Analysis method and system for multi-tracer image, storage medium and computer equipment

Abstract

The invention discloses an analysis method and system for a multi-tracer image, a storage medium and computer equipment, and the method comprises the steps: obtaining PET images and CT images of a target object under each tracer, carrying out the lesion segmentation of each PET image to obtain a lesion segmentation result, and carrying out the organ segmentation of each CT image to obtain an organ segmentation result; performing registration processing on the CT image and the PET image through a unified CT coordinate system to obtain a primary registration result, and performing secondary registration on the primary registration result based on each organ segmentation result to obtain a final registration result; determining a development overlapping region of the final registration result, and calculating a first index parameter according to the development overlapping region, each focus segmentation result and each organ segmentation result; calculating a second index parameter according to the tracer developing area of each PET image, the focus segmentation result and the matched organ segmentation result; and performing state analysis on the target object according to the first index parameter and each second index parameter.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method and system for analyzing multi-tracer images, a storage medium, and a computer device. Background Technology

[0002] In the field of medical imaging diagnosis, with the rapid development of medical imaging technology, imaging technologies such as positron emission tomography (PET) and computed tomography (CT) have become important tools for clinical diagnosis and disease research. PET images can reflect the metabolic, physiological, and biochemical processes in an organism by observing the distribution of tracers within the body, providing crucial information for early disease diagnosis, staging, and treatment efficacy evaluation. CT images, on the other hand, have high spatial resolution and can clearly display the anatomical structures of human organs and tissues. In practical clinical applications, to gain a more comprehensive understanding of the occurrence and development of diseases and to evaluate treatment effects, it is often necessary to use multiple tracers to image the same target object, obtaining multi-tracer images. Different tracers have different properties and can reflect the lesion from different perspectives. For example, some tracers are sensitive to the metabolic activity of tumor cells, while others have a better tracking effect on the blood perfusion of specific tissues. Comprehensive analysis of multi-tracer images can provide doctors with richer and more accurate diagnostic evidence, helping to improve the accuracy of disease diagnosis and the level of treatment efficacy evaluation.

[0003] However, the analysis and processing of multi-tracer images still faces many challenges. On the one hand, due to factors such as the subject's position and respiration during imaging with different tracers, spatial discrepancies may exist between PET and CT images. Furthermore, positional mismatches may also occur between PET images using different tracers. This spatial inconsistency severely impacts subsequent comprehensive analysis of multi-tracer images, making it impossible to accurately obtain imaging information of each tracer at the same anatomical location, thus affecting the accurate assessment of disease status. On the other hand, in image analysis, most existing methods only perform simple lesion or organ segmentation on single-tracer images, lacking in-depth mining and comprehensive utilization of the correlation information between multi-tracer images, directly affecting the final analysis results. Summary of the Invention

[0004] In view of this, this application provides a method and system for analyzing multi-tracer images, a storage medium, and a computer device. Through two registration processes—first, primary registration based on a unified CT coordinate system, and then secondary registration using organ segmentation results—the accuracy of image registration is ensured, laying the foundation for accurate subsequent analysis. By processing image data under multiple tracers, the information provided by different tracers is fully utilized to calculate indicator parameters at multiple levels, including systemic, organ, and lesion levels. These indicator parameters at different levels can complement each other, comprehensively reflecting the state of the target object from macroscopic to microscopic perspectives, improving the comprehensiveness and accuracy of the analysis, and providing doctors with more comprehensive decision-making basis. Advanced image processing and analysis algorithms are employed to automate and intelligently calculate image segmentation, registration, and indicator parameters. This not only improves analysis efficiency but also reduces interference from human factors, enhancing the reliability and repeatability of the analysis results.

[0005] According to one aspect of this application, a method for analyzing multi-tracer images is provided, comprising:

[0006] Acquire PET images of the target object under each tracer and CT images matching the PET images, and perform lesion segmentation on the PET images under each tracer to obtain the lesion segmentation results corresponding to each PET image, and perform organ segmentation on the CT images under each tracer to obtain the organ segmentation results corresponding to each CT image.

[0007] Based on the CT images under each tracer, the entire CT image and PET image are registered using a unified CT coordinate system to obtain the primary registration result. Then, based on the organ segmentation results corresponding to the CT images under each tracer, the primary registration result is further registered to obtain the final registration result.

[0008] Based on the final registration result, the overlapping regions of each tracer on the final registration result are determined, and a first index parameter of the target object is calculated according to the overlapping regions, the lesion segmentation results of each PET image, and the organ segmentation results of each CT image. The first index parameter includes a first systemic index parameter, a first organ-level index parameter, and a first lesion-level index parameter. Furthermore, for each PET image under each tracer, a second index parameter of the target object under the tracer is calculated according to the tracer-developed region of the PET image, the lesion segmentation result, and the organ segmentation result of the CT image matching the PET image. The second index parameter includes a second systemic index parameter, a second organ-level index parameter, and a second lesion-level index parameter.

[0009] Based on the first indicator parameter and the second indicator parameter corresponding to each tracer, the state of the target object is analyzed.

[0010] According to another aspect of this application, an analysis system for multi-tracer images is provided, comprising:

[0011] The image registration module is used to acquire PET images of the target object under each tracer and CT images matching the PET images, and to perform lesion segmentation on the PET images under each tracer to obtain lesion segmentation results for each PET image, and to perform organ segmentation on the CT images under each tracer to obtain organ segmentation results for each CT image; based on the CT images under each tracer, the module performs registration processing on all CT images and PET images through a unified CT coordinate system to obtain a primary registration result, and performs secondary registration processing on the primary registration result based on the organ segmentation results corresponding to the CT images under each tracer to obtain a final registration result;

[0012] The image analysis module is used to determine the overlapping regions of each tracer on the final registration result, and calculate a first index parameter of the target object based on the overlapping regions, lesion segmentation results of each PET image, and organ segmentation results of each CT image. The first index parameter includes a first systemic index parameter, a first organ-level index parameter, and a first lesion-level index parameter. Furthermore, for each tracer-treated PET image, based on the tracer-treated regions and lesion segmentation results of the PET image, and the organ segmentation results of the CT image matching the PET image, the module calculates a second index parameter of the target object under the tracer. The second index parameter includes a second systemic index parameter, a second organ-level index parameter, and a second lesion-level index parameter. Based on the first index parameter and the second index parameter corresponding to each tracer, the module performs a state analysis on the target object.

[0013] According to another aspect of this application, a storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described method for analyzing multi-tracer images.

[0014] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-described method for analyzing multiple tracer images.

[0015] By employing the above technical solutions, this application provides a method and system for analyzing multi-tracer images, a storage medium, and a computer device. Through two registration processes—first, primary registration based on a unified CT coordinate system, and then secondary registration using organ segmentation results—the accuracy of image registration is ensured, laying the foundation for accurate subsequent analysis. By processing image data under multiple tracers, the information provided by different tracers is fully utilized to calculate indicator parameters at multiple levels, including systemic, organ, and lesion levels. These indicator parameters at different levels can complement each other, comprehensively reflecting the state of the target object from macroscopic to microscopic perspectives, improving the comprehensiveness and accuracy of the analysis, and providing doctors with more comprehensive decision-making support. Advanced image processing and analysis algorithms are adopted to automate and intelligently calculate image segmentation, registration, and indicator parameters. This not only improves analysis efficiency but also reduces interference from human factors, enhancing the reliability and repeatability of the analysis results.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0018] Figure 1 A schematic flowchart of an analysis method for multiple tracer images provided in an embodiment of this application is shown;

[0019] Figure 2 A schematic diagram of the structure of an analysis system for multi-tracer images provided in an embodiment of this application is shown;

[0020] Figure 3 A schematic diagram of the device structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0021] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0022] This embodiment provides a method for analyzing multi-tracer images, such as... Figure 1 As shown, the method includes:

[0023] Step 101: Obtain PET images of the target object under each tracer and CT images matching the PET images, and perform lesion segmentation on the PET images under each tracer to obtain lesion segmentation results corresponding to each PET image, and perform organ segmentation on the CT images under each tracer to obtain organ segmentation results corresponding to each CT image.

