Target region positioning method, apparatus, device, and medium

Abstract

The application provides a target region positioning method, device, equipment and medium. The method comprises: obtaining an enhanced equivalent image pair for a target region in a radiotherapy process; the enhanced equivalent image pair is used to represent a virtual monochrome image pair in the case of injecting contrast agent into the target region; based on the enhanced equivalent image pair, a plurality of virtual monochrome images are generated according to a plurality of energy values; the plurality of virtual monochrome images correspond to a plurality of preset single energy values one by one; based on the energy value and the first CT value corresponding to each virtual monochrome image, a spectral curve is generated; a first energy point and a second energy point are determined from the spectral curve; based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point, a target region enhancement image is generated; the target region enhancement image is used to position the target region. In this way, the automation and objectivity of target region visualization energy selection can be realized, and the accuracy and efficiency of energy judgment in the target region positioning process can be improved.

Description

Technical Field

[0001] This application relates to the field of computer vision technology, and more specifically, to a target area localization method, apparatus, device, and medium. Background Technology

[0002] Radiation therapy is a treatment for tumors that uses high-dose ionizing radiation (commonly X-rays, gamma rays, or protons) to kill cancer cells and shrink tumors. Based on the delivery method, it is divided into external beam radiation therapy (external irradiation), brachytherapy (intracavitary / interstitial radiotherapy), and systemic radionuclide therapy. It is widely used in comprehensive tumor treatment. When performing radiation therapy on a tumor, it is necessary to determine the location of the target area. The target area is the collective term for the anatomical volume that needs to receive a sufficient dose to achieve the therapeutic goal.

[0003] During radiotherapy, virtual monochrome images at different energy levels display different effects on specific areas, and these images can assist in target localization. Currently, most virtual monochrome images used for target localization are selected manually, which results in low accuracy and efficiency in determining the optimal energy level for the target area, thus affecting the radiotherapy outcome. Summary of the Invention

[0004] In view of this, this application provides a target area localization method, apparatus, device and medium, which generates multiple virtual monochromatic images by enhancing equivalent image pairs, and then generates energy spectrum curves corresponding to the multiple virtual monochromatic images. Based on the energy spectrum curves, the energy points corresponding to the virtual monochromatic images used for target area localization are determined, which helps to improve the accuracy and efficiency of energy judgment for target area localization.

[0005] Specifically, this application is implemented through the following technical solution: According to a first aspect of this application, a target localization method is provided, the method comprising: During radiotherapy, enhanced equivalent image pairs are acquired for the target area; these enhanced equivalent image pairs are used to characterize virtual monochrome image pairs when contrast agent is injected into the target area. Based on the enhanced equivalent image pair, multiple virtual monochrome images are generated according to multiple preset single energy values; the multiple virtual monochrome images correspond one-to-one with the multiple preset single energy values. Based on the energy value and the first CT value corresponding to each of the virtual monochrome images, an energy spectrum curve is generated; the energy spectrum curve is used to represent the mapping relationship between the energy value and the first CT value; the first CT value corresponding to the virtual monochrome image is the CT value corresponding to the first region of interest located in the target area of ​​the virtual monochrome image; A first energy point and a second energy point are determined from the energy spectrum curve; the first energy point is used to represent the position in the target area where the contrast changes most drastically with energy, and the second energy point is used to represent the energy point where the CT value deviates the most from the first energy point. Based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point, a target enhancement image is generated; the target enhancement image is used to locate the target area.

[0006] In one alternative implementation, a first energy point is determined from the energy spectrum curve through the following steps: The energy spectrum curve is subjected to first-order difference processing to obtain the first-order difference values ​​corresponding to each energy point except the first energy point. The energy point with the largest absolute value of the corresponding first-order difference value is determined as the first energy point. Alternatively, the energy spectrum curve can be subjected to second-order difference processing to obtain the second-order difference values ​​corresponding to each energy point except the first energy point, and the energy point with the largest absolute value of the corresponding second-order difference value can be determined as the first energy point. Alternatively, peak detection can be performed on the energy spectrum curve, and the energy point with the largest corresponding first CT value can be determined as the first energy point.

[0007] In one optional implementation, generating a target enhancement image based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point includes: The virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point are subtracted to generate a target area difference image. Determine the contrast-to-noise ratio corresponding to each of the virtual monochrome images, and select the virtual monochrome image with the largest contrast-to-noise ratio from among the virtual monochrome images; The target region difference image and the virtual monochrome image with the largest contrast-to-noise ratio are added together to generate a target region enhancement image.

[0008] In one optional implementation, determining the contrast-to-noise ratio corresponding to each of the virtual monochrome images includes: For each of the virtual monochrome images, a first region of interest located within the target area and a second region of interest located within the normal tissue area are determined from the virtual monochrome image; Based on the first CT value corresponding to the first region of interest, the second CT value corresponding to the second region of interest, and the noise value corresponding to the second region of interest, the contrast-to-noise ratio of the virtual monochrome image is determined.

[0009] In one optional implementation, obtaining the enhanced equivalent image pair for the target region includes: Acquire a pair of non-enhanced virtual monochrome images of the target area during the treatment phase; the pair of non-enhanced virtual monochrome images is used to characterize a pair of virtual monochrome images of the target area without the injection of contrast agent. The unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

[0010] In one optional implementation, the method further includes: Based on the body part where the target area is located, select the substance decomposition method that matches the target area; The non-enhanced virtual monochrome image pair is subjected to material decomposition processing according to the described material decomposition method to obtain a bone component image; The bone component image is rigidly registered with the planned bone component image to obtain the adjustment parameters of the bone component image relative to the planned bone component image; the planned bone component image is obtained by performing material decomposition processing on the planned non-enhanced virtual monochrome image pair according to the material decomposition method; the planned non-enhanced virtual monochrome image pair is acquired for the target area during the simulation positioning stage; the simulation positioning stage is the stage of simulating target positioning of the human body before radiotherapy; The unenhanced virtual monochrome image pair is adjusted according to the adjustment parameters to obtain a rigidly registered unenhanced virtual monochrome image pair. The step of inputting the non-enhanced virtual monochrome image pair into the image prediction model to obtain the enhanced equivalent image pair corresponding to the non-enhanced virtual monochrome image pair output by the image prediction model includes: The rigidly registered unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

[0011] In one optional implementation, the image prediction model is trained through the following steps: Acquire pairs of unenhanced virtual monochrome images and pairs of enhanced virtual monochrome images of the sample target area during the simulation positioning phase; the unenhanced virtual monochrome image in the unenhanced virtual monochrome image pair is the virtual monochrome image before the injection of contrast agent into the sample target area, and the enhanced virtual monochrome image in the enhanced virtual monochrome image pair is the virtual monochrome image after the injection of contrast agent into the sample target area. Based on the sample non-enhanced virtual monochrome image pair, image alignment processing is performed on the sample enhanced virtual monochrome image pair to obtain the aligned sample enhanced virtual monochrome image pair; The sample unenhanced virtual monochrome image pairs are input into a neural network for image prediction to obtain the predicted enhanced virtual monochrome image pairs output by the neural network; the neural network includes a contrast agent distribution simulation module and an enhancement mode identification module; Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the target error loss is determined; Based on the target error loss, the parameters of the neural network are adjusted until the neural network meets the training cutoff condition. Based on the contrast agent distribution simulation module included in the neural network when the training cutoff condition is met, the image prediction model is obtained.

[0012] In one optional implementation, determining the target error loss based on the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair includes: Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the pixel regression loss, adversarial loss, and region of interest loss are determined. Subtraction processing is performed on the unenhanced virtual monochrome image pairs and the aligned enhanced virtual monochrome image pairs to obtain contrast-enhanced residual image pairs. Subtraction processing is performed on the sample unenhanced virtual monochrome image pair and the predicted enhanced virtual monochrome image pair to obtain the predicted contrast-enhanced residual image pair; Based on the sample contrast agent-enhanced residual image pair and the predicted contrast agent-enhanced residual image pair, the iodine distribution sensing loss is determined; The target error loss is determined based on the pixel regression loss, the adversarial loss, the region of interest loss, and the iodine distribution sensing loss.

[0013] In one optional implementation, the neural network is determined to meet the training cutoff condition through the following steps: If the CT value deviation between the first region of interest located within the target area in the predicted enhanced virtual monochrome image pair and the first region of interest located within the target area in the aligned sample enhanced virtual monochrome image pair meets a preset condition, the target error loss converges, and the peak signal-to-noise ratio of the predicted enhanced virtual monochrome image pair is stable, then the neural network is determined to meet the training cutoff condition.

[0014] In one optional implementation, the method further includes: The enhanced target region image is elastically registered, and the enhanced target region image after elastic registration is labeled to obtain the labeling result; the labeling result is used to indicate the location of the target region and the location of the organs at risk. Based on the annotation results, an adaptive radiotherapy treatment plan is generated.

[0015] According to a second aspect of this application, a target area localization device is provided, the device comprising: An image acquisition module is used to acquire enhanced equivalent image pairs of a target area during radiotherapy; the enhanced equivalent image pairs are used to characterize virtual monochrome image pairs when a contrast agent is injected into the target area; The image generation module is used to generate multiple virtual monochrome images based on the enhanced equivalent image pair and according to multiple preset single energy values; the multiple virtual monochrome images correspond one-to-one with the multiple preset single energy values; The curve generation module is used to generate an energy spectrum curve based on the energy value and the first CT value corresponding to each of the virtual monochrome images; the energy spectrum curve is used to represent the mapping relationship between the energy value and the first CT value; the first CT value corresponding to the virtual monochrome image is the CT value corresponding to the first region of interest located in the target area of ​​the virtual monochrome image; An energy selection module is used to determine a first energy point and a second energy point from the energy spectrum curve; the first energy point is used to represent the position in the target area where the contrast changes most drastically with energy, and the second energy point is used to represent the energy point where the CT value deviates the most from the first energy point. The image enhancement module is used to generate a target enhancement image based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point; the target enhancement image is used to locate the target area.

[0016] According to a third aspect of this application, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the target localization method described in the first aspect above.

[0017] According to a fourth aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the target localization method described in the first aspect above.

