image data, the spatial information of each organ is obtained, and the spatial information includes at least one of geometric morphological features, contour and boundary information, center point coordinates, and spatial occupancy regions. Specifically, the spatial information is extracted from the target image data, and the target image data already has a unified spatial coordinate system, which can clearly represent the positions and relationships of multiple organs in three-dimensional space. The spatial information may include the center point coordinates, boundary contours, volume labels, structure masks, relative position vectors, distance ratios, etc. of each organ, and are all obtained based on the fused and registered images. After spatial normalization processing, these information can be directly used for three-dimensional modeling and spatial distribution construction. For example, the spatial information extracted from the target image data may include the segmentation results of multiple organs such as the liver, kidney, pancreas, etc. and their center points, boundary ranges, adjacency matrices, etc. in the unified voxel space, so as to reflect the tissue form of their anatomical structures in space.

[0099] In operation S720, a three-dimensional spatial distribution map is generated according to the spatial information. Specifically, by projecting or reconstructing the spatial information of each organ into a three-dimensional coordinate system, a complete spatial distribution map can be constructed. This distribution map can intuitively express the relative spatial relationship and topological structure between organs, facilitating subsequent target area recognition by the model under anatomical structure constraints. The generation of the three-dimensional spatial distribution map can be realized by methods such as voxel reconstruction, volume rendering, and boundary grid splicing, and structure codes can be added to each organ in the figure to achieve parallel visualization and structure modeling. For example, the system can generate a three-dimensional surface model of each organ using the Marching Cubes algorithm according to the extracted organ masks and center positions, and arrange and combine these models according to the unified coordinate axes to construct a three-dimensional spatial distribution map showing the geometric positions, proximity relationships, and spatial occupancy of each organ for use by downstream models.

[0100] According to the embodiments of the present disclosure, by fusing the spatial structure information between organs and the target image data, the segmentation accuracy and anatomical consistency are significantly improved. First, by introducing the spatial distance ratio between organs at risk as structural prior information, the model can reasonably infer the relative positions of the target areas, reducing the risks of overfitting and missegmentation, and performing excellently especially in scenarios where organs are dense and tumor boundaries are blurred. Second, through the spatial structure modeling based on the three-dimensional spatial distribution map, the adaptability of the model to different individual anatomical differences is enhanced, the generalization ability of the model is improved, and it can effectively support the automatic identification of target areas under various body types, ages, or pathological conditions. In addition, the constructed target loss function simultaneously optimizes the image-level segmentation accuracy and the spatial-level structural consistency, strengthening the interpretability and clinical acceptability of the model, and solving the problem that traditional image segmentation methods only focus on pixel accuracy while ignoring anatomical spatial relationships. By extracting spatial information from the fused and registered target image data, the spatial offset caused by the conduction of registration errors is avoided, further improving the stability of spatial relationship modeling.

[0101] Figure 8 A block diagram of a medical target area generation device according to an embodiment of the present disclosure is schematically shown.

[0102] As Figure 8 shown, the medical target area generation device 800 may include a first acquisition module 810, a first generation module 820, a second acquisition module 830, and a second generation module 840.

[0103] The first acquisition module 810 is configured to acquire target image data, and the target image data at least represents multiple organ images of a target object. In some embodiments, the first acquisition module 810 may be configured to perform the operation S110 in the above-mentioned medical target area generation method, which will not be elaborated here.

[0104] The first generation module 820 is configured to generate a three-dimensional spatial distribution map according to the target image data, and the three-dimensional spatial distribution map at least represents the three-dimensional images of multiple organs. In some embodiments, the first generation module 820 may be configured to perform the operation S120 in the above-mentioned medical target area generation method, which will not be elaborated here.

[0105] The second acquisition module 820 is configured to acquire spatial structure information according to the three-dimensional spatial distribution map, and the spatial structure information represents the spatial relationship between different organs. In some embodiments, the second acquisition module 820 may be configured to perform the operation S130 in the above-mentioned medical target area generation method, which will not be elaborated here.

[0106] The second generation module 820 is configured to generate a target area segmentation result according to the spatial structure information and the target image data. In some embodiments, the second generation module 820 may be configured to perform the operation S140 in the above-mentioned medical target area generation method, which will not be elaborated here.