[0024] Step 102: Based on the CT images under each tracer, the entire CT image and PET image are registered using a unified CT coordinate system to obtain a primary registration result. Based on the organ segmentation results corresponding to the CT images under each tracer, the primary registration result is then registered a second time to obtain the final registration result.

[0025] Step 103: Based on the final registration result, determine the overlapping regions of each tracer on the final registration result, and calculate the first index parameter of the target object according to the overlapping regions, the lesion segmentation results of each PET image, and the organ segmentation results of each CT image. The first index parameter includes a first systemic index parameter, a first organ-level index parameter, and a first lesion-level index parameter. Furthermore, for each tracer-treated PET image, calculate the second index parameter of the target object under the tracer according to the tracer-treated region of the PET image, the lesion segmentation result, and the organ segmentation result of the CT image matching the PET image. The second index parameter includes a second systemic index parameter, a second organ-level index parameter, and a second lesion-level index parameter.

[0026] Step 104: Perform state analysis on the target object based on the first index parameter and the second index parameter corresponding to each tracer.

[0027] This application provides an analysis method for multi-tracer images, which can be applied to comprehensive medical image processing and analysis platforms. By performing in-depth processing and analysis on PET images under multiple tracers and their matching CT images, it provides a comprehensive and accurate assessment of the target object's status for clinical diagnosis, drug development, and biomedical research. The method sequentially completes image registration and image analysis based on the registration results, ensuring that valuable information is extracted from the original image data.

[0028] Specifically, PET images of the target object under each tracer and corresponding CT images (in the same orientation) are first acquired from relevant medical imaging equipment. PET images can reflect the distribution of the tracer in the body by detecting the radioactive decay signal of the tracer, thereby indirectly displaying information such as tissue metabolism and function. CT images, on the other hand, utilize the attenuation characteristics of X-rays after penetrating the human body to provide detailed anatomical information, including the morphology, location, and size of organs. The combination of both provides more comprehensive information for subsequent analysis. For example, in tumor diagnosis, PET images can show the metabolic activity of the tumor, while CT images can clearly define the location of the tumor and its relationship with surrounding tissues.

[0029] Next, lesion segmentation can be performed on PET images under each tracer to locate the lesion region. Specifically, advanced image segmentation algorithms can be used for lesion segmentation, such as threshold-based segmentation methods, region-based segmentation methods, or deep learning-based segmentation methods. For example, deep learning-based segmentation methods can train a large amount of labeled data, allowing the model to automatically learn the features of the lesion, thereby achieving accurate lesion segmentation. The segmented lesion results will be used for subsequent parameter calculations and state analysis.

[0030] Simultaneously, organ segmentation can be performed on CT images under each tracer, providing anatomical references for subsequent registration and analysis. Organ segmentation also requires image segmentation algorithms, but unlike lesion segmentation, organ segmentation needs to more accurately identify and separate different organs. For example, in chest CT images, organs such as the lungs, heart, and liver need to be accurately segmented. The quality of the organ segmentation results will directly affect the accuracy of registration and subsequent analysis.

[0031] After obtaining the segmentation results for each PET and CT image, the next step is to perform registration processing on these images. Specifically, a unified CT coordinate system can be used to register all CT and PET images, yielding a preliminary registration result. A unified CT coordinate system ensures the spatial comparability of different images. Since images acquired at different times or using different devices may have spatial differences, a unified coordinate system aligns these images to the same reference frame, enabling accurate subsequent analysis. During preliminary registration using a unified CT coordinate system, algorithms such as rigid registration can be used.

[0032] Next, based on the organ segmentation results corresponding to the CT images under each tracer, a secondary registration process is performed on the primary registration results to obtain the final registration result. Using the organ segmentation results for secondary registration can further improve the registration accuracy. This is because organ segmentation results can more accurately reflect the position and shape of anatomical structures. By using the organ segmentation results as a constraint condition for registration, errors that may exist in the primary registration can be eliminated, especially in areas with significant changes in organ boundaries or morphology.

[0033] After obtaining the final registration result, the overlapping regions of each tracer in the final registration result are determined. The overlapping regions reflect the common action areas of different tracers in vivo, which is of great significance for understanding the interrelationships between different tracers and the overall state of the target object.

[0034] Furthermore, based on the overlapping areas of contrast imaging, the lesion segmentation results of each PET image, and the organ segmentation results of each CT image, the first index parameters of the target object are calculated. Here, the first index parameters include the first systemic index parameter, the first organ-level index parameter, and the first lesion-level index parameter. Among them, the systemic index parameter can reflect the overall metabolic activity distribution and tracer uptake of the target object macroscopically; the organ-level index parameter can assess the functional status and metabolic level of a specific organ; and the lesion-level index parameter can describe the characteristics of the lesion, such as its size, shape, and metabolic activity. These index parameters reflect the comprehensive situation of the target object under the action of multiple tracers from different levels (systemic, organ, and lesion).

[0035] Furthermore, for each tracer-induced PET image, a second index parameter of the target object under that tracer can be calculated based on the tracer-enhanced area and lesion segmentation results of the PET image, and the organ segmentation results of the CT image matching the PET image. Here, the second index parameter includes a second systemic index parameter, a second organ-level index parameter, and a second lesion-level index parameter. These index parameters focus on the individual effects of each tracer, helping to understand the characteristics of the effects of different tracers on the target object. For example, different tracers may exhibit different uptake characteristics in different organs or lesions; calculating the second index parameter can provide a deeper understanding of these differences.

[0036] Subsequently, a state analysis of the target subject can be performed based on the primary indicator parameter and the secondary indicator parameters corresponding to each tracer. By comprehensively analyzing these multi-level indicator parameters, the physiological or pathological state of the target subject can be assessed more comprehensively and accurately. For example, in tumor diagnosis, combining systemic, organ-level, and lesion-level indicator parameters can determine the severity, progression, and presence of metastasis of the tumor. The results of the state analysis can support clinical decision-making. Doctors can develop personalized treatment plans, evaluate treatment effectiveness, or predict disease prognosis based on the analysis results. For example, if the analysis results show a significant decrease in tumor metabolic activity, it may indicate that the treatment plan is effective; if new lesions are found or lesions enlarge, the treatment plan may need to be adjusted.

[0037] In one specific embodiment, to facilitate doctors' understanding and use of the analysis results, the results can be visualized, such as by generating images and charts. Simultaneously, detailed reports can be generated, including the values ​​of indicator parameters, conclusions of the status analysis, and recommendations, providing intuitive references for clinical decision-making.