[0018] The target localization method, apparatus, device, and medium provided in this application acquire enhanced equivalent image pairs of virtual monochrome image pairs characterizing the state of contrast agent injection into the target area. Based on the enhanced equivalent image pairs, multiple virtual monochrome images corresponding to multiple preset single energy values ​​are generated. An energy spectrum curve representing the mapping relationship between the energy value and the first CT value is generated according to the first CT value. Based on the energy spectrum curve, a first energy point representing the most dramatic change in contrast with energy in the target area and a second energy point with the largest CT value deviation from the first energy point are determined and screened. The enhanced image of the target area is generated by combining the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point for target localization. In this way, the advantages of energy spectrum imaging and enhanced CT images can be fully utilized to highlight the boundary between the target area and the surrounding normal tissue, effectively reducing the problem of insufficient identification in conventional imaging localization. At the same time, the automation and objectification of target area visualization energy selection can be achieved, improving the accuracy and efficiency of energy judgment in the target localization process.

[0019] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of this disclosure.

[0020] To make the above-mentioned objects, features and advantages of this disclosure more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a target area localization method according to an exemplary embodiment of this application; Figure 2 This is a schematic diagram of a neural network structure shown in an exemplary embodiment of this application; Figure 3 This is a schematic diagram illustrating a target area localization process according to an exemplary embodiment of this application; Figure 4 This is a schematic diagram of a target area positioning device shown in an exemplary embodiment of this application; Figure 5 This is a schematic diagram of the structure of a computer device shown in an exemplary embodiment of this application.

[0022] In the diagram: 200 - Neural Network; 210 - Contrast Agent Distribution Simulation Module; 220 - Enhancement Mode Identification Module; 211 - Encoder; 212 - Bottleneck Layer; 213 - Decoder; 400 - Target Area Localization Device; 401 - Image Acquisition Module; 402 - Image Generation Module; 403 - Curve Generation Module; 404 - Energy Selection Module; 405 - Image Enhancement Module; 406 - Model Training Module; 407 - Plan Generation Module; 500 - Computer Equipment; 510 - Processor; 520 - Memory; 530 - Bus; 521 - Main Memory; 522 - External Memory. Detailed Implementation

[0023] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0024] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0025] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0026] To facilitate understanding of the technical solutions disclosed herein, the technical terms used in the embodiments of this disclosure will be explained first.

[0027] Medical linear accelerators (LINACs) are the most commonly used treatment machines in external beam radiotherapy. They accelerate electrons to high energies and then form a therapeutic beam in one of two ways: the electrons are used directly for treatment (electron beam, suitable for superficial lesions); or the electrons strike a high-Z metal target to generate megavolt-level X-rays for treating deep tumors. Modern LINACs can also precisely shape the beam and, in conjunction with image guidance, perform high-precision treatment.

[0028] LINAC's working principle and key components: Electron gun: Generates electrons. Microwave / RF source (magnetron or klystron) and accelerating structure / waveguide: Accelerates electrons to the required energy using a microwave electric field. Beam transmission and deflection (bending magnet, etc.): Guides electrons to the target position. Target / flattening filter or scattering foil: When irradiating X-rays, electrons bombard a metal target to produce braking radiation; traditionally, flattening filters (FF) are used to make the dose distribution more uniform at a certain depth; when irradiating an electron beam, scattering foil or magnetic scanning is used for spreading. Collimation and shaping: Primary / secondary collimators + multi-leaf collimators (MLC) dynamically open and close according to the planned shape, shaping and intensity modulating the beam. Ionization monitoring chamber: Monitors dose and beam stability online and interlocks. Gantt / Treatment Table / Imaging: The gantry rotates around the isocenter, and the treatment table is positioned with three or six degrees of freedom; equipped with kilovoltage (KV) or megavoltage (MV) imaging or cone-beam computed tomography (CBCT) for image-guided radiation therapy (IGRT).

[0029] LINAC typically provides multiple megavolt X-ray beams (approximately 4–25 MV) and electron beams (several MeV to over twenty MeV) for target areas at different depths. By removing the flattened filter, the central axis dose rate is significantly increased, while scattering and secondary head radiation are reduced. This is commonly used for efficient delivery in stereotactic radiosurgery (SRS) and stereotactic body radiation therapy (SBRT). LINAC can be used to perform three-dimensional conformal radiation therapy (3D-CRT), intensity-modulated radiation therapy (IMRT), volumetric modulated arc therapy (VMAT), spiral volumetric modulated arc therapy (SVMAT), and stereotactic radiotherapy (SRS / SBRT). IGRT can be used before / during treatment to correct the target area and organ position, improving accuracy and reducing the dose to normal tissues.

[0030] When performing radiotherapy on tumors, it is necessary to determine the location of the target volume. The target volume is the collective term for the anatomical volume that needs to receive a sufficient dose to achieve the therapeutic goal. It is not a single volume, but a set of stratified standard volumes (tumor volume / clinical target volume / internal target volume / planning target volume) used to specify prescriptions, plans, and assessments. The tumor volume (GTV) is the extent of the tumor and metastases that can be seen / confirmed by imaging, physical examination, or microscopy, and it forms the basis for all subsequent volumes. The clinical target volume (CTV) is the GTV plus any potential subclinical infiltrations (e.g., risk areas along anatomical structures or lymphatic drainage pathways); this is a clinical / anatomical concept representing the tumor-related tissue that "must be adequately irradiated." The internal target volume (ITV) is the volume that covers the internal movements of the tumor or organ within the body, such as during respiration, by adding an "internal margin (IM)" to the CTV; areas with minimal movement, such as the head and neck, may not require a separate ITV definition. The Planning Target Volume (PTV) is the external margin (Safety Margin, SM) added to the CTV (or ITV) to obtain the geometric volume used for planning optimization and dose assessment. The significance of PTV is to ensure that the CTV can still obtain a sufficient dose under various small errors.

[0031] Spectral imaging is a novel CT imaging technique based on dual-energy computed tomography (CT) or photon-counting CT. By acquiring data at different energy levels, it can generate various types of images, including material separation images, virtual monochrome images, iodine-based images, effective atomic number maps, and electron density maps. The main advantages of spectral CT include: material decomposition capability: it can decompose CT images into density images of different substances, such as bone components, soft tissue components, and water components; virtual monochrome images: it can generate images at any single energy level within the range of 40-200 keV, which helps optimize tissue contrast. Spectral imaging includes contrast-free imaging and contrast-enhanced imaging. Contrast-enhanced imaging utilizes the spectral characteristics of iodine contrast agents to more clearly display blood vessels and highly vascularized tumors.

[0032] Adaptive Radiation Therapy (ART) is a radiotherapy technique that dynamically adjusts the treatment plan based on changes in the patient's anatomy. During treatment, the location and morphology of the target area may change significantly due to tumor shrinkage, organ movement, weight changes, and other factors. Conventional radiotherapy involves pre-determining the treatment plan and executing it accordingly, making it impossible to monitor changes in the target area's morphology during treatment. Adaptive radiation therapy, by acquiring images during treatment, assessing changes, and redesigning the plan when necessary, can determine at each treatment session whether to use a pre-determined plan or a newly determined plan in real time. This allows for precise targeting of the target area while better protecting normal tissues.

[0033] Adaptive radiotherapy typically includes the following steps: image guidance, acquiring patient images before or during treatment, such as cone-beam CT, conventional CT, or MRI images; image registration, registering the images acquired during treatment with simulated positioning images to determine changes in target location; dose assessment, evaluating the impact of anatomical changes on the original planned dose distribution, where dose refers to the ionizing radiation dose absorbed by the body; plan adjustment, redesigning the plan as necessary, adjusting target delineation and the ionizing radiation dose distribution absorbed by the body; and quality verification, ensuring that the adjusted plan meets dosimetric requirements.

[0034] In radiotherapy, target localization is typically achieved using the following methods: Method 1, image-guided positioning and correction based on conventional CT images, is currently the mainstream clinical approach for IGRT. The main steps include: acquiring images of the patient's current treatment position using airborne CBCT or conventional CT before treatment; aligning the treatment image with the simulated positioning image (planning CT) using a rigid registration algorithm; calculating the target area's positioning offset in three-dimensional / six-dimensional space based on the registration results; and correcting the position by maneuvering the bed when the offset exceeds the clinically tolerable threshold to ensure precise alignment between the irradiation field and the target area. However, conventional CT images have limited soft tissue contrast, especially for tumors with similar density to surrounding tissues, making it difficult to clearly distinguish target boundaries; furthermore, they cannot obtain contrast enhancement information similar to enhanced CT, potentially affecting the assessment of some target areas or the boundaries of organs at risk.

[0035] Method two, marker-based tracking, involves implanting markers near the tumor and indirectly tracking the target area by tracing the markers' positions. The main steps include: implanting metallic markers or using anatomical landmarks near the tumor; obtaining marker positions before treatment using X-ray fluoroscopy or cone-beam CBCT; and calculating target area position changes based on marker positions. However, markers only reflect local location and cannot show changes in tumor morphology; implanting markers is invasive and can easily cause tissue damage; and certain anatomical locations are unsuitable for marker implantation.

[0036] Method three, based on respiratory gating, involves monitoring the patient's respiratory cycle and irradiating at specific respiratory phases. The main steps include: monitoring the patient's breathing using respiratory monitoring equipment (such as infrared trackers, strain sensors, etc.); triggering the radiation beam irradiation at a predetermined respiratory phase (usually end-expiration); and treating only when the target area is in the intended location. However, this method can only handle target area movement related to respiratory motion and cannot address anatomical changes caused by other factors, resulting in lower treatment efficiency.

[0037] Method four is MRI-guided radiotherapy. This method equips the treatment room with an MRI system to achieve real-time imaging of soft tissues. The main steps include: integrating the MRI system next to the treatment accelerator; acquiring real-time images of the patient's soft tissues; and delineating the target area online and adjusting the treatment plan. However, this method has extremely high equipment costs, and the design of MRI-compatible treatment equipment is complex.

[0038] In radiotherapy, spectral CT virtual monochrome images are often used. Virtual monochrome images at different energy levels display different effects on specific areas, and these images can assist in target localization. Currently, most virtual monochrome images for target localization are manually selected based on experience or subjective assessment. This experiential selection or subjective assessment has low accuracy in determining the optimal energy level for the target area, cannot achieve automated and objective energy selection, and is time-consuming and inefficient, affecting the consistency and repeatability of target localization results, thus impacting radiotherapy efficacy.