[0107] Any of a plurality of modules, sub-modules, units, and sub-units according to embodiments of the present disclosure, or at least part of the functions of any of them may be implemented in one module. Any one or more of the modules, sub-modules, units, and sub-units according to embodiments of the present disclosure may be split into multiple modules for implementation. Any one or more of the modules, sub-modules, units, and sub-units according to embodiments of the present disclosure may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), programmable logic array (PLA), system on chip, system on substrate, system on package, application specific integrated circuit (ASIC), or may be implemented by any other reasonable way of integrating or packaging circuits, in hardware or firmware, or implemented in any one of the three implementation manners of software, hardware, and firmware, or in an appropriate combination of any several of them. Alternatively, one or more of the modules, sub-modules, units, and sub-units according to embodiments of the present disclosure may be at least partially implemented as a computer program module, which can execute corresponding functions when the computer program module is run.

[0108] For example, any of the first acquisition module 810, the first generation module 820, the second acquisition module 830, and the second generation module 840 may be combined and implemented in one module / unit / sub-unit, or any one of the module / unit / sub-unit may be split into multiple modules / units / sub-units. Alternatively, at least part of the functions of one or more of these modules / units / sub-units may be combined with at least part of the functions of other modules / units / sub-units and implemented in one module / unit / sub-unit. According to embodiments of the present disclosure, at least one of the first acquisition module 810, the first generation module 820, the second acquisition module 830, and the second generation module 840 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), programmable logic array (PLA), system on chip, system on substrate, system on package, application specific integrated circuit (ASIC), or may be implemented by any other reasonable way of integrating or packaging circuits, etc., in hardware or firmware, or implemented in any one of the three implementation manners of software, hardware, and firmware, or in an appropriate combination of any several of them. Alternatively, at least one of the first acquisition module 810, the first generation module 820, the second acquisition module 830, and the second generation module 840 may be at least partially implemented as a computer program module, which can execute corresponding functions when the computer program module is run.

[0109] It should be noted that the data processing system part in the embodiments of the present disclosure corresponds to the data processing method part in the embodiments of the present disclosure. For the description of the data processing system part, please refer to the data processing method part specifically, and details are not described herein again.

[0110] Figure 9 Schematically shows a block diagram of an electronic device suitable for implementing the method described above according to an embodiment of the present disclosure. Figure 9 The electronic device shown is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of the present disclosure.

[0111] As Figure 9 shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage section 908 into a random access memory (RAM) 903. The processor 901 can include, for example, a general microprocessor (such as a CPU), an instruction set processor, and / or a related chipset, and / or a dedicated microprocessor (such as an application specific integrated circuit (ASIC)), and so on. The processor 901 can also include on-board memory for caching purposes. The processor 901 can include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0112] In the RAM 903, various programs and data required for the operation of the electronic device 900 are stored. The processor 901, the ROM 902, and the RAM 903 are connected to each other through a bus 904. The processor 901 performs various operations of the method flow according to an embodiment of the present disclosure by executing the programs in the ROM 902 and / or the RAM 903. It should be noted that the program can also be stored in one or more memories other than the ROM 902 and the RAM 903. The processor 901 can also perform various operations of the method flow according to an embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0113] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, and the input / output (I / O) interface 905 is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the input / output (I / O) interface 905: an input portion 906 including a keyboard, a mouse, etc.; an output portion 907 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 908 including a hard disk, etc.; and a communication portion 909 including a network interface card such as a LAN card, a modem, etc. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read from it can be installed into the storage portion 908 as needed.

[0114] According to an embodiment of the present disclosure, the method flow according to the embodiment of the present disclosure may be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes program codes for executing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from the network through the communication portion 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the above-described systems, devices, apparatuses, modules, units, etc. may be implemented by computer program modules.

[0115] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiment; or may exist separately without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0116] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium. For example, it may include but is not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device.

[0117] For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the above-described ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903.

[0118] Embodiments of the present disclosure further include a computer program product, which includes a computer program. The computer program contains program code for executing the method provided by the embodiments of the present disclosure. When the computer program product runs on an electronic device, the program code is used to enable the electronic device to implement the control method provided by the embodiments of the present disclosure.