[0038] By applying the technical solution of this embodiment, a two-stage registration process is used: first, primary registration is performed based on a unified CT coordinate system, and then secondary registration is performed using organ segmentation results. This ensures the accuracy of image registration and lays the foundation for accurate subsequent analysis. By processing image data under multiple tracers, the information provided by different tracers is fully utilized to calculate indicator parameters at multiple levels, including systemic, organ, and lesion levels. The indicator parameters at different levels can complement each other, comprehensively reflecting the state of the target object from macroscopic to microscopic perspectives, improving the comprehensiveness and accuracy of the analysis, and providing doctors with more comprehensive decision-making basis. Advanced image processing and analysis algorithms are adopted to automate and intelligently calculate image segmentation, registration, and indicator parameters. This not only improves analysis efficiency but also reduces interference from human factors, improving the reliability and repeatability of analysis results.

[0039] Optionally, in this embodiment, step 102, "based on the CT images under each tracer, registering all CT images and PET images using a unified CT coordinate system to obtain a preliminary registration result," includes: taking any CT image from the CT images under each tracer as a reference image; calculating the image deviation between each target CT image in the remaining CT images and the reference image using a unified CT coordinate system; and calculating the adjustment parameters corresponding to the target CT image based on the image deviation, the adjustment parameters including translation parameters, scaling parameters, and rotation parameters; for each target CT image, sequentially performing image processing on the target CT image and the PET image matching the target CT image according to the adjustment parameters corresponding to each target CT image; and comparing the processed image with the reference image and the reference image. Alignment is performed with the corresponding PET images. After the processed images corresponding to each target CT image are aligned, the primary registration result is obtained. Step 102, "based on the organ segmentation results corresponding to the CT images under each tracer, a secondary registration process is performed on the primary registration result to obtain the final registration result", includes: for each processed target CT image in the primary registration result, based on the organ segmentation results of the processed target CT image and the organ segmentation results of the reference image, the mapping relationship between each pixel in the processed target CT image and the corresponding pixel in the reference image is calculated, and the processed target CT image and the corresponding processed PET image are re-registered according to the mapping relationship. After each processed target CT image and the corresponding processed PET image are re-registered, the final registration result is obtained.

[0040] In this embodiment, during primary registration based on CT images from each tracer, one image is first selected as the reference image. The selection of the reference image is flexible; specifically, it can be an image with good quality, clear anatomical structure display, and representativeness. For example, if one of the multi-tracer CT images has low noise, moderate contrast, and clearly displays the boundaries of major organs during acquisition, then it can be used as the reference image.

[0041] Next, using the reference image as a reference, the image deviation between the remaining CT images (each of which can be referred to as the target CT image) and the reference image is calculated in a unified CT coordinate system. This process can employ feature matching methods from image registration algorithms. For example, feature points (such as corner points and edge points) in the image can be extracted, and then the positional differences of these feature points in the target CT image and the reference image can be calculated to obtain the image deviation. These deviations reflect the spatial differences between the target CT image and the reference image, including changes in translation, rotation, and scaling. Based on the calculated image deviations, the adjustment parameters corresponding to the target CT image are further calculated. These parameters include translation parameters, scaling parameters, and rotation parameters. Translation parameters describe the amount of positional movement of the target CT image in the X, Y, and Z directions; scaling parameters adjust the size ratio of the target CT image in each direction; and rotation parameters determine the rotation angle of the target CT image around different axes. In a unified CT coordinate system, each feature point has a definite coordinate position. By comparing the coordinates of corresponding feature points in the target CT image and the reference image, their positional differences in the X, Y, and Z directions can be calculated. These differences are the image deviations. For example, if a feature point has coordinates (10, 20, 30) in the reference image and coordinates (12, 22, 32) in the target CT image, then its deviation in the X direction is 2, its deviation in the Y direction is 2, and its deviation in the Z direction is 2.

[0042] There are various methods for calculating adjustment parameters, with the least squares-based optimization algorithm being a common one. This algorithm finds the optimal adjustment parameters by minimizing the differences between the target CT image and the reference image (such as differences in pixel values, distances between feature points, etc.). For example, when calculating translation parameters, the optimal translation parameters can be obtained by continuously adjusting the position of the target CT image to minimize the difference in pixel values ​​at corresponding positions between the target CT image and the reference image.

[0043] For each target CT image, image processing is performed sequentially on the target CT image and its corresponding PET image according to their corresponding adjustment parameters. For example, the image is moved in space according to translation parameters, resized according to scaling parameters, and rotated according to rotation parameters. Next, the processed target CT image and its corresponding PET image are aligned with the reference image and its corresponding PET image. The purpose of alignment is to ensure that images under different tracers are spatially consistent for subsequent analysis and comparison. Once all processed images corresponding to the target CT images have been aligned, the primary registration result is obtained. The primary registration result eliminates spatial differences between different CT images to some extent, but some errors may still exist due to factors such as changes in organ morphology.

[0044] Furthermore, for each processed target CT image in the initial registration result, the mapping relationship between each pixel in the processed target CT image and its corresponding pixel in the reference image can be calculated based on the organ segmentation results of the processed target CT image and the reference image. The organ segmentation results provide precise boundary information of different organs in the image. By comparing the morphology and position of organs in the processed target CT image and the reference image, the correspondence between pixels can be determined more accurately. The mapping relationship can be calculated using a deformation model-based method. The deformation model can perform non-rigid deformation on the image based on the anatomical features and physical properties of the organs, resulting in a better match between the organs in the processed target CT image and the organs in the reference image. For example, by optimizing the parameters of the deformation model, the similarity between corresponding organs in the processed target CT image and the reference image can be maximized, thereby obtaining the mapping relationship for each pixel.

[0045] Based on the calculated mapping relationship, the processed target CT image and its corresponding processed PET image are re-registered. Unlike primary registration, secondary registration is a non-rigid registration, which can better handle situations with significant changes in organ morphology. By performing non-rigid deformation on the images, the processed target CT and PET images are more anatomically consistent with the reference image. Once all processed target CT images and their corresponding processed PET images have been re-registered, the final registration result is obtained. The final registration result has higher accuracy and can more accurately reflect the spatial correspondence between images under different tracers, providing a reliable basis for subsequent image analysis and condition assessment.

[0046] This application's embodiments combine primary and secondary registration. First, by unifying the CT coordinate system, most spatial differences are eliminated. Then, non-rigid registration is performed using organ segmentation results, further improving image registration accuracy. This multi-level registration method can better handle complex variations between images under different tracers, reducing registration errors. Furthermore, this registration method can adapt to the characteristics of images under different tracers; image differences caused by different acquisition times, equipment variations, or physiological changes can all be corrected through registration processing.

[0047] Optionally, in this embodiment of the application, step 103, "determining the imaging overlap region of each tracer on the final registration result based on the final registration result," includes: determining the pixel set corresponding to the lesion location in the PET image after final registration for each tracer based on the final registration result, wherein the pixel set includes the coordinates of each pixel point contained in the lesion location; extracting common pixels in each pixel set based on the pixel set corresponding to each tracer, and determining the imaging overlap region of the final registration result based on the common pixels.

[0048] In this embodiment, the final registration result is obtained after primary and secondary registration processes. This eliminates spatial differences between images under different tracers, resulting in a high degree of consistency in anatomical structure across different images. After obtaining the final registration result, the lesion location in the finally registered PET image for each tracer can be determined based on this result. The coordinates of each pixel within all lesion locations in the PET image are obtained. In a digital image, each pixel has a unique coordinate location, which can be accurately obtained through image processing algorithms. Then, the coordinates of all pixels within these lesion locations are collected to form a pixel set. This pixel set represents the specific location and extent of the lesion under that tracer in the image. For example, for a circular lesion with a diameter of 1 cm, its pixel set will contain the coordinates of all pixels within that circular area.