[0039] Based on the above research, this application provides a target area localization method that can fully highlight the boundary between the target area and the surrounding normal tissue by relying on the advantages of energy spectrum imaging and enhanced CT images, effectively reducing the problem of insufficient identification in conventional imaging localization, and can also realize the automation and objectification of target area visualization energy selection, thereby improving the accuracy and efficiency of energy judgment in the target area localization process.

[0040] To facilitate understanding of this embodiment, a target area localization method disclosed in this application will first be described in detail. The execution entity of the target area localization method provided in this application is generally a computer device with certain computing power. This computer device can be a server, which can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, big data, and artificial intelligence platforms. In some possible implementations, the computer device can also be a terminal device, which can be a mobile device, terminal, handheld device, computing device, vehicle-mounted device, etc. In other implementations, this target area localization method can be applied to an implementation environment composed of a terminal device and a server. Furthermore, this target area localization method can also be implemented by a processor calling computer-readable instructions stored in memory.

[0041] The following description, in conjunction with the accompanying drawings, illustrates a target localization method provided in an embodiment of this application.

[0042] See Figure 1 The diagram shown is a flowchart illustrating a target area localization method according to an exemplary embodiment of this application. Figure 1 As shown in the figure, the target area localization method provided in this embodiment includes steps S101 to S105, wherein: S101: During radiotherapy, an enhanced equivalent image pair is acquired for the target area; the enhanced equivalent image pair is used to characterize a virtual monochrome image pair when a contrast agent is injected into the target area.

[0043] In this step, during radiotherapy, a pair of enhanced equivalent virtual monochromatic images characterizing the injection of contrast agent into the target area can be acquired. Here, the enhanced equivalent images in the enhanced equivalent image pair correspond to different energy values. For example, the enhanced equivalent images in the enhanced equivalent image pair correspond to 40 keV and 140 keV, respectively.

[0044] In practical applications, using contrast agents during the simulation localization phase can enhance the contrast between the tumor and surrounding tissues, facilitating target delineation. However, during actual treatment, because the contrast agent has been metabolized in the patient's body, and repeated injections pose safety risks, it is impossible to obtain enhanced images of the same quality as in the simulation localization phase, resulting in unclear target areas during treatment. Therefore, how to overcome the limitations of contrast agent use during treatment and obtain enhanced images of the same quality as in the simulation localization phase has become an urgent problem to be solved.

[0045] In some possible implementations, acquiring the enhanced equivalent image for the target region includes: Acquire a pair of non-enhanced virtual monochrome images of the target area during the treatment phase; the pair of non-enhanced virtual monochrome images is used to characterize a pair of virtual monochrome images of the target area without the injection of contrast agent. The unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

[0046] In the above steps, non-enhanced virtual monochromatic image pairs of the patient targeting the treatment area can be acquired using spectral CT during the treatment phase (i.e., before or during radiotherapy). These non-enhanced virtual monochromatic image pairs can be acquired by emitting kilovolt rays, while megavolt rays can be emitted during actual treatment. Here, the energy values ​​corresponding to the non-enhanced virtual monochromatic images in the non-enhanced virtual monochromatic image pair are different. The energy values ​​corresponding to the non-enhanced virtual monochromatic images are determined according to the actual treatment needs and are not specifically limited here. For example, the non-enhanced virtual monochromatic images in the non-enhanced virtual monochromatic image pair correspond to 40keV and 140keV, respectively. The non-enhanced virtual monochromatic image pairs are input into a trained image prediction model. The image prediction model can predict enhanced virtual monochromatic image pairs based on the non-enhanced virtual monochromatic image pairs and output the enhanced equivalent image pairs corresponding to the non-enhanced virtual monochromatic image pairs.

[0047] In practical applications, although deep learning image synthesis technology is used in radiotherapy, it is mostly used to predict conventional CT images based on low-dose CT or MRI images, with a focus on improving image quality. It has not yet been applied to the prediction of target enhancement images.

[0048] In this way, by using deep learning networks to learn the mapping relationship from unenhanced images to enhanced images, and by using image prediction models to predict enhanced equivalent image pairs, the impact of contrast agent usage restrictions can be reduced during the treatment phase. At the same time, enhanced equivalent image pairs of the same quality as those in the simulation localization phase can be generated, providing the necessary image basis data for subsequent target localization.

[0049] In some possible implementations, the image prediction model is trained through the following steps: Acquire pairs of unenhanced virtual monochrome images and pairs of enhanced virtual monochrome images of the sample target area during the simulation positioning phase; the unenhanced virtual monochrome image in the unenhanced virtual monochrome image pair is the virtual monochrome image before the injection of contrast agent into the sample target area, and the enhanced virtual monochrome image in the enhanced virtual monochrome image pair is the virtual monochrome image after the injection of contrast agent into the sample target area. Based on the sample non-enhanced virtual monochrome image pair, image alignment processing is performed on the sample enhanced virtual monochrome image pair to obtain the aligned sample enhanced virtual monochrome image pair; The sample unenhanced virtual monochrome image pairs are input into a neural network for image prediction to obtain the predicted enhanced virtual monochrome image pairs output by the neural network; the neural network includes a contrast agent distribution simulation module and an enhancement mode identification module; Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the target error loss is determined; Based on the target error loss, the parameters of the neural network are adjusted until the neural network meets the training cutoff condition. Based on the contrast agent distribution simulation module included in the neural network when the training cutoff condition is met, the image prediction model is obtained.

[0050] In the above steps, non-enhanced virtual monochrome image pairs and enhanced virtual monochrome image pairs of the patient can be acquired during the simulation positioning phase targeting the target area. Here, the simulation positioning phase is the phase of simulating target area positioning of the human body before radiotherapy. The patient's position and scanning parameters are the same during the acquisition of the non-enhanced virtual monochrome image pairs and the enhanced virtual monochrome image pairs. Optionally, multiple sets of matched non-enhanced virtual monochrome image pairs and enhanced virtual monochrome image pairs can be acquired.

[0051] Before training the network using the unenhanced virtual monochrome image pairs and the enhanced virtual monochrome image pairs, preprocessing can be performed on them. Specifically, image alignment processing can be performed on the enhanced virtual monochrome image pairs based on the unenhanced virtual monochrome image pairs to obtain aligned enhanced virtual monochrome image pairs. In practice, the enhanced virtual monochrome image pairs can be translated, rotated, or otherwise aligned with the unenhanced virtual monochrome image pairs to obtain aligned enhanced virtual monochrome image pairs.

[0052] The sample unenhanced virtual monochrome image pairs are used as input X, and then input into a neural network for image prediction to obtain the predicted enhanced virtual monochrome image pairs output by the neural network. The aligned sample enhanced virtual monochrome image pairs are used as labels Y. Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, a target error loss is determined. Based on the target error loss, the parameters of the neural network are adjusted until the neural network meets the training cutoff condition. Based on the contrast agent distribution simulation module included in the neural network when the training cutoff condition is met, the image prediction model is obtained, such that F(X)≈Y.

[0053] In this way, by using paired images before and after enhancement in the simulated localization stage, and after image alignment processing, the neural network is trained, enabling the neural network to fully learn the mapping relationship from the non-enhanced image to the enhanced image. This effectively improves the accuracy and detail restoration of the model's prediction of the enhanced equivalent image, reduces problems such as texture distortion, grayscale deviation, and misalignment of the developing area in the predicted image, and ensures that the trained image prediction model has strong generalization ability and can stably output enhanced equivalent images that closely match the imaging effect of real injected contrast agents.

[0054] The neural network architecture in this embodiment can be a convolution-based encoder-decoder architecture (such as a U-shaped network, U-Net) or a deep learning architecture such as a generative adversarial network. It can also employ improved convolutional architectures such as residual networks and residual attention networks, as well as network architectures based on self-attention mechanisms such as Transformers. These networks exhibit superior performance in extracting complex image features and modeling long-range dependencies.

[0055] For example, taking a generative adversarial network architecture based on the decomposition of iodine-based substances using a neural network as an example, this network architecture can utilize the multi-energy characteristics of spectral CT to explicitly extract the features of iodine, thereby predicting the enhancement effect of contrast agents. For example, see here. Figure 2 This is a schematic diagram illustrating the structure of a neural network, as shown in an exemplary embodiment of this application. Figure 2As shown, the neural network 200 includes a contrast agent distribution simulation module 210 and an enhancement mode identification module 220. Here, the contrast agent distribution simulation module 210 acts as a generator, and the enhancement mode identification module 220 acts as a discriminator. The contrast agent distribution simulation module 210 and the enhancement mode identification module 220 form an adversarial training relationship. The contrast agent distribution simulation module 210 is used to convert sample non-enhanced virtual monochrome image pairs into predicted enhanced virtual monochrome image pairs that simulate the contrast agent enhancement effect. The enhancement mode identification module 220 is used to receive images (the predicted enhanced virtual monochrome image pairs or the predicted enhanced virtual monochrome image pairs) and determine their authenticity, thereby guiding the training of the contrast agent distribution simulation module 210.

[0056] The contrast agent distribution simulation module 210 includes an encoder 211, a bottleneck layer 212, and a decoder 213. The encoder 211, bottleneck layer 212, and decoder 213 are in series. Optionally, the contrast agent distribution simulation module 210 can be a variant of U-net. The encoder 211 is used to extract and compress features of the input image layer by layer. The bottleneck layer 212, located between the encoder 211 and the decoder 213, is the intermediate layer with the lowest feature dimension, used to integrate and preserve key information. The decoder 213 is used to restore spatial resolution layer by layer, outputting predicted enhanced virtual monochrome image pairs.

[0057] Here, the encoder 211 employs a multi-scale residual network (ResNet) and introduces an iodine feature-guided branch, specifically designed to capture the uptake patterns and metabolic characteristics of iodine contrast agents in different tissues (such as the liver, kidneys, and blood vessels). This avoids misidentifying bright areas in blood vessels as calcifications or ordinary noise, rather than simply extracting texture details from the image. Specifically, the input to the encoder 211 is the sample non-enhanced virtual monochrome image pair. The output of the encoder 211 is a multi-scale feature map (containing iodine uptake features extracted by the iodine feature-guided branch). The encoder 211 can extract and compress image features layer by layer, and specifically learn the iodine uptake patterns of different tissues through the iodine feature-guided branch.

