Organ virtual display method and system driven by three-dimensional medical image
By generating virtual display models based on 3D medical images and using classification and risk labeling models for virtual image comparison and correction, the problem that virtual models cannot reflect anatomical variations and pathological features is solved, achieving high-precision virtual display and interactive surgical training.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-02-09
- Publication Date
- 2026-03-13
AI Technical Summary
In existing technologies, virtual models are difficult to reflect the anatomical variations and tissue texture differences of patients, and cannot truly reflect key pathological features such as tumors and lesions. Moreover, the constructed virtual models are static geometric objects, which are difficult to provide preoperative planning support.
By generating a virtual display model based on 3D medical images, using classification and risk labeling models to extract suspicious areas, generating high-risk samples and combining them with 3D medical images, virtual image comparison and correction are performed, and corrected images are generated in conjunction with user operations, achieving accurate matching and real-time updates.
It improves the accuracy and realism of virtual display, enhances the relevance and effectiveness of surgical training, and ensures the precision and efficiency of virtual display results.
Smart Images
Figure CN121661219A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of medical imaging technology, specifically relating to a method and system for virtual display of organs driven by three-dimensional medical images. Background Technology
[0002] With the development of modern medical imaging technology, three-dimensional imaging technology, especially computed tomography, magnetic resonance imaging and positron emission tomography, detects and collects image data of the internal structure of the human body. In existing technologies, virtualization models are used in combination with virtual reality and augmented reality technologies to construct visualized three-dimensional virtual objects of organs based on the detected image data, so as to provide medical personnel with medical education, surgical simulation and doctor-patient communication.
[0003] In existing technologies, virtual models mostly use standardized general organ models, and then map patient image data to their approximate anatomical locations. This easily overlooks the patient's anatomical variations and differences in tissue texture, and it is difficult to truly reflect key pathological features such as tumors, lesions, and inflammation. Furthermore, the constructed virtual models are mostly static, smooth-looking, and structurally regular geometric shapes, which are difficult to reflect changes in tissue density, the adjacent relationships of surrounding blood vessels and tissues, etc., making the models only serve as basic anatomical references and difficult to provide preoperative planning support for clinical cases.
[0004] Based on the above-mentioned technical problems, this application proposes a method and system for virtual display of organs driven by three-dimensional medical images. Summary of the Invention
[0005] The purpose of this invention is to provide a method and system for virtual organ display driven by three-dimensional medical images. This system can construct a virtual organ display model based on three-dimensional medical images, achieving precise matching of virtual organ display and output for patients. This allows for highly realistic interaction methods based on virtual display during medical equipment development or medical student training, avoiding deviations in the matching process. Furthermore, it can summarize user interaction data and quickly generate corresponding matching values based on the summarized results, enabling real-time, accurate, and efficient output of virtual display results.
[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0007] A method for virtual organ display driven by three-dimensional medical images, comprising the following steps:
[0008] Based on the three-dimensional medical image of the organ to be detected and multiple training samples, a virtual display image is generated and output;
[0009] When a user action is received on the virtual display image, the following steps are performed: determine the operation deviation and generate an evaluation result based on the user action; update the virtual display image based on the parameters indicated by the evaluation result.
[0010] The process of generating and outputting a virtual display image based on a 3D medical image of the organ to be detected and multiple training samples includes: processing multiple training samples based on a classification model to obtain a risk level classification result that divides the multiple training samples into at least high-risk samples and low-risk samples; generating multiple virtual image output results by combining the 3D medical image with the high-risk samples as input; comparing the multiple virtual image output results to determine the comparison deviation between the multiple virtual image output results; and correcting one of the multiple virtual image output results according to the comparison deviation to generate a corrected image as the virtual display image.
[0011] Preferably, the processing of multiple training samples based on the classification model includes:
[0012] For each training sample, based on the risk labeling model, multiple suspicious sites are extracted from the organ structure corresponding to the training sample, and the location of each suspicious site is defined as a suspicious region. Each suspicious region is labeled as a unit to be evaluated, and a center point is determined for each unit to be evaluated as a standard evaluation point. Multiple additional evaluation points are created at the location of the standard evaluation point, and the set of points consisting of the standard evaluation point and multiple additional evaluation points is defined as the evaluation point set. Evaluation is performed based on the evaluation point set to output multiple evaluation scores. The multiple evaluation scores are compared with a preset threshold to generate high-risk or low-risk labels, and the high-risk or low-risk labels are used as classification criteria and input into the risk processing model to determine the risk level of the training sample through the risk processing model.