[0049] Next, by comparing the pixel sets corresponding to different tracers, the common pixels are identified. This can be achieved programmatically, for example, using the intersection operation of sets. Treating the pixel set of each tracer as a set, the set of common pixels can be obtained by calculating the intersection of these sets. For example, if there are three tracers A, B, and C, with corresponding pixel sets S... A S B S C Then the set of pixels in total is S. A ∩S B ∩S CIn other words, common pixels refer to pixels present in the pixel sets corresponding to multiple tracers. These pixels represent areas that show abnormal signals under different tracers, and may be the core area of ​​a lesion or areas with common physiological and pathological characteristics. For example, in tumor diagnosis, different tracers may reflect the metabolic activity and blood perfusion of the tumor, while common pixels may correspond to areas with dense tumor cells, vigorous metabolism, and rich blood flow. In a specific embodiment, to reduce computational load, the pixel set can be preprocessed before comparison, such as removing some pixels that obviously do not belong to the lesion area, or downsampling the pixel set. Downsampling can reduce the number of pixels while maintaining a certain level of accuracy, thereby improving the comparison speed.

[0050] Subsequently, based on the extracted common pixels, the imaging overlap region of the final registration result is determined. The imaging overlap region refers to the area that shows imaging signals under different tracers, reflecting the common range of action of different tracers at the lesion site. For example, in tumor diagnosis, the imaging overlap region may correspond to an area with a higher degree of malignancy and a poorer prognosis. Specifically, the boundary of the imaging overlap region can be determined by connecting the common pixels. Commonly used methods include convex hull algorithms and boundary tracking algorithms. The convex hull algorithm can enclose the common pixels within a minimal convex polygon, which is the approximate boundary of the imaging overlap region. The boundary tracking algorithm, on the other hand, gradually determines the boundary of the region by tracking the distribution of the common pixels.

[0051] Once the overlapping areas of contrast imaging are identified, further analysis can be performed on these areas to extract their characteristic information. For example, the area, volume, and average radioactive uptake of the region can be calculated. This characteristic information can be used to assess the nature and severity of the lesion.

[0052] The embodiments of this application use different tracers to reflect the physiological and pathological processes in the body from different perspectives. By determining the overlapping areas of imaging, this information can be integrated, which can more accurately identify the core area of ​​the lesion and areas with common characteristics, thus helping to improve the accuracy of disease diagnosis.

[0053] In this embodiment of the application, optionally, step 103, "calculating the first index parameter of the target object based on the contrast-enhanced overlapping region, the lesion segmentation results of each PET image, and the organ segmentation results of each CT image," includes: calculating the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to the contrast-enhanced overlapping region based on the contrast-enhanced overlapping region, and using the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis as the first systemic index parameter; determining the organ registration results of each CT image based on the final registration results, and dividing the contrast-enhanced overlapping region into regions based on the organ registration results to obtain the first overlapping sub-region corresponding to each organ, and calculating the first overlapping sub-region based on the first overlapping sub-region corresponding to each organ. The mean SUV, maximum SUV, FDG-metabolized tumor volume, and total lesion glycolysis corresponding to each overlapping sub-region are used as the first organ-level index parameters. Based on the final registration result, the lesion registration result of each PET image is determined, and the imaging overlapping region is divided into regions according to the lesion registration result to obtain the second overlapping sub-region corresponding to each lesion. Based on the second overlapping sub-region corresponding to each lesion, the mean SUV, maximum SUV, FDG-metabolized tumor volume, and total lesion glycolysis corresponding to the second overlapping sub-region are calculated. The mean SUV, maximum SUV, FDG-metabolized tumor volume, and total lesion glycolysis corresponding to each second overlapping sub-region are used as the first lesion-level index parameters.

[0054] In this embodiment, the first systemic indicator parameters were selected based on the following criteria: the mean SUV (Standardized Uptake Value), the maximum SUV, the FDG-metabolic tumor volume (MTV), and the total lesion glycolysis (TLG). These parameters reflect the metabolic characteristics of the contrast-overlapping region from different perspectives. SUV measures the extent of tracer uptake by the tissue; the mean reflects the average metabolic level of the entire contrast-overlapping region, while the maximum represents the highest metabolic activity within that region. MTV represents the metabolically active tumor volume in the contrast-overlapping region, and TLG is the product of the mean SUV and MTV. By comprehensively considering both metabolic intensity and tumor volume, TLG provides a more complete reflection of the tumor's metabolic burden.

[0055] The SUV mean is calculated as follows: First, obtain the SUV values ​​of all pixels within the overlapping development area. Then, sum these SUV values ​​and divide by the total number of pixels to obtain the SUV mean. For example, if the overlapping development area has n pixels, the corresponding SUV values ​​are SUV1, SUV2, ..., SUV2. n Then the average value of SUVs = (SUV1 + SUV2 + ... + SUV n ) / n.

[0056] SUV maximum value calculation: The maximum value of SUV is found directly in the overlapping area of ​​the developing region.

[0057] FDG-mediated tumor volume calculation: A threshold method can be used to determine the median volume (MTV). An appropriate SUV threshold is set, and pixels within the overlapping imaging region with SUV values ​​higher than this threshold are considered metabolically active tumor tissue. The number of these pixels is counted, and then the MTV is calculated based on the physical dimensions of the pixels (e.g., pixel pitch). For example, if the actual volume represented by each pixel is V, and the number of pixels within the overlapping imaging region with SUV values ​​higher than the threshold is m, then MTV = m × V.

[0058] Total lesion glycolysis calculation: TLG = mean SUV × MTV. TLG is obtained by multiplying the mean SUV by the MTV. This index comprehensively reflects the metabolic intensity and tumor volume in the contrast-overlapping region.

[0059] Furthermore, the first organ-level index parameters can be calculated. Specifically, the organ registration results for each CT image can be determined based on the final registration results. The purpose of organ registration is to ensure that the position and shape of the same organ are consistent in different CT images, providing an accurate basis for subsequent region division. For example, in chest CT images, organ registration can accurately identify the position and boundaries of organs such as the heart, lungs, and liver. Then, based on the organ registration results, the overlapping regions are divided to obtain the first overlapping sub-region corresponding to each organ. This can be achieved by determining whether the pixels within the overlapping regions belong to the registration region of a certain organ. For example, for the registration region corresponding to the heart, the pixels within the overlapping regions are assigned to the first overlapping sub-region of the heart.

[0060] Similar to the calculation of the first systemic level parameters, the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis are calculated for each organ based on its corresponding first overlapping sub-region. These parameters reflect the metabolic characteristics of the overlapping regions within different organs, helping to understand metabolic changes in different organs during disease progression. For example, in cancer patients, the overlapping regions within different organs may reflect tumor metastasis; by calculating the parameters for each organ, the metabolic activity and burden of the tumor in different organs can be assessed.

[0061] In addition, first-level lesion parameters can be calculated. Specifically, the lesion registration results for each PET image can be determined based on the final registration results. It's important to note that the lesion registration results do not mean that lesions in different PET images are completely identical, but rather that after final registration, the co-occurrence positions of each lesion in the same coordinate space can be a collection of lesions from different PET images. Next, based on the lesion registration results, the overlapping regions are divided to obtain a second overlapping sub-region corresponding to each lesion. This can be achieved by determining whether pixels within the overlapping regions belong to a specific lesion region. For example, for a tumor lesion in the liver, pixels within the overlapping regions that are located within the lesion's registration region are assigned to the second overlapping sub-region of that lesion. Based on the second overlapping sub-region corresponding to each lesion, the mean SUV, maximum SUV, FDG metabolic tumor volume, and total lesion glycolysis are calculated. These indicators reflect the metabolic characteristics of individual lesions and help assess the malignancy, growth activity, and response to treatment.