[0058] The bottleneck layer 212 includes an anatomical structure preservation module that introduces structural information from the original plain CT scan via skip connections. This ensures that the patient's original anatomical morphology (such as organ size and bone position) is not altered during the generation of enhanced images; only the CT values ​​are changed. Specifically, the input to the bottleneck layer 212 is the deepest layer of compressed image features output by the encoder 211 and the original plain CT structural information introduced via skip connections. The output of the bottleneck layer 212 is a fused feature, which is a compressed feature representation fused with anatomical structure constraints. The bottleneck layer 212 can integrate encoded features and anatomical priors at the lowest feature dimension, ensuring that the patient's anatomical morphology is not altered during subsequent decoding.

[0059] The decoder 213 takes as input the fused features output from the bottleneck layer 212 and the skip connection features (multi-scale) of each layer output from the encoder 211. The output of the decoder 213 is the predicted enhanced virtual monochrome image pair. The decoder 213 can upsample the input features layer by layer to restore the spatial resolution of the original image, reconstructing the compressed features into a complete predicted enhanced virtual monochrome image pair. The predicted enhanced virtual monochrome image pair significantly improves the CT values ​​in areas such as blood vessels and hypervascular lesions, realistically simulating the image features after iodine contrast agent filling.

[0060] The enhancement mode identification module 220 employs a multi-task discriminator. Besides judging the overall realism of the image through the main branch, it also judges the rationality of the highlighted regions of interest (ROIs) through an auxiliary branch. This prevents the model from generating highlight artifacts in areas without blood vessels or from under-enhancing tumor areas that should be enhanced. Specifically, the input to the enhancement mode identification module 220 is either a real enhanced virtual monochrome image pair (i.e., an aligned sample enhanced virtual monochrome image pair) or a predicted enhanced virtual monochrome image pair output by the contrast agent distribution simulation module 210. The output of the main branch of the enhancement mode identification module 220 is a true / false probability value, specifically a value between 0 and 1; the output of the auxiliary branch of the enhancement mode identification module 220 is a rationality score for the highlighted ROIs. The enhancement mode identification module 220 can determine whether the input image is a real enhanced virtual monochrome image pair and evaluate whether the highlighted ROIs are anatomically reasonable.

[0061] Optionally, the sample non-enhanced virtual monochrome image pair is input to the encoder 211; the encoder 211 extracts features through multi-scale residual convolution, wherein the iodine feature-guided branch is specifically used to capture the uptake pattern of iodine contrast agent; the compressed image features encoded by the encoder 211 are input to the bottleneck layer 212; the bottleneck layer 212 introduces the structural information of the original plain CT scan through skip connections to ensure that the anatomical morphology remains unchanged; the decoder 213 receives the fusion features output by the bottleneck layer 212, upsamples layer by layer to restore spatial resolution, and outputs the predicted enhanced virtual monochrome image pair. The predicted enhanced virtual monochrome image pair or the aligned sample enhanced virtual monochrome image pair is input to the enhancement mode identification module 220. The main branch of the enhancement mode identification module 220 judges the overall authenticity of the image, and the auxiliary branch of the enhancement mode identification module 220 judges the rationality of the highlighted ROI region.

[0062] In some possible implementations, during neural network training, in step A, the parameters of the enhancement mode discrimination module 220 are fixed, and the sample non-enhanced virtual monochrome image pair is input into the contrast agent distribution simulation module 210 to obtain the predicted enhanced virtual monochrome image pair. The parameters of the contrast agent distribution simulation module 210 are updated according to the gradient of the target error loss. In step B, the parameters of the contrast agent distribution simulation module 210 are fixed, and the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair are respectively input into the enhancement mode discrimination module 220. The parameters of the enhancement mode discrimination module 220 are updated according to the gradient of the discrimination loss. Steps A and B are iterated alternately until the neural network meets the training cutoff condition, and training is considered complete. The contrast agent distribution simulation module 210 included in the trained neural network is used as the image prediction model. The image prediction model can generate enhanced equivalent image pairs that are as consistent as possible with the real enhanced virtual monochrome image pairs in terms of visual effect and CT value distribution.

[0063] In some possible implementations, determining the target error loss based on the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair includes: Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the pixel regression loss, adversarial loss, and region of interest loss are determined. Subtraction processing is performed on the unenhanced virtual monochrome image pairs and the aligned enhanced virtual monochrome image pairs to obtain contrast-enhanced residual image pairs. Subtraction processing is performed on the sample unenhanced virtual monochrome image pair and the predicted enhanced virtual monochrome image pair to obtain the predicted contrast-enhanced residual image pair; Based on the sample contrast agent-enhanced residual image pair and the predicted contrast agent-enhanced residual image pair, the iodine distribution sensing loss is determined; The target error loss is determined based on the pixel regression loss, the adversarial loss, the region of interest loss, and the iodine distribution sensing loss.

[0064] Here, by determining the target error loss, the target error loss quantifies the gap between the currently generated image (i.e., the predicted enhanced virtual monochrome image pair) and the desired image (i.e., the aligned sample enhanced virtual monochrome image pair). The gradient of the target error loss is determined by backpropagation, and the update direction and magnitude of the parameters of the contrast agent distribution simulation module 210 are determined according to the gradient of the target error loss, so that the network can simulate the real contrast agent enhancement effect. Here, the target error loss includes constraints on CT values ​​and iodine distribution. Specifically, the target error loss can be obtained by weighted summation of the pixel regression loss, the adversarial loss, the region of interest loss, and the iodine distribution perception loss. For example, the target error loss can be expressed by the following formula (1): (1) in, Indicates the target error loss; This represents pixel regression loss, which is used to ensure that the CT value of the overall background tissue does not drift. This represents the adversarial loss, which is used to ensure the overall visual naturalness of the generated image; This refers to the iodine distribution sensing loss, which is used to enable the network to learn the correct spatial distribution of the iodine contrast agent and to make the generated image clearer. The region of interest loss is used to guide the contrast agent distribution simulation module 210 to focus its learning on clinically relevant regions, ensuring that the enhancement effect in these regions meets clinical standards, so that the generated image can achieve the same level of enhancement in clinically relevant regions as the actual injected contrast agent. This represents the weight coefficients corresponding to the pixel regression loss. This represents the weighting coefficient corresponding to the adversarial loss. This represents the weighting coefficient corresponding to the iodine distribution perception loss. This represents the weighting coefficients corresponding to the region of interest loss. The weighting coefficients for pixel regression loss, adversarial loss, iodine distribution sensing loss, and region of interest loss can be determined based on actual treatment needs; no specific limitations are imposed here.

[0065] Here, the pixel regression loss can be obtained by calculating the difference between the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair pixel by pixel (e.g., using L1 or L2 norm). The pixel regression loss ensures that the generated predicted enhanced virtual monochrome image pair is closer to the aligned sample enhanced virtual monochrome image pair in terms of overall CT value, preventing meaningless drift of the CT value in the background tissue region, and providing pixel-level alignment constraints for network training.

[0066] Here, the adversarial loss is a feedback signal from the enhancement mode identification module 220. When the enhancement mode identification module 220 can distinguish between the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair, the loss value increases, prompting the contrast agent distribution simulation module 210 to generate a more realistic image; when the enhancement mode identification module 220 has difficulty distinguishing between the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair, the loss value decreases. This adversarial loss can improve the overall visual naturalness and texture realism of the synthesized image, avoiding the generation of overly smoothed or obviously artifact-laden images.

[0067] Here, subtraction processing is performed on the sample non-enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair. Specifically, the aligned sample enhanced virtual monochrome image pair is used to subtract the sample non-enhanced virtual monochrome image pair pixel-by-pixel to obtain the sample contrast agent enhanced residual image pair. Subtraction processing is then performed on the sample non-enhanced virtual monochrome image pair and the predicted enhanced virtual monochrome image pair. Specifically, the predicted enhanced virtual monochrome image pair is used to subtract the sample non-enhanced virtual monochrome image pair pixel-by-pixel to obtain the predicted contrast agent enhanced residual image pair. In this way, the contrast agent enhanced residual image pair is used as the key learning target for neural network training, allowing the neural network to focus on learning the region where the contrast agent enhancement effect occurs.

[0068] Optionally, a pre-trained energy dispersive spectroscopy (EDS) network can be used to extract a first virtual iodine map corresponding to the sample contrast agent enhancement residual image pair and a second virtual iodine map corresponding to the predicted contrast agent enhancement residual image pair. Based on the deviation between the first and second virtual iodine maps, the iodine distribution difference between the sample contrast agent enhancement residual image pair and the predicted contrast agent enhancement residual image pair is determined; based on the iodine distribution difference, the iodine distribution sensing loss is determined. The iodine distribution sensing loss can constrain the contrast agent distribution simulation module 210 at the material level (rather than just the pixel level), ensuring that the contrast agent distribution simulation module 210 learns the correct spatial distribution law of iodine contrast agent, such as the tissue requiring iodine uptake enhancement and the required enhancement amplitude.

[0069] Here, the region of interest loss can be obtained by determining the CT value deviation between the first region of interest located within the target area and the second region of interest located within the normal tissue area in the predicted enhanced virtual monochrome image pair.

[0070] During training, the four loss terms (pixel regression loss, adversarial loss, iodine distribution sensing loss, and region of interest loss) are weighted by coefficients ( to The target error loss can be obtained by weighted summation. .

[0071] Here, the discriminative loss is obtained through the following steps: The predicted enhanced virtual monochrome image pair (labeled "pseudo") and the aligned sample enhanced virtual monochrome image pair (labeled "true") are input into the enhancement pattern discrimination module 220. The main branch loss is calculated using the standard adversarial loss of a Generative Adversarial Network (GAN) (e.g., binary classification loss, commonly cross-entropy loss). The main branch loss represents the probability that the predicted enhanced virtual monochrome image pair output by the enhancement pattern discrimination module 220 is a true enhanced virtual monochrome image pair. The main branch loss measures the deviation between the label classification result of the enhancement pattern discrimination module 220 and the true label. The auxiliary branch loss is obtained based on the reasonableness score of the highlighted region ROI of the predicted enhanced virtual monochrome image pair. The discrimination loss is obtained based on the main branch loss and the auxiliary branch loss.

[0072] During training, an optimizer can be used to generate the gradient of the target error loss and the gradient of the discrimination loss, and update the parameters along the gradient descent direction. For example, an Adaptive Moment Estimation with Weight Decay Decoupling (AdamW) optimizer can be used, with an initial learning rate of 1e-4 (0.0001) and a weight decay coefficient of 1e-5 (0.00001).