[0013] Preferably, the comparison deviation includes differences in image similarity and differences in display latency.
[0014] Preferably, correcting one of the multiple virtual image outputs based on the comparison deviation to generate a corrected image as a virtual display image includes:
[0015] Based on differences in image similarity and display latency, one of the multiple virtual image outputs is adjusted to generate a corrected image; and the multiple virtual image outputs are compared with a 3D medical image to determine the deviation value corresponding to each virtual image output; from the multiple virtual image outputs, the virtual image output with the smallest deviation value is selected as the reference image; based on the reference image, the display brightness and display duration are adjusted to generate the corrected image.
[0016] A three-dimensional medical image-driven organ virtual display system, used to implement the above-mentioned three-dimensional medical image-driven organ virtual display method, includes the following modules:
[0017] The virtual display generation module is used to generate and output virtual display images based on the three-dimensional medical images of the organ to be detected and multiple training samples;
[0018] The user operation acquisition module is used to acquire user operations on the virtual display image.
[0019] The operation evaluation module is used to determine the operation deviation and generate evaluation results based on the user operation obtained by the response user operation acquisition module.
[0020] The image update module is used to update the virtual display image based on the parameters indicated by the evaluation results.
[0021] Preferably, the process of generating and outputting a virtual display image based on a three-dimensional medical image of the organ to be detected and multiple training samples includes:
[0022] Based on a classification model, multiple training samples are processed to obtain risk level classification results that divide the multiple training samples into at least high-risk level samples and low-risk level samples; multiple virtual image output results are generated by combining three-dimensional medical images with high-risk level samples as input; the multiple virtual image output results are compared with each other to determine the comparison deviation between the multiple virtual image output results; based on the comparison deviation, one of the multiple virtual image output results is corrected to generate a corrected image as a virtual display image.
[0023] Preferably, the processing of multiple training samples based on the classification model includes:
[0024] For each training sample, based on the risk labeling model, multiple suspicious parts are extracted from the organ structure corresponding to the training sample, and the location of each suspicious part is defined as a suspicious region; each suspicious region is labeled as a unit to be evaluated, and a center point is determined for each unit to be evaluated as a standard evaluation point; multiple additional evaluation points are created at the location of the standard evaluation point, and the set of points consisting of the standard evaluation point and multiple additional evaluation points is defined as the evaluation point set.
[0025] Preferably, the process of processing multiple training samples based on the classification model further includes: evaluating based on the set of evaluation points to output multiple evaluation scores; comparing the multiple evaluation scores with a preset threshold to generate a high-risk label or a low-risk label; and inputting the high-risk label or low-risk label as a classification basis into the risk processing model to determine the risk level of the training samples through the risk processing model.
[0026] Preferably, the comparison deviation includes differences in image similarity and differences in display latency;
[0027] Furthermore, based on the comparison deviation, one of the multiple virtual image output results is corrected to generate a corrected image as a virtual display image, including: adjusting one of the multiple virtual image output results based on image similarity differences and display latency differences.
[0028] Preferably, the process of correcting one of the multiple virtual image outputs based on the comparison deviation to generate a corrected image as a virtual display image further includes:
[0029] Multiple virtual image outputs are compared with a 3D medical image to determine the deviation value for each virtual image output. The virtual image output with the smallest deviation value is selected from the multiple virtual image outputs as the reference image. Based on the reference image, the display brightness and display duration are adjusted to generate the corrected image.
[0030] Beneficial effects
[0031] 1. This invention combines a three-dimensional medical image of the organ to be detected with a high-risk sample as input to generate multiple virtual image outputs and compares them to determine the comparison deviation. Then, it corrects the comparison deviation to generate a virtual display image, thereby reducing the error introduced in the generation process and ensuring the accuracy of the generated virtual display image, overcoming the defect of poor virtual display effect.