[0062] This application's embodiments calculate indicator parameters at three levels: systemic, organ, and lesion. This allows for a comprehensive and in-depth understanding of the target subject's metabolic characteristics and disease status. The indicator parameters at different levels complement each other, providing richer information for clinical diagnosis and treatment. By calculating specific indicator parameters, quantitative assessment of contrast-enhanced areas is achieved. Compared to traditional qualitative analysis, quantitative assessment is more objective and accurate, contributing to improved diagnostic accuracy and evaluation of treatment efficacy.

[0063] Optionally, in this embodiment, PET images under each tracer are segmented based on a pre-defined lesion segmentation model. The pre-defined lesion segmentation model for each tracer is constructed as follows: multiple PET image samples corresponding to the target tracer are obtained, each PET image sample is labeled with a lesion, and each lesion is labeled with a lesion tag; the nnU-Net framework is invoked, and multi-dimensional features are extracted from each PET image sample using the nnU-Net framework; based on the multi-dimensional feature extraction results, the network depth, network width, encoder structure, and decoder structure of the initial lesion segmentation model are determined; and based on the network depth, network width, encoder structure, and decoder structure, a new lesion segmentation model is constructed. An initial lesion segmentation model is established. Each PET image sample is downsampled to obtain a downsampled sample. The initial lesion segmentation model is then trained based on the downsampled samples to obtain an intermediate lesion segmentation model. Each PET image sample is input into the intermediate lesion segmentation model to obtain a coarse-grained lesion segmentation map corresponding to each PET image sample. The coarse-grained lesion segmentation map is then upsampled to obtain an upsampled sample corresponding to each PET image sample. Based on the upsampled samples corresponding to each PET image sample, the intermediate lesion segmentation model is retrained to obtain the preset lesion segmentation model.

[0064] In this embodiment, PET images under each tracer have unique imaging characteristics and lesion features. For example, different tracers have different distribution and metabolic pathways in vivo, resulting in differences in the appearance of lesions in the images (such as signal intensity, morphology, etc.). Therefore, it is necessary to construct a preset lesion segmentation model for each tracer. The preset lesion segmentation model corresponding to each tracer can be constructed through the following steps: First, obtain multiple PET image samples corresponding to the target tracer. These samples can cover different cases, different lesion types, and lesion degrees to ensure the diversity and representativeness of the samples. For example, in tumor diagnosis, the samples should include tumor lesions of different sizes and locations, as well as tumor images of different pathological types. Diverse samples help the model learn a wider range of features and improve the model's generalization ability. In addition, the lesions in each PET image sample are accurately labeled to clarify the location and extent of the lesions. This can be done by professional medical experts to ensure the accuracy of the labeling. For example, in PET images, lesions may appear as areas of increased radioactive uptake, and experts can manually delineate the boundaries of the lesions using image analysis software. Appropriate labels can be designed for each lesion, and these labels should accurately reflect the characteristics and nature of the lesion. For example, in tumor diagnosis, labels can include information such as tumor type (e.g., benign, malignant) and grade (e.g., low-grade, high-grade). The design of the labels should take into account the needs of subsequent model training and evaluation, ensuring that the labels have clear semantics and are distinguishable.

[0065] Next, an initial lesion segmentation model can be constructed using nnU-Net. nnU-Net is a deep learning-based medical image segmentation framework with advantages such as automatic adjustment of network structure and optimization of training parameters, enabling it to adapt to different types and scales of medical image data. It can automatically learn multi-dimensional features of the image based on the input PET image samples, including spatial and textural features. Specifically, feature extraction is performed on each PET image sample using the nnU-Net framework. The framework can perform operations such as convolution and pooling on the image to progressively extract features at different levels. These features will be used for subsequent model structure determination and training.

[0066] Based on the multidimensional feature extraction results, the network depth and width of the initial lesion segmentation model can be determined. Network depth refers to the number of convolutional layers in the network; deeper networks can learn more complex features, but may also lead to problems such as vanishing gradients or overfitting. Network width refers to the number of kernels in each convolutional layer; wider networks can extract more features, but will also increase the computational cost of the model. By analyzing the feature extraction results, an appropriate network depth and width are selected to balance model performance and computational efficiency. The final network depth of the initial lesion segmentation model can be determined as follows: construct networks of different depths, plot training and validation loss curves, and calculate the segmentation accuracy, Dice coefficient, and other metrics on the validation set to evaluate the performance of networks of different depths. The network depth that best performs on the validation set, exhibits a stable decreasing trend in training and validation losses, high segmentation accuracy and Dice coefficient, and is also reasonable in terms of computational resources and training efficiency is selected as the final network depth of the initial lesion segmentation model. The final network width of the initial lesion segmentation model can be determined as follows: Construct networks of different widths to determine the network width at which the model achieves ideal segmentation accuracy and Dice coefficient on the validation set, which is sufficient to extract enough features for accurate lesion segmentation without overfitting.

[0067] Furthermore, the encoder and decoder structures can be determined. The encoder is typically used to extract image features, progressively compressing the input image into a low-dimensional feature representation; the decoder is used to restore the segmentation result to the original image size from the low-dimensional feature representation. The structural design of the encoder and decoder can be found in common architectures such as U-Net and ResNet. Specifically, a suitable encoder and decoder can be determined by validating the lesion segmentation performance of different common structures.

[0068] Once the network depth, network width, encoder structure, and decoder structure are determined, an initial lesion segmentation model can be built.

[0069] Next, each PET image sample can be downsampled to obtain a downsampled sample for each PET image sample. The purpose of downsampling is to reduce the image size and computational cost, thus speeding up model training. Simultaneously, the downsampled image can highlight the main features of the image, helping the model learn more robust features. Common downsampling methods include max pooling and average pooling. During the downsampling process, an appropriate downsampling ratio can be selected to ensure that the downsampled image still retains sufficient information for model training.

[0070] Subsequently, the initial lesion segmentation model is trained based on downsampled samples. During training, downsampled samples are input into the model, and the model outputs predicted lesion segmentation results, which are then compared with the actual lesion segmentation results to calculate the loss function. The model parameters are updated using the backpropagation algorithm to minimize the loss function, thereby improving the model's segmentation accuracy. When the model loss value of the initial lesion segmentation model is less than a first preset loss threshold, an intermediate lesion segmentation model is obtained.

[0071] Next, the intermediate lesion segmentation model can be retrained. Specifically, each PET image sample is input into the intermediate lesion segmentation model to obtain a coarse-grained lesion segmentation map corresponding to each PET image sample. The coarse-grained lesion segmentation map is the prediction result of the model trained on downsampled samples for the original-size image. Due to the influence of downsampling, the resolution of the coarse-grained lesion segmentation map is low, and the segmentation result may not be refined enough. Therefore, the coarse-grained lesion segmentation map is further upsampled to obtain an upsampled sample corresponding to each PET image sample. The purpose of upsampling is to restore the resolution of the coarse-grained lesion segmentation map to the original image size. Common upsampling methods include deconvolution and interpolation. The upsampled samples will be used for subsequent model retraining.