[0073] Optionally, training can be conducted in two phases. In the first phase, reinforcement learning is performed on large blood vessels, prioritizing the training of the network to recognize and enhance large blood vessel structures with relatively high contrast agent concentrations and well-defined boundaries, such as the aorta and vena cava, allowing the network to learn basic enhancement patterns. In the second phase, reinforcement learning is performed on tissue structures with relatively low contrast agent concentrations and progressive changes, such as solid organs and small lesions. Building upon the accurate enhancement of large blood vessels, the network is further fine-tuned to learn the progressive enhancement characteristics of solid organs such as the liver and kidneys, as well as the blood supply enhancement characteristics of small lesions. This training method improves training stability and effectiveness.

[0074] In some possible implementations, the neural network is determined to meet the training cutoff condition through the following steps: If the CT value deviation between the first region of interest located within the target area in the predicted enhanced virtual monochrome image pair and the first region of interest located within the target area in the aligned sample enhanced virtual monochrome image pair meets a preset condition, the target error loss converges, and the peak signal-to-noise ratio (PSNR) of the predicted enhanced virtual monochrome image pair is stable, then the neural network is determined to meet the training cutoff condition.

[0075] The preset condition is that the CT value deviation between the first region of interest located within the target area in the predicted enhanced virtual monochrome image pair and the first region of interest located within the target area in the aligned sample enhanced virtual monochrome image pair is less than the clinically permissible error and no longer decreases after multiple consecutive iterations.

[0076] Optionally, if the target error loss tends to stabilize and no longer decreases significantly, it can be determined that the target error loss has converged. If the peak signal-to-noise ratio (PSNR) fluctuation of the predicted enhanced virtual monochrome image pair narrows and becomes basically stable, it can be determined that the PSNR of the predicted enhanced virtual monochrome image pair is stable.

[0077] In this way, the CT value deviation between the first region of interest (ROI) within the target area in the predicted enhanced virtual monochrome image pair and the first ROI within the target area in the aligned sample enhanced virtual monochrome image pair is used as the core judgment index. The core convergence criterion is that the CT value deviation between the first ROI within the target area in the predicted enhanced virtual monochrome image pair and the first ROI within the target area in the aligned sample enhanced virtual monochrome image pair is less than the clinically permissible error and does not decrease for multiple consecutive iterations. The target error loss and the peak signal-to-noise ratio (PSNR) of the predicted enhanced virtual monochrome image pair are used as auxiliary judgment indicators. The convergence of the target error loss and the stability of the PSNR of the predicted enhanced virtual monochrome image pair are used as auxiliary convergence criteria, which helps to improve the comprehensiveness and accuracy of network convergence judgment.

[0078] Here, in the two-stage training scenario, both training stages determine the training termination when the neural network meets the training termination conditions, but the training objectives and priorities differ. Specifically, in the first training stage, the focus is on monitoring the CT value deviations corresponding to large blood vessel structures such as the aorta and vena cava in the first region of interest, prioritizing the enhancement effect of these structures. When the CT value deviations in the regions corresponding to the large blood vessel structures meet preset conditions, the target error loss converges, and the peak signal-to-noise ratio of the predicted enhanced virtual monochrome image pair stabilizes, the first training stage is considered to have ended, indicating that the enhancement of the large blood vessel structures has met the target, and the second training stage begins. In the second training stage, based on the assumption that the enhancement of the large blood vessel structures has met the target, monitoring is added for solid organs, micro-lesions, and small blood vessels. When the CT value deviations in the entire first region of interest meet preset conditions, the target error loss converges, and the peak signal-to-noise ratio of the predicted enhanced virtual monochrome image pair stabilizes, the second training stage is considered to have ended, meaning the neural network training is complete.

[0079] In other possible implementations, implicit neural representation (INR) can be used to model the sample unenhanced virtual monochrome image pairs as continuous functions. By learning the nonlinear mapping field from flat scan to enhancement, the gradual filling process of contrast agent in microvessels can be expressed more delicately, avoiding the partial volumetric blurring caused by traditional convolutional networks.

[0080] In some possible implementations, the method further includes: Based on the body part where the target area is located, select the substance decomposition method that matches the target area; The non-enhanced virtual monochrome image pair is subjected to material decomposition processing according to the described material decomposition method to obtain a bone component image; The bone component image is rigidly registered with the planned bone component image to obtain the adjustment parameters of the bone component image relative to the planned bone component image; the planned bone component image is obtained by performing material decomposition processing on the planned non-enhanced virtual monochrome image pair according to the material decomposition method; the planned non-enhanced virtual monochrome image pair is acquired for the target area during the simulation positioning stage; the simulation positioning stage is the stage of simulating target positioning of the human body before radiotherapy; The unenhanced virtual monochrome image pair is adjusted according to the adjustment parameters to obtain a rigidly registered unenhanced virtual monochrome image pair. The step of inputting the non-enhanced virtual monochrome image pair into the image prediction model to obtain the enhanced equivalent image pair corresponding to the non-enhanced virtual monochrome image pair output by the image prediction model includes: The rigidly registered unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

[0081] Traditional rigid registration methods have limited accuracy when handling soft tissue deformation. When directly using conventional CT images for registration, skeletal information may dominate the registration process, affecting the registration accuracy of soft tissue regions. Traditional rigid registration methods require extracting skeletal structures from images, and their accuracy is greatly affected by boundary thresholds and soft tissue interference. When directly using conventional CT images for registration, the extraction accuracy of skeletal structures is often low, resulting in low registration accuracy. In this embodiment, the material decomposition capability of spectral CT is used to generate bone component images, and these bone component images are used for rigid registration, thereby eliminating soft tissue interference and improving the accuracy and stability of rigid registration.

[0082] Specifically, based on the body part where the target area is located, a material decomposition method matching the target area is selected. Optionally, for body parts with relatively few organs, such as the chest, lungs, and pelvis, a bone-water dual-material decomposition algorithm can be used with spectral CT to decompose the non-enhanced virtual monochrome image pairs into bone component images and water component images. For body parts with relatively many organs, such as the abdomen, a bone-blood dual-material decomposition algorithm can be used with spectral CT to decompose the non-enhanced virtual monochrome image pairs into bone component images and blood component images. In this embodiment, only the bone component image will be used subsequently, and the water and blood component images will not be used. The bone component image removes soft tissue interference, which can improve the stability of rigid registration.

[0083] The bone component image is rigidly registered with the planned bone component image to obtain adjustment parameters for the bone component image relative to the planned bone component image. Here, rigidity refers to unchanging parts within the human body, such as bones; therefore, rigid registration is bone-based registration. The adjustment parameters include displacement parameters, rotation parameters, etc. The planned bone component image is obtained by performing material decomposition processing on the planned non-enhanced virtual monochrome image pair according to the selected material decomposition method. The planned non-enhanced virtual monochrome image pair is acquired for the target area during the simulation positioning phase, which is the phase of target area simulation positioning of the human body before radiotherapy.

[0084] Given the adjustment parameters, the unenhanced virtual monochrome image pair can be adjusted according to the adjustment parameters to obtain a rigidly registered unenhanced virtual monochrome image pair. Then, the rigidly registered unenhanced virtual monochrome image pair is input into an image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

[0085] In this way, rigid registration guided by material decomposition utilizes the material decomposition capability of spectral CT to extract bone component images for rigid registration, eliminating soft tissue interference and improving registration accuracy. Based on the non-enhanced virtual monochromatic image pairs after rigid registration, the enhanced equivalent image pairs are predicted, effectively improving the prediction accuracy and imaging consistency of the enhanced equivalent image pairs.

[0086] S102: Based on the enhanced equivalent image pair, generate multiple virtual monochrome images according to multiple preset single energy values; the multiple virtual monochrome images correspond one-to-one with the multiple preset single energy values.

[0087] In this step, multiple preset single energy values ​​can be obtained, and virtual monochromatic images corresponding to each single energy value can be generated by energy spectrum CT based on the enhanced equivalent image pairs.

[0088] Here, the multiple preset single energy values ​​can be determined according to actual treatment needs, and no specific limitation is made here. For example, the multiple preset single energy values ​​can be 30keV, 31keV, 32keV, ..., 89keV, 90keV, a total of 61 single energy values, with an interval of 1keV, thereby generating 61 virtual monochrome images.

[0089] S103: Generate an energy spectrum curve based on the energy value and the first CT value corresponding to each of the virtual monochrome images; the energy spectrum curve is used to represent the mapping relationship between the energy value and the first CT value; the first CT value corresponding to the virtual monochrome image is the CT value corresponding to the first region of interest located in the target area of ​​the virtual monochrome image.

[0090] In this step, for each virtual monochrome image, a first region of interest (ROI) located within the target region is determined from the virtual monochrome image. Optionally, the first ROI located within the target region in the virtual monochrome image can be manually selected. Alternatively, an automatic segmentation algorithm (such as nnU-Net) can be used to automatically identify the first ROI located within the target region in the virtual monochrome image. Here, the first ROI located within the target region in the virtual monochrome image is generally an enhanced region within the target region, which is used to indicate the area to which the contrast agent is visualized, such as blood vessels.

[0091] Based on the average CT values ​​of all pixels in the first region of interest located within the target area in the virtual monochrome image, the CT value corresponding to the first region of interest located within the target area in the virtual monochrome image is determined, that is, the first CT value corresponding to the virtual monochrome image is determined.

[0092] In this way, energy points corresponding to each of the virtual monochrome images can be obtained, where the horizontal axis of the energy point represents the energy value and the vertical axis represents the first CT value. By sorting the energy points corresponding to each of the virtual monochrome images in ascending order of energy value, an energy spectrum curve can be obtained.

[0093] S104: Determine a first energy point and a second energy point from the energy spectrum curve; the first energy point is used to represent the position in the target area where the contrast changes most drastically with energy, and the second energy point is used to represent the energy point where the CT value deviates the most from the first energy point.

[0094] In this step, the first energy point in the target region with the most drastic contrast change with energy and the second energy point with the largest CT value deviation from the first energy point can be determined from the energy spectrum curve. Specifically, the largest CT value deviation from the first energy point refers to the largest absolute value of the difference in CT values ​​between the two energy points. Optionally, if multiple energy points have the largest CT value deviation from the first energy point, the energy point with the smallest corresponding energy value can be determined as the second energy point.