[0119] When the computer program is executed by the processor 901, the above functions defined in the system / apparatus of the embodiments of the present disclosure are executed. According to an embodiment of the present disclosure, the above-described systems, apparatuses, modules, units, etc. may be implemented by computer program modules.

[0120] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices and magnetic storage devices. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication section 909, and / or installed from the removable medium 911. The program code included in the computer program may be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above. According to the embodiments of the present disclosure, the program code for executing the computer program provided by the embodiments of the present disclosure may be written in any combination of one or more programming languages. Specifically, these computing programs may be implemented using high-level procedures and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include but are not limited to, such as Java, C++, python, the "C" language, or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., by connecting through the Internet using an Internet service provider).

[0121] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram or flowchart, and combinations of blocks in the block diagram or flowchart, may be implemented by a dedicated hardware-based system for performing the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions. Those skilled in the art can understand that the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways. All such combinations and / or combinations fall within the scope of the present disclosure.

[0122] The embodiments of the present disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the embodiments have been described separately above, this does not mean that the measures in each embodiment cannot be used advantageously in combination. Without departing from the scope of the present disclosure, those skilled in the art can make various substitutions and modifications, and all such substitutions and modifications should fall within the scope of the present disclosure.

Claims

1. A method for generating a medical target area, comprising: Obtaining target image data, where the target image data at least represents images of multiple organs of a target object; Generating a three-dimensional spatial distribution map according to the target image data, where the three-dimensional spatial distribution map at least represents three-dimensional images of multiple organs; Obtaining spatial structure information according to the three-dimensional spatial distribution map, where the spatial structure information represents the spatial relationship between different organs; Generating a target segmentation result according to the spatial structure information and the target image data.

2. The method according to claim 1, wherein, The organs include organs at risk. The obtaining of the spatial structure information according to the three-dimensional spatial distribution map includes: Determining multiple organs at risk according to the three-dimensional spatial distribution map; Generating at least one piece of the spatial structure information according to the spatial information of each organ at risk.

3. The method according to claim 2, wherein The spatial structure information includes the spatial distance ratio between each organ at risk.

4. The method according to claim 1, wherein The generating of the target segmentation result includes: Inputting the spatial structure information and the target image data into a target model to generate the target segmentation result.

5. The method according to claim 4, wherein, The training process of the target model includes: Obtaining sample spatial structure information, sample image data, and a sample target segmentation result, where the sample spatial structure information represents the spatial relationship between at least some organs in the sample image data; Inputting the sample spatial structure information and the sample image data into the target model to obtain the output information of the target model; Calculating a target loss according to the output information and the sample target segmentation result; Adjusting the parameters of the target model according to the target loss; Repeating the above operations until the target model converges.

6. The method according to claim 5, wherein The calculating of the target loss includes: Generating a target loss function according to the sample spatial structure information, where the loss function is at least used to calculate the error between the spatial information between each organ in the output information of the target model and the sample spatial structure information; Calculating the target loss according to the output information, the sample target segmentation result, and the target loss function.

7. The method according to claim 1, wherein Generating a three-dimensional spatial distribution map according to the target image data, including: Obtaining the spatial information of each organ according to the target image data, where the spatial information includes at least one of geometric morphological features, contour and boundary information, central point coordinates, and spatial occupancy area; Generating the three-dimensional spatial distribution map according to the spatial information.

8. The method according to claim 1, wherein, It further includes: Obtaining multiple initial image data, where the data formats of the initial image data are different, and the initial image data is one of CT image data, magnetic resonance image data, or positron emission tomography data; Performing a fusion registration operation on the multiple initial image data to obtain the target image data.

9. A medical target area generating device, comprising: A first obtaining module, configured to obtain target image data, where the target image data at least represents images of multiple organs of a target object; A first generating module, configured to generate a three-dimensional spatial distribution map according to the target image data, where the three-dimensional spatial distribution map at least represents three-dimensional images of multiple organs; A second acquisition module, configured to acquire spatial structure information according to the three-dimensional space distribution map, where the spatial structure information characterizes the spatial relationship between different organs; A second generation module, configured to generate a target region segmentation result according to the spatial structure information and the target image data.

10. An electronic device, comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 8.

Images