[0072] Subsequently, the intermediate lesion segmentation model is retrained based on the upsampled samples corresponding to each PET image sample. The purpose of retraining is to further improve the segmentation accuracy of the model, enabling it to better adapt to images of the original size. By comparing the upsampled samples with the actual lesion segmentation, the loss function is calculated and the model parameters are updated, allowing the model to obtain more accurate segmentation results even on the upsampled images. When the model loss value of the intermediate lesion segmentation model is less than a second preset loss threshold, the preset lesion segmentation model is obtained.

[0073] The embodiments of this application utilize the nnU-Net framework to automatically determine the model structure, reducing the workload of manual model design, and can adapt to different types and scales of PET image data, exhibiting strong versatility and adaptability. By employing a hierarchical training method of first training on downsampled samples and then retraining on upsampled samples, training efficiency is improved, and the segmentation accuracy of the model is gradually increased, resulting in more accurate lesion segmentation results.

[0074] Optionally, in this embodiment, step 101, "performing lesion segmentation on PET images under each tracer to obtain lesion segmentation results corresponding to each PET image," includes: for each PET image under each tracer, retrieving a preset lesion segmentation model and a preset sliding window mechanism matching the tracer; using the preset sliding window mechanism, acquiring sub-image regions of the PET image each time, and inputting the sub-image regions into the matching preset lesion segmentation model to obtain lesion segmentation results corresponding to the sub-image regions; performing reliability calculations on the lesion segmentation results corresponding to each sub-image region of the PET image, and obtaining the lesion segmentation results corresponding to the PET image based on the reliability calculation results.

[0075] In this embodiment, when processing a new PET image, a pre-defined lesion segmentation model matching the type of tracer used in the image is retrieved from a pre-defined model library. This matching mechanism ensures that the model can better adapt to the characteristics of the current image, thereby improving the segmentation accuracy. A sliding window mechanism is used to segment large PET images into multiple smaller sub-image regions. This is because directly segmenting the entire large image may face problems such as insufficient computational resources and difficulty in model processing. Using a sliding window, local regions of the image can be processed one by one, reducing computational complexity. The size and stride of the sliding window are two key parameters. The window size can be determined based on the typical size of the lesion and the image resolution. If the window is too small, it may not contain complete lesion information; if the window is too large, it will increase the computational load and may lead to a decrease in segmentation accuracy. The stride determines the interval at which the window slides across the image; a stride that is too small will result in excessive computation, while a stride that is too large may miss some image regions.

[0076] Following a pre-defined sliding window mechanism, the window slides across the PET image at certain step sizes. After each slide, the image region within the window is extracted as a sub-image region. At image edges, the sliding window may extend beyond the image area; in this case, padding is used to fill the image edges with appropriate pixel values ​​(such as 0 or the pixel values ​​of the image edge) so that the window completely covers the image area. Each sub-image region is input into a matching pre-defined lesion segmentation model. The model processes the sub-image region and outputs the lesion segmentation result for that region, typically a binary image where lesion areas are marked as 1 and non-lesion areas as 0. The segmentation results for each sub-image region are stored for subsequent integration processing.

[0077] After obtaining the lesion segmentation results corresponding to each sub-image region, reliability calculations can be performed on these results. Reliability calculation aims to evaluate the accuracy and confidence of each sub-image region segmentation result. For example, a pre-set lesion segmentation model can simultaneously output a confidence score. The confidence score reflects the model's grasp of the segmentation of that sub-image region; a higher score indicates higher reliability. Based on the reliability calculation results, a reliability weight is assigned to the segmentation result of each sub-image region. Segmentation results with high reliability will have a greater influence in subsequent integration. Finally, the segmentation results of each sub-image region can be integrated based on the reliability weight calculation results. Weighted averaging, voting, and other methods can be used. For example, for each pixel location, considering the segmentation results of all sub-image regions containing that pixel and their reliability weights, a weighted average is used to obtain the final segmentation value for that pixel. Furthermore, post-processing can be performed on the integrated segmentation results to further improve segmentation quality. Post-processing operations can include removing small isolated regions (which are likely noise or missegmented areas) and smoothing segmentation boundaries (making the boundaries more natural).

[0078] This application embodiment constructs a matching preset lesion segmentation model for each tracer, which can better adapt to the characteristics of PET images under different tracers and improve the accuracy of segmentation; the sliding window mechanism can be adjusted according to image size and computing resources, so that the scheme can handle images of different sizes and balances computing efficiency and segmentation accuracy to a certain extent; by integrating the segmentation results of sub-image regions through reliability calculation, the occurrence of missegmentation can be reduced and the reliability of the final segmentation result can be improved.

[0079] Optionally, in this embodiment of the application, step 104 includes: receiving a target analysis request, parsing the target analysis request to obtain a state to be analyzed, retrieving the state analysis model corresponding to the state to be analyzed, inputting the first indicator parameter and the second indicator parameter corresponding to each tracer into the state analysis model, and obtaining the state analysis result corresponding to the target object. The state to be analyzed includes at least one of overall survival state, progression-free survival state, and disease recurrence rate state.

[0080] In this embodiment, the state of the target object can be analyzed based on the acquired first indicator parameter and the second indicator parameter corresponding to each tracer. Specifically, a target analysis request can be received. This request can come from clinicians, researchers, or other relevant personnel. For example, in clinical research, doctors may want to understand the overall survival of patients with a certain disease to assess the effectiveness of treatment plans; in drug development, researchers may focus on the relapse rate to determine the efficacy and safety of new drugs. Requests can be presented in various forms, such as text descriptions, specific request codes, or options entered through a user interface. For example, a user can select "Analyze Overall Survival Status" as the target analysis request on a graphical user interface (GUI).

[0081] Next, the target analysis request is parsed. The purpose of parsing is to extract key information from the received target analysis request and clarify the state to be analyzed. Parsing can use Natural Language Processing (NLP) technology to parse the text-based request and identify keywords related to the state to be analyzed. For request codes or interface options with specific formats, the state to be analyzed can be directly determined through preset rules or mapping relationships. For example, when receiving a text request "Please analyze the patient's overall survival," the NLP algorithm can identify the keyword "overall survival," thereby determining the state to be analyzed as the overall survival state.

[0082] Different states to be analyzed correspond to different state analysis models. These models are pre-built based on a large amount of clinical data and related research, and each model is optimized for a specific state analysis task. The models can be constructed using machine learning algorithms (such as logistic regression, random forest, etc.). After determining the state to be analyzed, the corresponding state analysis model is retrieved from the model library according to a pre-defined mapping relationship. The model library stores various pre-trained models, each with its unique identifier and corresponding state analysis information. By querying this information, the required state analysis model can be quickly and accurately found and retrieved.

[0083] Furthermore, the first indicator parameters and the second indicator parameters corresponding to each tracer are organized and input according to the format required by the model. After receiving the input parameters, the state analysis model performs calculations based on its internal learning algorithm and parameters. Specifically, it performs operations such as feature extraction, weight allocation, and combination on the input parameters, and finally outputs the state analysis results corresponding to the target object. For example, for the overall survival state analysis model, the output result can be a predicted overall survival time range or survival probability; for the disease recurrence rate state analysis model, the output result can be a predicted recurrence rate.

[0084] This application embodiment, by retrieving the corresponding state analysis model according to different states to be analyzed, can more accurately perform state analysis for specific problems, improving the accuracy and reliability of the analysis results; by comprehensively considering the first indicator parameter and the second indicator parameter corresponding to each tracer, it makes full use of the information from multiple sources of data, and can more comprehensively reflect the health status and disease characteristics of the target object; from receiving the requirements and parsing the requirements to retrieving the model and outputting the results, the entire process is automated, reducing manual intervention and improving analysis efficiency and consistency.