[0032] 2. This invention uses a risk labeling model to extract suspicious regions from training samples and generates risk labels based on the set of evaluation points. The risk processing model determines the risk level of the training samples based on the risk labels, realizing a progressive risk level classification from suspicious region location to refined assessment, improving the accuracy and reliability of risk level classification, and providing a foundation for the subsequent generation of virtual display images.
[0033] 3. This invention acquires user operations on virtual display images, determines operational deviations and generates evaluation results, and updates virtual display images based on the evaluation results, constructing a feedback process of user operations, evaluation, and image updates. This enables virtual display images to be updated according to user operations, enhancing the realism of the interaction while improving the targeting and effectiveness of surgical training. Attached Figure Description
[0034] Figure 1 This is a flowchart of the method provided by the invention;
[0035] Figure 2 This is a system module diagram provided by the present invention. Detailed Implementation
[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the scope of protection of the invention.
[0037] Example 1
[0038] Please refer to Figure 1 This embodiment provides a method for virtual display of organs driven by three-dimensional medical images, including the following steps:
[0039] To acquire three-dimensional medical images of the organs to be examined, multi-dimensional data including cross-sections, coronal planes, and sagittal planes are collected by performing computed tomography or magnetic resonance imaging scans on the organs to be examined in the patient's body, generating three-dimensional medical images and extracting structural features.
[0040] Specifically, by constructing a standardized organ morphology benchmark to accurately extract the structural features of the organ to be detected, the following steps are taken: extracting a set of multiple image samples from a basic image database containing multiple three-dimensional medical images to jointly constitute a statistically significant population reference for the organ to be detected; extracting the structural parameters of each image sample in the basic image database and defining the structural parameters as basic parameters, wherein the structural parameters include the organ's volume, average density, surface area, and offset position relative to the standard anatomical coordinate system.
[0041] The basic parameters are normalized to eliminate the influence of inconsistent dimensions and numerical ranges caused by imaging equipment, scanning parameters or individual differences in different image samples. Based on the normalized parameters, all image samples are registered by alignment and the spatial position differences of corresponding voxels or anatomical landmarks at multiple preset stages of different phases of heartbeat or different positions of respiratory movement are calculated and quantified into similarity scores.
[0042] Specifically, the functional relationship for calculating the similarity score is:
[0043] ;
[0044] ;
[0045] In the formula, The similarity score is the reciprocal of the average spatial variance of all corresponding landmarks within the region. The higher the score, the more stable the morphology of the region is and the better the consistency among different samples. This indicates that within a certain region of multiple registered image samples... A set of spatial location vectors corresponding to anatomical landmarks; This represents the position vector, which means the first position vector. In the nth image sample The three-dimensional spatial coordinates of an anatomical landmark; This represents the average position vector, which means the first... anatomical landmarks in all Average spatial location in each sample; This indicates the number of samples, which means the total number of image samples used for calculation; This indicates the number of landmarks, which represents the total number of anatomical landmarks defined within the area.
[0046] Then, the region with the highest similarity score is selected as the candidate region. The candidate region represents the part with the most stable and consistent morphology under different samples and states. The anatomical reference region determined based on the candidate region is used as the standard region for accurately setting the evaluation points. Within the standard region, multiple evaluation points are set to accurately delineate the key anatomical boundaries of the organ based on prior anatomical knowledge. According to the distribution of multiple evaluation points, the organ contour is generated by spline interpolation or mesh generation geometry construction technology. The organ contour is marked as the structural feature that characterizes the key morphology of the organ to be tested and recorded in the computing processing platform.
[0047] Furthermore, multiple training samples were collected from the medical imaging database, and a two-stage assessment process was adopted to classify the risk level. The first-stage processing rules were used to quickly locate suspicious areas, and the second-stage processing rules were used to comprehensively judge the location results, thereby improving the accuracy and efficiency of classification.
[0048] Specifically, the two-stage evaluation process includes: for each training sample, according to the preset image segmentation processing rules, analyzing the organ structure corresponding to the training sample voxel by voxel to identify and delineate multiple suspicious parts with specific visual features.