[0095] In some possible implementations, the first energy point is determined from the energy spectrum curve by the following steps: The energy spectrum curve is subjected to first-order difference processing to obtain the first-order difference values ​​corresponding to each energy point except the first energy point. The energy point with the largest absolute value of the corresponding first-order difference value is determined as the first energy point. Alternatively, the energy spectrum curve can be subjected to second-order difference processing to obtain the second-order difference values ​​corresponding to each energy point except the first energy point, and the energy point with the largest absolute value of the corresponding second-order difference value can be determined as the first energy point. Alternatively, peak detection can be performed on the energy spectrum curve, and the energy point with the largest corresponding first CT value can be determined as the first energy point.

[0096] In the above steps, the energy spectrum curve can be processed by first-order difference to obtain the first-order difference values ​​corresponding to each energy point except the first energy point. The first-order difference value represents the slope between the energy point and its preceding adjacent energy point. The energy point with the largest absolute value of the corresponding first-order difference value is determined as the first energy point. Optionally, if multiple energy points have the largest absolute value of their corresponding first-order difference values, the energy point with the smallest corresponding energy value can be determined as the first energy point.

[0097] The energy spectrum curve can also be processed using second-order difference to obtain the second-order difference values ​​corresponding to each energy point except the first energy point. The energy point with the largest absolute value of the corresponding second-order difference value is determined as the first energy point. Optionally, if multiple energy points have the largest absolute value of their corresponding second-order difference values, the energy point with the smallest corresponding energy value can be determined as the first energy point.

[0098] Alternatively, peak detection can be performed on the energy spectrum curve, and the energy point with the largest corresponding first CT value can be determined as the first energy point. Optionally, if multiple energy points have the largest corresponding first CT values, the energy point with the smallest corresponding energy value can be determined as the first energy point.

[0099] In this way, multiple methods such as first-order difference, second-order difference, or peak detection can be flexibly selected from the energy spectrum curve to screen the first energy point. First-order difference can quickly determine the energy point where the CT value of the target area changes with the fastest rate of change with energy. Second-order difference can quickly determine the energy point where the curvature of the energy spectrum curve changes most significantly. Peak detection can directly select the energy point with the highest CT value. Multiple methods for determining the first energy point are adapted to the attenuation characteristics of different tissue lesions. This can objectively and efficiently determine the position in the target area where the contrast changes most drastically with energy, reduce the limitations of manual subjective selection of energy points, effectively improve the accuracy of feature energy point selection, and ensure that when generating target area enhancement images based on the virtual monochrome images corresponding to the first and second energy points, the imaging differences between the target area and surrounding tissues can be maximized, effectively improving the accuracy of target area localization.

[0100] S105: Based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point, a target enhancement image is generated; the target enhancement image is used to locate the target area.

[0101] In this step, a target enhancement image is generated based on the difference in CT values ​​of corresponding pixels in the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point. This allows for precise localization and identification of the lesion target area based on the boundary between the target area and normal tissue in the target enhancement image.

[0102] In some possible implementations, generating a target enhancement image based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point includes: The virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point are subtracted to generate a target area difference image. Determine the contrast-to-noise ratio corresponding to each of the virtual monochrome images, and select the virtual monochrome image with the largest contrast-to-noise ratio from among the virtual monochrome images; The target region difference image and the virtual monochrome image with the largest contrast-to-noise ratio are added together to generate a target region enhancement image.

[0103] In the above steps, the CT values ​​of each pair of pixels corresponding to the same position in the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point can be subtracted to generate a target area difference image. It can be understood that the pixels in the target area difference image indicate the CT difference value. The target area difference image can highlight the difference between the target area and the surrounding tissue to the greatest extent.

[0104] Conventional CT images have limitations in terms of soft tissue contrast, especially for tumors with similar density to surrounding tissues (such as some abdominal tumors, pelvic tumors, etc.). Insufficient soft tissue contrast makes it difficult to accurately distinguish the target area boundary, affecting the accuracy of subsequent registration and boundary delineation.

[0105] In this embodiment, by determining the contrast-to-noise ratio of each of the virtual monochrome images, the virtual monochrome image with the highest contrast-to-noise ratio is selected. This virtual monochrome image with the highest contrast-to-noise ratio is then combined with the target region difference image to generate a target region enhancement image. Specifically, the CT values ​​of each pair of pixels corresponding to the same position in the target region difference image and the virtual monochrome image with the highest contrast-to-noise ratio can be added together. The target region enhancement image combines feature selection from the energy spectrum curve and image quality optimization based on the contrast-to-noise ratio.

[0106] In this way, the difference between the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point is first processed to obtain the target area difference image, which effectively amplifies the difference in energy spectral attenuation between the target area and the surrounding normal tissue. Then, by comparing the contrast-to-noise ratio of each virtual monochrome image, the virtual monochrome image with the largest contrast-to-noise ratio is selected, that is, the virtual monochrome image with the best imaging quality is selected. The target area difference image is added to the virtual monochrome image with the largest contrast-to-noise ratio. This process not only relies on the target area difference image to highlight the advantages of the target area and the surrounding tissue, but also relies on the high-quality imaging of the virtual monochrome image with the largest contrast-to-noise ratio to suppress image noise interference. The resulting target area enhancement image has clear layers and high boundary recognition, which can more intuitively and accurately present the target area position, effectively improving the accuracy and reliability of subsequent target area localization.

[0107] In some possible implementations, determining the contrast-to-noise ratio corresponding to each of the virtual monochrome images includes: For each of the virtual monochrome images, a first region of interest located within the target area and a second region of interest located within the normal tissue area are determined from the virtual monochrome image; Based on the first CT value corresponding to the first region of interest, the second CT value corresponding to the second region of interest, and the noise value corresponding to the second region of interest, the contrast-to-noise ratio of the virtual monochrome image is determined.

[0108] In the above steps, for each virtual monochrome image, a first region of interest (ROI) located within the target area and a second region of interest located within the normal tissue area are determined from the virtual monochrome image. Optionally, the first ROI located within the target area and the second ROI located within the normal tissue area in the virtual monochrome image can be manually selected. Alternatively, an automatic segmentation algorithm (such as nnU-Net) can be used to automatically identify the first ROI located within the target area and the second ROI located within the normal tissue area in the virtual monochrome image; here, the first ROI located within the target area in the virtual monochrome image is generally an enhanced region within the target area, which indicates the area to be contrasted by the contrast agent, such as blood vessels; the second ROI located within the normal tissue area in the virtual monochrome image is generally an organ region in the normal tissue area adjacent to the enhanced region.

[0109] Based on the average CT values ​​of all pixels in the first region of interest (ROI) located within the target area in the virtual monochrome image, a first CT value corresponding to the first ROI located within the target area in the virtual monochrome image is determined. Based on the average CT values ​​of all pixels in the second ROI located within the normal tissue area in the virtual monochrome image, a second CT value corresponding to the second ROI located within the normal tissue area in the virtual monochrome image is determined.

[0110] The absolute value of the difference between the first CT value corresponding to the first region of interest and the second CT value corresponding to the second region of interest, and the ratio of the noise value corresponding to the second region of interest, can be determined as the contrast-to-noise ratio of the virtual monochrome image. Specifically, the contrast-to-noise ratio can be determined by the following formula (2): (2) in, Indicates the contrast-to-noise ratio. This represents the first CT value corresponding to the first region of interest. This represents the second CT value corresponding to the second region of interest. This represents the noise value corresponding to the second region of interest.

[0111] In this way, by determining the first region of interest within the target area and the second region of interest within the normal tissue area in each virtual monochrome image, and determining the contrast-to-noise ratio based on the CT values ​​of the two areas and the noise value of the second region of interest, the imaging sharpness and target area resolution of each virtual monochrome image can be objectively quantified. This enables standardized quantitative judgment of image imaging quality, providing reliable data for selecting the virtual monochrome image with the best target area display effect (i.e., the virtual monochrome image with the highest contrast-to-noise ratio), and ensuring the imaging quality of the images used to generate subsequent target area enhancement images.

[0112] In some possible implementations, the method further includes: The enhanced target region image is elastically registered, and the enhanced target region image after elastic registration is labeled to obtain the labeling result; the labeling result is used to indicate the location of the target region and the location of the organs at risk. Based on the annotation results, an adaptive radiotherapy treatment plan is generated.

[0113] In the above steps, non-rigid transformations can be used to perform elastic registration on the target enhancement image, such as B-splines, free deformation, the Demons algorithm, optical flow, and deep learning registration networks. The elastically registered target enhancement image is then delineated and labeled to obtain the labeling results, which include the delineation of the target area and organs at risk. Based on the labeling results, an adaptive radiotherapy treatment plan is generated.

[0114] In this way, flexible registration of enhanced images of the target area can capture local deformation of soft tissue and effectively adapt to the dynamic changes in the morphology of human tissue during diagnosis and treatment. Based on this, the target area and organs at risk are labeled and an adaptive radiotherapy treatment plan is generated. This plan can be tailored to the patient's current actual condition in real time, effectively reducing the radiation deviation caused by tissue deformation. It ensures sufficient radiation dose to the target area while avoiding surrounding normal tissue, thus effectively improving the effect of adaptive radiotherapy.

[0115] For a clearer illustration of the target localization process, please refer to [link / reference]. Figure 3 This is a schematic diagram illustrating a target area localization process, as shown in an exemplary embodiment of this application. Figure 3 As shown, during the simulated positioning phase, sample non-enhanced virtual monochrome image pairs and sample enhanced virtual monochrome image pairs are acquired for model training to obtain an image prediction model. Non-enhanced virtual monochrome image pairs are acquired for the treatment phase. The non-enhanced virtual monochrome image pairs are processed according to the selected material decomposition method to obtain bone component images. The bone component images are rigidly registered with the planned bone component images to obtain adjustment parameters for the bone component images relative to the planned bone component images. The non-enhanced virtual monochrome image pairs are adjusted according to the adjustment parameters to obtain rigidly registered non-enhanced virtual monochrome image pairs. The image prediction model predicts enhanced equivalent image pairs based on the rigidly registered non-enhanced virtual monochrome image pairs. Energy spectrum curves corresponding to the enhanced equivalent image pairs are generated, and a first energy point and a second energy point are determined from the energy spectrum curves. Based on the virtual monochrome images corresponding to the first energy point and the second energy point, a target enhancement image is generated. The target enhancement image is elastically registered and the target area is delineated to generate an adaptive radiotherapy plan. Specific steps are described in the aforementioned embodiments and will not be repeated here.