[0085] In one specific embodiment, after determining the state to be analyzed, the attribute values ​​of other attributes of the target object that match the state to be analyzed can be obtained, such as gender, age, smoking history, medical history, etc. Furthermore, the required target indicator parameters can be determined from the first indicator parameter and the second indicator parameter based on the state to be analyzed. These attribute values ​​and target indicator parameters are then used as input to the state analysis model corresponding to the state to be analyzed, ultimately obtaining the desired prediction result.

[0086] Furthermore, as Figure 1 In terms of specific implementation, this application provides an analysis system for multi-tracer images, such as... Figure 2 As shown, the system includes:

[0087] The image registration module is used to acquire PET images of the target object under each tracer and CT images matching the PET images, and to perform lesion segmentation on the PET images under each tracer to obtain lesion segmentation results for each PET image, and to perform organ segmentation on the CT images under each tracer to obtain organ segmentation results for each CT image; based on the CT images under each tracer, the module performs registration processing on all CT images and PET images through a unified CT coordinate system to obtain a primary registration result, and performs secondary registration processing on the primary registration result based on the organ segmentation results corresponding to the CT images under each tracer to obtain a final registration result;

[0088] The image analysis module is used to determine the overlapping regions of each tracer on the final registration result, and calculate a first index parameter of the target object based on the overlapping regions, lesion segmentation results of each PET image, and organ segmentation results of each CT image. The first index parameter includes a first systemic index parameter, a first organ-level index parameter, and a first lesion-level index parameter. Furthermore, for each tracer-treated PET image, based on the tracer-treated regions and lesion segmentation results of the PET image, and the organ segmentation results of the CT image matching the PET image, the module calculates a second index parameter of the target object under the tracer. The second index parameter includes a second systemic index parameter, a second organ-level index parameter, and a second lesion-level index parameter. Based on the first index parameter and the second index parameter corresponding to each tracer, the module performs a state analysis on the target object.

[0089] Optionally, the image registration module is further configured to:

[0090] Using any CT image from the CT images under each of the tracers as a reference image, and through a unified CT coordinate system, the image deviation between each target CT image in the remaining CT images and the reference image is calculated respectively. Based on the image deviation, the adjustment parameters corresponding to the target CT image are calculated. The adjustment parameters include translation parameters, scaling parameters, and rotation parameters.

[0091] For each target CT image, the target CT image and the PET image matching the target CT image are processed sequentially according to the adjustment parameters corresponding to each target CT image. The processed image is then aligned with the reference image and the PET image corresponding to the reference image. After the processed images corresponding to each target CT image have been aligned, the primary registration result is obtained.

[0092] The image registration module is also used for:

[0093] For each processed target CT image in the primary registration result, based on the organ segmentation results of the processed target CT image and the organ segmentation results of the reference image, the mapping relationship between each pixel in the processed target CT image and the corresponding pixel in the reference image is calculated. According to the mapping relationship, the processed target CT image and the corresponding processed PET image are re-registered. After all processed target CT images and the corresponding processed PET images are re-registered, the final registration result is obtained.

[0094] Optionally, the image analysis module is used for:

[0095] Based on the final registration result, a set of pixels corresponding to the lesion location in the final registered PET image for each tracer is determined, wherein the set of pixels includes the coordinates of each pixel point contained in the lesion location;

[0096] Based on the pixel set corresponding to each tracer, common pixels in each pixel set are extracted, and the development overlap region of the final registration result is determined based on the common pixels.

[0097] Optionally, the image analysis module is further configured to:

[0098] Based on the overlapping imaging region, the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to the overlapping imaging region are calculated, and the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis are used as the first systemic index parameters.

[0099] Based on the final registration result, the organ registration result of each CT image is determined, and the imaging overlap region is divided according to the organ registration result to obtain the first overlap sub-region corresponding to each organ. Based on the first overlap sub-region corresponding to each organ, the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to the first overlap sub-region are calculated. The mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to each first overlap sub-region are used as the first organ-level index parameters.

[0100] Based on the final registration result, the lesion registration result of each PET image is determined, and the overlapping region is divided according to the lesion registration result to obtain the second overlapping sub-region corresponding to each lesion. Based on the second overlapping sub-region corresponding to each lesion, the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis are calculated for the second overlapping sub-region. The mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to each second overlapping sub-region are used as the first lesion-level index parameters.

[0101] Optionally, PET images under each tracer are segmented based on a matched preset lesion segmentation model, which is constructed as follows:

[0102] Multiple PET image samples corresponding to the target tracer are obtained, each PET image sample is marked with a lesion, and each lesion is marked with a lesion label;

[0103] The nnU-Net framework is invoked to extract multidimensional features from each PET image sample. Based on the multidimensional feature extraction results, the network depth, network width, encoder structure, and decoder structure of the initial lesion segmentation model are determined. The initial lesion segmentation model is then constructed based on the network depth, network width, encoder structure, and decoder structure.

[0104] Each PET image sample is downsampled to obtain a downsampled sample for each PET image sample. The initial lesion segmentation model is trained based on the downsampled sample to obtain an intermediate lesion segmentation model.

[0105] Each PET image sample is input into the intermediate lesion segmentation model to obtain a coarse-grained lesion segmentation map corresponding to each PET image sample. The coarse-grained lesion segmentation map is then upsampled to obtain an upsampled sample corresponding to each PET image sample.

[0106] Based on the upsampled samples corresponding to each PET image sample, the intermediate lesion segmentation model is retrained to obtain the preset lesion segmentation model.

[0107] Optionally, the image registration module is further configured to:

[0108] For each tracer-based PET image, a preset lesion segmentation model and a preset sliding window mechanism matching the tracer are retrieved. Through the preset sliding window mechanism, a sub-image region of the PET image is acquired each time, and the sub-image region is input into the matching preset lesion segmentation model to obtain the lesion segmentation result corresponding to the sub-image region.

[0109] The reliability of the lesion segmentation results corresponding to each sub-image region of the PET image is calculated, and the lesion segmentation results corresponding to the PET image are obtained based on the reliability calculation results.

[0110] Optionally, the image analysis module is further configured to:

[0111] The system receives a target analysis request, parses the target analysis request to obtain the state to be analyzed, and retrieves the state analysis model corresponding to the state to be analyzed. The first indicator parameter and the second indicator parameter corresponding to each tracer are input into the state analysis model to obtain the state analysis result corresponding to the target object. The state to be analyzed includes at least one of the following: overall survival state, progression-free survival state, and disease recurrence rate state.

[0112] It should be noted that other corresponding descriptions of the functional units involved in the analysis system for multi-tracer images provided in this application embodiment can be found in the following references. Figure 1 The corresponding descriptions in the method will not be repeated here.