[0049] Among them, specific visual features include: grayscale values, textures or gradient changes exceeding the normal range, which can be specifically manifested in the visual features of mucosal tears, abnormal proliferation or linear lesions in medicine;
[0050] Then, the location of each suspicious part is defined as a suspicious area, and each suspicious area is marked as a unit to be evaluated. The center point of each unit to be evaluated is determined and defined as a standard evaluation point. Around the location of the standard evaluation point, multiple additional evaluation points are created according to the preset geometric distribution such as cube or sphere shape. The set of points composed of the standard evaluation point and multiple additional evaluation points is defined as the evaluation point set, in order to capture the local environmental characteristics of the suspicious area, avoid the randomness brought by single-point evaluation, and thus perform robust feature evaluation of the unit.
[0051] Based on the set of evaluation points, a series of image statistics are calculated, such as the mean, variance, gradient magnitude, and texture complexity index of voxel gray values within the area covered by the set of evaluation points, and these image statistics are combined into multiple evaluation scores.
[0052] Specifically, the functional relationship for calculating the test score is as follows:
[0053] ;
[0054] In the formula, The score represents the weighted sum of multiple image statistics and is used to quantify the risk level of the unit being evaluated. This represents the weighting coefficient, which is a preset value assigned to the first... The weights of each image statistic are used to reflect the importance of that statistic in risk assessment;
[0055] In the formula, This represents a set of evaluation points extracted from the set of evaluation points of the unit to be evaluated. Image statistics; This represents the image statistics, which means the first... Image features such as quantized grayscale mean or texture complexity; This indicates the number of statistics, the total number of image statistics used to calculate the score.
[0056] Multiple test scores are compared with preset thresholds determined based on receiver operating characteristic curve analysis. High-risk or low-risk labels are generated for each unit to be evaluated. Then, high-risk or low-risk labels are used as classification criteria. Through preset comprehensive evaluation rules, multiple labels from the same training sample are weighted statistically or logically combined to determine the risk level of the training sample and obtain the risk level classification result of dividing multiple training samples into at least high-risk samples and low-risk samples.
[0057] Furthermore, the three-dimensional medical images of the organs to be detected with extracted structural features are combined with the selected high-risk samples to generate multiple virtual image outputs. Then, the multiple virtual image outputs are compared to determine the comparison deviation between them, thereby evaluating the consistency and stability of each virtual image output.
[0058] Among them, high-risk samples can be selected to generate clinically challenging virtual scenes by superimposing or fusing the pathological or structural variation features contained in the high-risk samples onto the three-dimensional medical image of the organ to be tested.
[0059] Specifically, it can simulate tumors at different growth stages or different degrees of vascular stenosis, thereby stress testing the performance capabilities of virtual displays.
[0060] The process of generating multiple virtual image output results uses computer graphics techniques such as skeleton skinning, texture mapping, and deformation interpolation to generate one or more virtual image output results for each high-risk sample.
[0061] Specifically, the comparison deviation includes image similarity difference and display latency difference. The image similarity difference is calculated using quantifiable image quality evaluation indicators such as structural similarity index or peak signal-to-noise ratio. The display latency difference records and compares the computation and rendering time required to generate each virtual image, so as to measure the efficiency consistency of image generation in application scenarios that require real-time interaction.
[0062] Furthermore, after determining the comparison deviation between multiple virtual image outputs, one of the multiple virtual image outputs is corrected based on the comparison deviation to generate a corrected image;
[0063] One type of correction method is to adjust one of the output results of multiple virtual images based on the differences in image similarity and display latency. Specifically, if the difference in image similarity of a certain virtual image output result is less than a preset similarity threshold, but its difference in display latency is greater than a preset latency threshold, the parameters of its rendering process can be adjusted by reducing the rendering resolution or simplifying the lighting calculation to shorten the latency, thereby achieving a balance between fidelity and smoothness.