[0116] The target localization method provided in this application acquires an enhanced equivalent image pair of virtual monochrome images representing the state of contrast agent injection into the target area. Based on the enhanced equivalent image pair, multiple virtual monochrome images corresponding to multiple preset single energy values ​​are generated. An energy spectrum curve representing the mapping relationship between the energy value and the first CT value is generated according to the first CT value. Based on the energy spectrum curve, a first energy point representing the most dramatic change in contrast with energy in the target area and a second energy point with the largest CT value deviation from the first energy point are determined and screened. The enhanced image of the target area is generated by combining the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point to perform target localization. In this way, the advantages of energy spectrum imaging and enhanced CT images can be fully utilized to highlight the boundary between the target area and the surrounding normal tissue, effectively reducing the problem of insufficient identification in conventional imaging localization. At the same time, it can realize the automation and objectification of target area visualization energy selection, and improve the accuracy and efficiency of energy judgment in the target localization process.

[0117] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0118] Corresponding to the aforementioned embodiments of the target area localization method, this application also provides embodiments of the target area localization device.

[0119] Please see Figure 4 This is a schematic diagram illustrating a target area positioning device according to an exemplary embodiment of this application. Figure 4 As shown in the figure, the target area localization device 400 provided in this application embodiment includes: The image acquisition module 401 is used to acquire enhanced equivalent image pairs of the target area during radiotherapy; the enhanced equivalent image pairs are used to characterize virtual monochrome image pairs when contrast agent is injected into the target area; Image generation module 402 is used to generate multiple virtual monochrome images based on the enhanced equivalent image pair and according to multiple preset single energy values; the multiple virtual monochrome images correspond one-to-one with the multiple preset single energy values; The curve generation module 403 is used to generate an energy spectrum curve based on the energy value and the first CT value corresponding to each of the virtual monochrome images; the energy spectrum curve is used to represent the mapping relationship between the energy value and the first CT value; the first CT value corresponding to the virtual monochrome image is the CT value corresponding to the first region of interest located in the target area of ​​the virtual monochrome image. The energy selection module 404 is used to determine a first energy point and a second energy point from the energy spectrum curve; the first energy point is used to represent the position in the target area where the contrast changes most drastically with energy, and the second energy point is used to represent the energy point with the largest CT value deviation from the first energy point. The image enhancement module 405 is used to generate a target enhancement image based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point; the target enhancement image is used to locate the target area.

[0120] In one optional implementation, the curve generation module 403 is used to determine a first energy point from the energy spectrum curve through the following steps: The energy spectrum curve is subjected to first-order difference processing to obtain the first-order difference values ​​corresponding to each energy point except the first energy point. The energy point with the largest absolute value of the corresponding first-order difference value is determined as the first energy point. Alternatively, the energy spectrum curve can be subjected to second-order difference processing to obtain the second-order difference values ​​corresponding to each energy point except the first energy point, and the energy point with the largest absolute value of the corresponding second-order difference value can be determined as the first energy point. Alternatively, peak detection can be performed on the energy spectrum curve, and the energy point with the largest corresponding first CT value can be determined as the first energy point.

[0121] In one optional implementation, the image enhancement module 405 is specifically used for: The virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point are subtracted to generate a target area difference image. Determine the contrast-to-noise ratio corresponding to each of the virtual monochrome images, and select the virtual monochrome image with the largest contrast-to-noise ratio from among the virtual monochrome images; The target region difference image and the virtual monochrome image with the largest contrast-to-noise ratio are added together to generate a target region enhancement image.

[0122] In one optional implementation, the image enhancement module 405, when determining the contrast-to-noise ratio corresponding to each of the virtual monochrome images, specifically performs the following: For each of the virtual monochrome images, a first region of interest located within the target area and a second region of interest located within the normal tissue area are determined from the virtual monochrome image; Based on the first CT value corresponding to the first region of interest, the second CT value corresponding to the second region of interest, and the noise value corresponding to the second region of interest, the contrast-to-noise ratio of the virtual monochrome image is determined.

[0123] In one optional implementation, the image acquisition module 401, when acquiring enhanced equivalent image pairs for the target region, is specifically used for: Acquire a pair of non-enhanced virtual monochrome images of the target area during the treatment phase; the pair of non-enhanced virtual monochrome images is used to characterize a pair of virtual monochrome images of the target area without the injection of contrast agent. The unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

[0124] In an optional implementation, the image acquisition module 401 is further configured to: Based on the body part where the target area is located, select the substance decomposition method that matches the target area; The non-enhanced virtual monochrome image pair is subjected to material decomposition processing according to the described material decomposition method to obtain a bone component image; The bone component image is rigidly registered with the planned bone component image to obtain the adjustment parameters of the bone component image relative to the planned bone component image; the planned bone component image is obtained by performing material decomposition processing on the planned non-enhanced virtual monochrome image pair according to the material decomposition method; the planned non-enhanced virtual monochrome image pair is acquired for the target area during the simulation positioning stage; the simulation positioning stage is the stage of simulating target positioning of the human body before radiotherapy; The unenhanced virtual monochrome image pair is adjusted according to the adjustment parameters to obtain a rigidly registered unenhanced virtual monochrome image pair. When the image acquisition module 401 is used to input the non-enhanced virtual monochrome image pair into the image prediction model to obtain the enhanced equivalent image pair corresponding to the non-enhanced virtual monochrome image pair output by the image prediction model, it is specifically used for: The rigidly registered unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

[0125] In an optional embodiment, the target area localization device 400 further includes a model training module 406, which is used to train the image prediction model through the following steps: Acquire pairs of unenhanced virtual monochrome images and pairs of enhanced virtual monochrome images of the sample target area during the simulation positioning phase; the unenhanced virtual monochrome image in the unenhanced virtual monochrome image pair is the virtual monochrome image before the injection of contrast agent into the sample target area, and the enhanced virtual monochrome image in the enhanced virtual monochrome image pair is the virtual monochrome image after the injection of contrast agent into the sample target area. Based on the sample non-enhanced virtual monochrome image pair, image alignment processing is performed on the sample enhanced virtual monochrome image pair to obtain the aligned sample enhanced virtual monochrome image pair; The sample unenhanced virtual monochrome image pairs are input into a neural network for image prediction to obtain the predicted enhanced virtual monochrome image pairs output by the neural network; the neural network includes a contrast agent distribution simulation module and an enhancement mode identification module; Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the target error loss is determined; Based on the target error loss, the parameters of the neural network are adjusted until the neural network meets the training cutoff condition. Based on the contrast agent distribution simulation module included in the neural network when the training cutoff condition is met, the image prediction model is obtained.

[0126] In an optional implementation, the model training module 406, when determining the target error loss based on the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair, is specifically used for: Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the pixel regression loss, adversarial loss, and region of interest loss are determined. Subtraction processing is performed on the unenhanced virtual monochrome image pairs and the aligned enhanced virtual monochrome image pairs to obtain contrast-enhanced residual image pairs. Subtraction processing is performed on the sample unenhanced virtual monochrome image pair and the predicted enhanced virtual monochrome image pair to obtain the predicted contrast-enhanced residual image pair; Based on the sample contrast agent-enhanced residual image pair and the predicted contrast agent-enhanced residual image pair, the iodine distribution sensing loss is determined; The target error loss is determined based on the pixel regression loss, the adversarial loss, the region of interest loss, and the iodine distribution sensing loss.

[0127] In one optional implementation, the model training module 406 is used to determine whether the neural network meets the training cutoff condition through the following steps: If the CT value deviation between the first region of interest located within the target area in the predicted enhanced virtual monochrome image pair and the first region of interest located within the target area in the aligned sample enhanced virtual monochrome image pair meets a preset condition, the target error loss converges, and the peak signal-to-noise ratio of the predicted enhanced virtual monochrome image pair is stable, then the neural network is determined to meet the training cutoff condition.

[0128] In an optional embodiment, the target area localization device 400 further includes a plan generation module 407, the plan generation module 407 being used for: The enhanced target region image is elastically registered, and the enhanced target region image after elastic registration is labeled to obtain the labeling result; the labeling result is used to indicate the location of the target region and the location of the organs at risk. Based on the annotation results, an adaptive radiotherapy treatment plan is generated.

[0129] The specific implementation process of the functions and roles of each module in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0130] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0131] Based on the same technical concept, this application also provides a computer device 500, referring to... Figure 5 The diagram shown is a schematic representation of the structure of a computer device according to an exemplary embodiment of this application, comprising: The processor 510, memory 520, and bus 530 are included. The memory 520 is used to store execution instructions and includes main memory 521 and external memory 522. The main memory 521, also known as internal memory, is used to temporarily store the operation data in the processor 510 and the data exchanged with external memory 522 such as hard disk. The processor 510 exchanges data with external memory 522 through main memory 521.

[0132] In this embodiment, the memory 520 is specifically used to store application code that executes the solution of this application, and its execution is controlled by the processor 510. That is, when the computer device 500 is running, the processor 510 communicates with the memory 520 through the bus 530, or the processor 510 communicates with the memory 520 through other means, so that the processor 510 executes the application code stored in the memory 520, and then executes the steps of the target area localization method described in any of the foregoing embodiments.

[0133] The memory 520 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0134] Processor 510 may be an integrated circuit chip with signal processing capabilities. The aforementioned processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0135] It is understood that the structures illustrated in the embodiments of this application do not constitute a specific limitation on the computer device 500. In other embodiments of this application, the computer device 500 may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0136] This disclosure also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the target localization method described in the above-described method embodiments. The storage medium can be a volatile or non-volatile computer-readable storage medium.

[0137] This disclosure also provides a computer program product, which stores a computer program. When the computer program is run by a processor, it executes the steps of the target area localization method provided in any of the above embodiments of this disclosure. For details, please refer to the above method embodiments, which will not be repeated here.

[0138] The aforementioned computer program product can be implemented through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied in a computer storage medium, which can be a volatile or non-volatile computer-readable storage medium. In another optional embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0139] Furthermore, embodiments of the subject matter and functional operation described in this specification can be implemented in the following ways: digital electronic circuits, tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or combinations thereof. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier for execution by a data processing apparatus or for controlling the operation of a data processing apparatus. Alternatively or additionally, program instructions may be encoded on artificially generated propagation signals, such as machine-generated electrical, optical, or electromagnetic signals, which are generated to encode information and transmit it to a suitable receiving device for execution by the data processing apparatus. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or combinations thereof.