[0113] This application also provides a computer device, which may specifically be a personal computer, a server, a network device, etc. Figure 3 As shown, the computer device includes a bus, a processor, memory, and a communication interface, and may also include an input / output interface and a display device. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores location information. The network interface allows communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0114] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0115] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0116] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0117] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0118] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0119] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0120] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for analyzing multi-tracer images, characterized in that, include: Acquire PET images of the target object under each tracer and CT images matching the PET images, and perform lesion segmentation on the PET images under each tracer to obtain the lesion segmentation results corresponding to each PET image, and perform organ segmentation on the CT images under each tracer to obtain the organ segmentation results corresponding to each CT image. Based on the CT images under each tracer, the entire CT image and PET image are registered using a unified CT coordinate system to obtain the primary registration result. Then, based on the organ segmentation results corresponding to the CT images under each tracer, the primary registration result is further registered to obtain the final registration result. Based on the final registration result, the overlapping regions of each tracer on the final registration result are determined, and a first index parameter of the target object is calculated according to the overlapping regions, the lesion segmentation results of each PET image, and the organ segmentation results of each CT image. The first index parameter includes a first systemic index parameter, a first organ-level index parameter, and a first lesion-level index parameter. Furthermore, for each PET image under each tracer, a second index parameter of the target object under the tracer is calculated according to the tracer-developed region of the PET image, the lesion segmentation result, and the organ segmentation result of the CT image matching the PET image. The second index parameter includes a second systemic index parameter, a second organ-level index parameter, and a second lesion-level index parameter. Based on the first indicator parameter and the second indicator parameter corresponding to each tracer, the state of the target object is analyzed.

2. The method according to claim 1, characterized in that, The CT images based on each tracer are registered using a unified CT coordinate system, yielding a preliminary registration result, including: Using any CT image from the CT images under each of the tracers as a reference image, and through a unified CT coordinate system, the image deviation between each target CT image in the remaining CT images and the reference image is calculated respectively. Based on the image deviation, the adjustment parameters corresponding to the target CT image are calculated. The adjustment parameters include translation parameters, scaling parameters, and rotation parameters. For each target CT image, the target CT image and the PET image matching the target CT image are processed sequentially according to the adjustment parameters corresponding to each target CT image. The processed image is then aligned with the reference image and the PET image corresponding to the reference image. After the processed images corresponding to each target CT image have been aligned, the primary registration result is obtained. The organ segmentation results corresponding to the CT images under each tracer are used to perform secondary registration processing on the primary registration results to obtain the final registration results, including: For each processed target CT image in the primary registration result, based on the organ segmentation results of the processed target CT image and the organ segmentation results of the reference image, the mapping relationship between each pixel in the processed target CT image and the corresponding pixel in the reference image is calculated. According to the mapping relationship, the processed target CT image and the corresponding processed PET image are re-registered. After all processed target CT images and the corresponding processed PET images are re-registered, the final registration result is obtained.

3. The method according to claim 1, characterized in that, The step of determining the overlapping development regions of each tracer on the final registration result includes: Based on the final registration result, a set of pixels corresponding to the lesion location in the final registered PET image for each tracer is determined, wherein the set of pixels includes the coordinates of each pixel point contained in the lesion location; Based on the pixel set corresponding to each tracer, common pixels in each pixel set are extracted, and the development overlap region of the final registration result is determined based on the common pixels.

4. The method according to claim 3, characterized in that, The calculation of the first index parameter of the target object based on the overlapping imaging region, the lesion segmentation results of each PET image, and the organ segmentation results of each CT image includes: Based on the overlapping imaging region, the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to the overlapping imaging region are calculated, and the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis are used as the first systemic index parameters. Based on the final registration result, the organ registration result of each CT image is determined, and the imaging overlap region is divided according to the organ registration result to obtain the first overlap sub-region corresponding to each organ. Based on the first overlap sub-region corresponding to each organ, the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to the first overlap sub-region are calculated. The mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to each first overlap sub-region are used as the first organ-level index parameters. Based on the final registration result, the lesion registration result of each PET image is determined, and the overlapping region is divided according to the lesion registration result to obtain the second overlapping sub-region corresponding to each lesion. Based on the second overlapping sub-region corresponding to each lesion, the mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis are calculated for the second overlapping sub-region. The mean SUV, maximum SUV, FDG-metabolic tumor volume, and total lesion glycolysis corresponding to each second overlapping sub-region are used as the first lesion-level index parameters.

5. The method according to claim 1, characterized in that, For each tracer, PET images were segmented based on a pre-defined lesion segmentation model. This pre-defined lesion segmentation model for each tracer was constructed as follows: Multiple PET image samples corresponding to the target tracer are obtained, each PET image sample is marked with a lesion, and each lesion is marked with a lesion label; The nnU-Net framework is invoked to extract multidimensional features from each PET image sample. Based on the multidimensional feature extraction results, the network depth, network width, encoder structure, and decoder structure of the initial lesion segmentation model are determined. The initial lesion segmentation model is then constructed based on the network depth, network width, encoder structure, and decoder structure. Each PET image sample is downsampled to obtain a downsampled sample for each PET image sample. The initial lesion segmentation model is trained based on the downsampled sample to obtain an intermediate lesion segmentation model. Each PET image sample is input into the intermediate lesion segmentation model to obtain a coarse-grained lesion segmentation map corresponding to each PET image sample. The coarse-grained lesion segmentation map is then upsampled to obtain an upsampled sample corresponding to each PET image sample. Based on the upsampled samples corresponding to each PET image sample, the intermediate lesion segmentation model is retrained to obtain the preset lesion segmentation model.

6. The method according to claim 5, characterized in that, The step of segmenting lesions in PET images under each tracer to obtain lesion segmentation results for each PET image includes: For each tracer-based PET image, a preset lesion segmentation model and a preset sliding window mechanism matching the tracer are retrieved. Through the preset sliding window mechanism, a sub-image region of the PET image is acquired each time, and the sub-image region is input into the matching preset lesion segmentation model to obtain the lesion segmentation result corresponding to the sub-image region. The reliability of the lesion segmentation results corresponding to each sub-image region of the PET image is calculated, and the lesion segmentation results corresponding to the PET image are obtained based on the reliability calculation results.

7. The method according to any one of claims 1 to 6, characterized in that, The step of performing state analysis on the target object based on the first index parameter and the second index parameter corresponding to each tracer includes: The system receives a target analysis request, parses the target analysis request to obtain the state to be analyzed, and retrieves the state analysis model corresponding to the state to be analyzed. The first indicator parameter and the second indicator parameter corresponding to each tracer are input into the state analysis model to obtain the state analysis result corresponding to the target object. The state to be analyzed includes at least one of the following: overall survival state, progression-free survival state, and disease recurrence rate state.

8. An analysis system for multi-tracer images, characterized in that, include: The image registration module is used to acquire PET images of the target object under each tracer and CT images matching the PET images, and to perform lesion segmentation on the PET images under each tracer to obtain lesion segmentation results for each PET image, and to perform organ segmentation on the CT images under each tracer to obtain organ segmentation results for each CT image; based on the CT images under each tracer, the module performs registration processing on all CT images and PET images through a unified CT coordinate system to obtain a primary registration result, and performs secondary registration processing on the primary registration result based on the organ segmentation results corresponding to the CT images under each tracer to obtain a final registration result; The image analysis module is used to determine the overlapping regions of each tracer on the final registration result, and calculate a first index parameter of the target object based on the overlapping regions, lesion segmentation results of each PET image, and organ segmentation results of each CT image. The first index parameter includes a first systemic index parameter, a first organ-level index parameter, and a first lesion-level index parameter. Furthermore, for each tracer-treated PET image, based on the tracer-treated regions and lesion segmentation results of the PET image, and the organ segmentation results of the CT image matching the PET image, the module calculates a second index parameter of the target object under the tracer. The second index parameter includes a second systemic index parameter, a second organ-level index parameter, and a second lesion-level index parameter. Based on the first index parameter and the second index parameter corresponding to each tracer, the module performs a state analysis on the target object.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images