[0064] The second correction method involves comparing multiple virtual image outputs with the original 3D medical image to determine the root mean square error of the voxel intensity value corresponding to each virtual image output as its deviation value. From the multiple virtual image outputs, the virtual image output with the smallest deviation value is selected as the rendering result that is closest to the real anatomical structure, and this virtual image output is defined as the reference image. Furthermore, based on the reference image, perceptual parameters such as display brightness, contrast, stereoscopic intensity, or display duration are adjusted to optimize the user's visual experience or meet specific task requirements.
[0065] Furthermore, to ensure the anatomical rationality of the correction process, the second correction method also uses low-risk samples as a basis to assist in the correction of the virtual image. Specifically, it uses the normal anatomical structure boundary defined by the low-risk samples as a constraint on the correction operation to ensure that the correction process does not generate morphological structures that exceed the boundary and do not conform to physiological common sense. Based on the corrected image, a virtual display image of the organ to be detected is output.
[0066] Furthermore, the generated virtual display images are presented to users such as medical experts or trainees, and user operations on the virtual display images are obtained. In some scenarios, multiple virtual image output results are presented to the user, and the user's selection of multiple virtual image output results is received to determine the user's evaluation result. The selection result can be used as a reference for adjusting the parameters or priorities used in the subsequent virtual image generation process.
[0067] Furthermore, during the interaction, the user's actions, such as the movement trajectory of the virtual surgical instruments, are compared in real time with the anatomical structures, such as the boundaries of key blood vessels, presented in the virtual display image to determine operational deviations. Then, based on the operational deviations, evaluation results are generated to indicate the specific parameters that need to be adjusted. If the evaluation results show that the user has a continuous deviation in depth perception, it may be necessary to adjust the stereoscopic intensity of the virtual display or the lighting rendering scheme. Based on the parameters indicated by the evaluation results, the virtual display image is updated, thus forming a closed-loop, continuously optimized interactive display process.
[0068] Example 2
[0069] Please refer to Figure 2 This embodiment provides a three-dimensional medical image-driven organ virtual display system for implementing the above-described three-dimensional medical image-driven organ virtual display method, including the following modules:
[0070] The virtual display generation module is used to generate and output virtual display images based on the three-dimensional medical images of the organ to be detected and multiple training samples. It performs a sample screening and image generation process based on a classification model: based on the classification model, it processes multiple training samples of three-dimensional medical images with annotation information of pathological diagnosis results collected in advance to classify multiple training samples into high-risk samples or low-risk samples.
[0071] Specifically, the classification model can consist of two sub-models: For each training sample, based on the trained convolutional neural network, i.e., the risk labeling model, multiple suspicious sites such as potential tumors, nodules, or abnormal tissues in the organ structure corresponding to the training sample are extracted, and the location of each suspicious site is defined as a suspicious region; each suspicious region is labeled as a unit to be evaluated, and its geometric center point is determined as a standard evaluation point for each unit to be evaluated; around the location of the standard evaluation point, multiple additional evaluation points are created according to preset distribution rules such as concentric spheres or grids in three-dimensional space, and the set of points consisting of the standard evaluation point and multiple additional evaluation points is defined as the evaluation point set, so as to more comprehensively evaluate the risk characteristics of the region;
[0072] Based on the set of evaluation points, image statistics such as image texture, grayscale, and gradient of the region are calculated for evaluation, and multiple evaluation scores are output. These evaluation scores are compared with one or more preset thresholds to generate high-risk or low-risk labels. These labels are used as classification criteria and input into the risk processing model of support vector machine (SVM) or decision tree. The risk processing model finally determines the risk level of the training sample.
[0073] After obtaining the risk level classification results, the virtual display generation module selects the data of the high-risk samples, inputs the three-dimensional medical image of the organ to be detected and the high-risk samples, and transfers or fuses the lesion features of the high-risk samples into the image to be detected through deep learning models such as generative adversarial networks, generating multiple virtual image output results.
[0074] The virtual display generation module compares multiple virtual image outputs to determine the comparison deviation, including differences in image similarity and display latency. Specifically, it calculates the structural similarity index or peak signal-to-noise ratio between different outputs to obtain image similarity differences, and generates the time difference required for each output as the display latency difference. Based on these differences, one of the multiple virtual image outputs is adjusted to generate a corrected image that serves as the final virtual display image. The image with the fastest generation speed and highest similarity to other images can be selected as the base and fine-tuned to ensure the quality and consistency of the final output virtual display image.