[0140] The processing and logic flow described in this specification can be executed by one or more programmable computers that execute one or more computer programs to perform corresponding functions by operating on input data and generating output. The processing and logic flow can also be executed by dedicated logic circuitry—such as FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits), and the device can also be implemented as dedicated logic circuitry.

[0141] Suitable computers for executing computer programs include, for example, general-purpose and / or special-purpose microprocessors, or any other type of central processing unit. Typically, the central processing unit receives instructions and data from read-only memory and / or random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include one or more mass storage devices for storing data, such as disks, magneto-optical disks, or optical disks, or the computer will be operatively coupled to such mass storage devices to receive data from or transfer data to them, or both. However, a computer is not required to have such devices. Furthermore, a computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name a few.

[0142] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, such as semiconductor memory devices (e.g., EPROM, EEPROM, and flash memory devices), magnetic disks (e.g., internal hard disks or removable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks. Processors and memory may be supplemented by or incorporated into dedicated logic circuitry.

[0143] While this specification contains numerous specific implementation details, these should not be construed as limiting the scope of any invention or the scope of the claims, but rather are primarily intended to describe features of specific embodiments of a particular invention. Certain features described in the various embodiments herein may also be implemented in combination in a single embodiment. Conversely, various features described in a single embodiment may also be implemented separately in various embodiments or in any suitable sub-combination. Furthermore, while features may function in certain combinations as described above and even initially claimed in this way, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may refer to a sub-combination or a variation thereof.

[0144] Similarly, although the operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order shown or sequentially, or requiring all illustrated operations to be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the above embodiments should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

[0145] Thus, specific embodiments of the subject matter have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings are not necessarily shown in a specific order or sequence to achieve the desired result. In some implementations, multitasking and parallel processing may be advantageous.

[0146] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A target area localization method, characterized in that, The method includes: During radiotherapy, enhanced equivalent image pairs are acquired for the target area; these enhanced equivalent image pairs are used to characterize virtual monochrome image pairs when contrast agent is injected into the target area. Based on the enhanced equivalent image pair, multiple virtual monochrome images are generated according to multiple preset single energy values; the multiple virtual monochrome images correspond one-to-one with the multiple preset single energy values. Based on the energy value and the first CT value corresponding to each of the virtual monochrome images, an energy spectrum curve is generated; the energy spectrum curve is used to represent the mapping relationship between the energy value and the first CT value; the first CT value corresponding to the virtual monochrome image is the CT value corresponding to the first region of interest located in the target area of ​​the virtual monochrome image; A first energy point and a second energy point are determined from the energy spectrum curve; the first energy point is used to represent the position in the target area where the contrast changes most drastically with energy, and the second energy point is used to represent the energy point where the CT value deviates the most from the first energy point. Based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point, a target enhancement image is generated; the target enhancement image is used to locate the target area.

2. The method according to claim 1, characterized in that, The first energy point is determined from the energy spectrum curve by the following steps: The energy spectrum curve is subjected to first-order difference processing to obtain the first-order difference values ​​corresponding to each energy point except the first energy point. The energy point with the largest absolute value of the corresponding first-order difference value is determined as the first energy point. Alternatively, the energy spectrum curve can be subjected to second-order difference processing to obtain the second-order difference values ​​corresponding to each energy point except the first energy point, and the energy point with the largest absolute value of the corresponding second-order difference value can be determined as the first energy point. Alternatively, peak detection can be performed on the energy spectrum curve, and the energy point with the largest corresponding first CT value can be determined as the first energy point.

3. The method according to claim 1, characterized in that, The step of generating a target enhancement image based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point includes: The virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point are subtracted to generate a target area difference image. Determine the contrast-to-noise ratio corresponding to each of the virtual monochrome images, and select the virtual monochrome image with the largest contrast-to-noise ratio from among the virtual monochrome images; The target region difference image and the virtual monochrome image with the largest contrast-to-noise ratio are added together to generate a target region enhancement image.

4. The method according to claim 3, characterized in that, Determining the contrast-to-noise ratio corresponding to each of the virtual monochrome images includes: For each of the virtual monochrome images, a first region of interest located within the target area and a second region of interest located within the normal tissue area are determined from the virtual monochrome image; Based on the first CT value corresponding to the first region of interest, the second CT value corresponding to the second region of interest, and the noise value corresponding to the second region of interest, the contrast-to-noise ratio of the virtual monochrome image is determined.

5. The method according to claim 1, characterized in that, The acquisition of enhanced equivalent image pairs for the target region includes: Acquire a pair of non-enhanced virtual monochrome images of the target area during the treatment phase; the pair of non-enhanced virtual monochrome images is used to characterize a pair of virtual monochrome images of the target area without the injection of contrast agent. The unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

6. The method according to claim 5, characterized in that, The method further includes: Based on the body part where the target area is located, select the substance decomposition method that matches the target area; The non-enhanced virtual monochrome image pair is subjected to material decomposition processing according to the described material decomposition method to obtain a bone component image; The bone component image is rigidly registered with the planned bone component image to obtain the adjustment parameters of the bone component image relative to the planned bone component image; the planned bone component image is obtained by performing material decomposition processing on the planned non-enhanced virtual monochrome image pair according to the material decomposition method; the planned non-enhanced virtual monochrome image pair is acquired for the target area during the simulation positioning stage; the simulation positioning stage is the stage of simulating target positioning of the human body before radiotherapy; The unenhanced virtual monochrome image pair is adjusted according to the adjustment parameters to obtain a rigidly registered unenhanced virtual monochrome image pair. The step of inputting the non-enhanced virtual monochrome image pair into the image prediction model to obtain the enhanced equivalent image pair corresponding to the non-enhanced virtual monochrome image pair output by the image prediction model includes: The rigidly registered unenhanced virtual monochrome image pair is input into the image prediction model to obtain the enhanced equivalent image pair corresponding to the unenhanced virtual monochrome image pair output by the image prediction model.

7. The method according to claim 5, characterized in that, The image prediction model is trained using the following steps: Acquire pairs of unenhanced virtual monochrome images and pairs of enhanced virtual monochrome images of the sample target area during the simulation positioning phase; the unenhanced virtual monochrome image in the unenhanced virtual monochrome image pair is the virtual monochrome image before the injection of contrast agent into the sample target area, and the enhanced virtual monochrome image in the enhanced virtual monochrome image pair is the virtual monochrome image after the injection of contrast agent into the sample target area. Based on the sample non-enhanced virtual monochrome image pair, image alignment processing is performed on the sample enhanced virtual monochrome image pair to obtain the aligned sample enhanced virtual monochrome image pair; The sample unenhanced virtual monochrome image pairs are input into a neural network for image prediction to obtain the predicted enhanced virtual monochrome image pairs output by the neural network; the neural network includes a contrast agent distribution simulation module and an enhancement mode identification module; Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the target error loss is determined; Based on the target error loss, the parameters of the neural network are adjusted until the neural network meets the training cutoff condition. Based on the contrast agent distribution simulation module included in the neural network when the training cutoff condition is met, the image prediction model is obtained.

8. The method according to claim 7, characterized in that, The determination of the target error loss based on the predicted enhanced virtual monochrome image pair and the aligned sample enhanced virtual monochrome image pair includes: Based on the predicted enhanced virtual monochrome image pairs and the aligned sample enhanced virtual monochrome image pairs, the pixel regression loss, adversarial loss, and region of interest loss are determined. Subtraction processing is performed on the unenhanced virtual monochrome image pairs and the aligned enhanced virtual monochrome image pairs to obtain contrast-enhanced residual image pairs. Subtraction processing is performed on the sample unenhanced virtual monochrome image pair and the predicted enhanced virtual monochrome image pair to obtain the predicted contrast-enhanced residual image pair; Based on the sample contrast agent-enhanced residual image pair and the predicted contrast agent-enhanced residual image pair, the iodine distribution sensing loss is determined; The target error loss is determined based on the pixel regression loss, the adversarial loss, the region of interest loss, and the iodine distribution sensing loss.

9. The method according to claim 7, characterized in that, The following steps are used to determine whether the neural network meets the training cutoff condition: If the CT value deviation between the first region of interest located within the target area in the predicted enhanced virtual monochrome image pair and the first region of interest located within the target area in the aligned sample enhanced virtual monochrome image pair meets a preset condition, the target error loss converges, and the peak signal-to-noise ratio of the predicted enhanced virtual monochrome image pair is stable, then the neural network is determined to meet the training cutoff condition.

10. The method according to any one of claims 1-9, characterized in that, The method further includes: The enhanced target region image is elastically registered, and the enhanced target region image after elastic registration is labeled to obtain the labeling result; the labeling result is used to indicate the location of the target region and the location of the organs at risk. Based on the annotation results, an adaptive radiotherapy treatment plan is generated.

11. A target area positioning device, characterized in that, The device includes: An image acquisition module is used to acquire enhanced equivalent image pairs of a target area during radiotherapy; the enhanced equivalent image pairs are used to characterize virtual monochrome image pairs when a contrast agent is injected into the target area; The image generation module is used to generate multiple virtual monochrome images based on the enhanced equivalent image pair and according to multiple preset single energy values; the multiple virtual monochrome images correspond one-to-one with the multiple preset single energy values; The curve generation module is used to generate an energy spectrum curve based on the energy value and the first CT value corresponding to each of the virtual monochrome images; the energy spectrum curve is used to represent the mapping relationship between the energy value and the first CT value; the first CT value corresponding to the virtual monochrome image is the CT value corresponding to the first region of interest located in the target area of ​​the virtual monochrome image; An energy selection module is used to determine a first energy point and a second energy point from the energy spectrum curve; the first energy point is used to represent the position in the target area where the contrast changes most drastically with energy, and the second energy point is used to represent the energy point where the CT value deviates the most from the first energy point. The image enhancement module is used to generate a target enhancement image based on the virtual monochrome image corresponding to the first energy point and the virtual monochrome image corresponding to the second energy point; the target enhancement image is used to locate the target area.

12. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the target area localization method according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the target area localization method according to any one of claims 1 to 10.

Images