[0075] Specifically, the correction steps include: comparing multiple virtual image output results with the original three-dimensional medical image to be detected to determine the sum of pixel-level differences or mean square error and other deviation values corresponding to each virtual image output result; selecting the virtual image output result with the smallest deviation value from the multiple virtual image output results as the reference image; adjusting its display brightness and display duration based on the reference image and in conjunction with a preset display strategy to generate a corrected image that is more suitable for observation; and outputting it as a virtual display image to the display device.
[0076] The user operation acquisition module is used to acquire user operations, connect and manage various input devices, including force feedback devices such as mice, keyboards, touch screens, virtual scalpels, spatial locators or eye trackers, etc.
[0077] Specifically, user operations include: clicking and selecting on the virtual display image to mark areas of interest; changing the viewing angle by dragging and rotating; or simulating surgical instruments to perform interactive actions such as cutting and separating, and converting the captured physical operations into digital signal output in real time.
[0078] The operation evaluation module receives user operations, determines operation deviations based on user operations, and generates evaluation results. If the current task requires the user to locate a tiny lesion from a high-risk sample implanted by the virtual display generation module in a virtual display image, the user-marked position is compared with the actual position of the lesion, and the calculated spatial distance is the operation deviation. If the task is to simulate a surgical resection path, the user's actual operation trajectory is compared with the preset optimal or safe path, and the calculated path deviation and proximity to surrounding key tissues can be used as operation deviations. Based on the calculated operation deviations, quantitative scores or specific levels of evaluation results are generated.
[0079] The image update module dynamically adjusts the currently displayed virtual image based on the generated evaluation results and the parameters indicated by the evaluation results. If the evaluation results indicate that the user has failed to find the target lesion, the parameters may instruct the correct location to be marked on the image by highlighting or flashing. If the evaluation results indicate that the user's surgical operation path has deviated, an ideal path can be superimposed on the image as a guide. Alternatively, the module may instruct the user to adjust the image's transparency and rendering mode to expose the blood vessels or nerve structures below the user's operation path, thereby providing a more intuitive risk warning.
[0080] By constructing an intelligent and personalized medical virtual interactive environment, it is possible to generate virtual display images that incorporate high-risk characteristics and have clinical reference value. It can also accurately evaluate and provide real-time feedback on user operations, enabling surgical simulation and assisted diagnosis of difficult cases.
[0081] Finally, it should be noted that the above examples are merely some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments and many variations are possible. All variations that can be directly derived or conceived by those skilled in the art from the disclosure of the present invention should be considered within the scope of protection of the present invention.
Claims
1. A method for virtual display of organs driven by three-dimensional medical images, characterized in that, Includes the following steps: Based on the three-dimensional medical image of the organ to be detected and multiple training samples, a virtual display image is generated and output; When a user action is received on the virtual display image, the following steps are performed: determine the operation deviation and generate an evaluation result based on the user action; update the virtual display image based on the parameters indicated by the evaluation result. The process of generating and outputting a virtual display image based on a 3D medical image of the organ to be detected and multiple training samples includes: processing multiple training samples based on a classification model to obtain a risk level classification result that divides the multiple training samples into at least high-risk samples and low-risk samples; generating multiple virtual image output results by combining the 3D medical image with the high-risk samples as input; comparing the multiple virtual image output results to determine the comparison deviation between the multiple virtual image output results; and correcting one of the multiple virtual image output results according to the comparison deviation to generate a corrected image as the virtual display image.
2. The method for virtual organ display driven by three-dimensional medical images according to claim 1, characterized in that, The processing of multiple training samples based on the classification model includes: For each training sample, based on the risk labeling model, multiple suspicious sites are extracted from the organ structure corresponding to the training sample, and the location of each suspicious site is defined as a suspicious region. Each suspicious region is labeled as a unit to be evaluated, and a center point is determined for each unit to be evaluated as a standard evaluation point. Multiple additional evaluation points are created at the location of the standard evaluation point, and the set of points consisting of the standard evaluation point and multiple additional evaluation points is defined as the evaluation point set. Evaluation is performed based on the evaluation point set to output multiple evaluation scores. The multiple evaluation scores are compared with a preset threshold to generate high-risk or low-risk labels, and the high-risk or low-risk labels are used as classification criteria and input into the risk processing model to determine the risk level of the training sample through the risk processing model.
3. The method for virtual organ display driven by three-dimensional medical images according to claim 1, characterized in that: The comparison deviation includes differences in image similarity and differences in display latency.
4. The method for virtual organ display driven by three-dimensional medical images according to claim 1, characterized in that, Based on the comparison deviation, one of the multiple virtual image outputs is corrected to generate a corrected image as a virtual display image, including: Based on differences in image similarity and display latency, one of the multiple virtual image outputs is adjusted to generate a corrected image; and the multiple virtual image outputs are compared with a 3D medical image to determine the deviation value corresponding to each virtual image output; from the multiple virtual image outputs, the virtual image output with the smallest deviation value is selected as the reference image; based on the reference image, the display brightness and display duration are adjusted to generate the corrected image.
5. A three-dimensional medical image-driven organ virtual display system, used to implement the three-dimensional medical image-driven organ virtual display method as described in any one of claims 1-4, characterized in that, Includes the following modules: The virtual display generation module is used to generate and output virtual display images based on the three-dimensional medical images of the organ to be detected and multiple training samples; The user operation acquisition module is used to acquire user operations on the virtual display image. The operation evaluation module is used to determine the operation deviation and generate evaluation results based on the user operation obtained by the response user operation acquisition module. The image update module is used to update the virtual display image based on the parameters indicated by the evaluation results.
6. A three-dimensional medical image-driven virtual organ display system according to claim 5, characterized in that, The process of generating and outputting a virtual display image based on a three-dimensional medical image of the organ to be detected and multiple training samples includes: Based on a classification model, multiple training samples are processed to obtain risk level classification results that divide the multiple training samples into at least high-risk level samples and low-risk level samples; multiple virtual image output results are generated by combining three-dimensional medical images with high-risk level samples as input; the multiple virtual image output results are compared with each other to determine the comparison deviation between the multiple virtual image output results; based on the comparison deviation, one of the multiple virtual image output results is corrected to generate a corrected image as a virtual display image.
7. A three-dimensional medical image-driven virtual organ display system according to claim 6, characterized in that, The processing of multiple training samples based on the classification model includes: For each training sample, based on the risk labeling model, multiple suspicious parts are extracted from the organ structure corresponding to the training sample, and the location of each suspicious part is defined as a suspicious region; each suspicious region is labeled as a unit to be evaluated, and a center point is determined for each unit to be evaluated as a standard evaluation point; multiple additional evaluation points are created at the location of the standard evaluation point, and the set of points consisting of the standard evaluation point and multiple additional evaluation points is defined as the evaluation point set.
8. A three-dimensional medical image-driven virtual organ display system according to claim 7, characterized in that, The processing of multiple training samples based on the classification model further includes: evaluating based on the set of evaluation points to output multiple evaluation scores; comparing the multiple evaluation scores with a preset threshold to generate high-risk or low-risk labels; and inputting the high-risk or low-risk labels as classification criteria into the risk processing model to determine the risk level of the training samples through the risk processing model.
9. A three-dimensional medical image-driven virtual organ display system according to claim 6, characterized in that: The comparison deviation includes differences in image similarity and differences in display latency; Furthermore, based on the comparison deviation, one of the multiple virtual image output results is corrected to generate a corrected image as a virtual display image, including: adjusting one of the multiple virtual image output results based on image similarity differences and display latency differences.
10. A three-dimensional medical image-driven virtual organ display system according to claim 6, characterized in that, The process of correcting one of the multiple virtual image outputs based on the comparison deviation to generate a corrected image as a virtual display image also includes: Multiple virtual image outputs are compared with a 3D medical image to determine the deviation value for each virtual image output. The virtual image output with the smallest deviation value is selected from the multiple virtual image outputs as the reference image. Based on the reference image, the display brightness and display duration are adjusted to generate the corrected image.