Intelligent management system for follow-up visit of pulmonary nodules

Abstract

The invention relates to the technical field of medical image processing, in particular to a pulmonary nodule follow-up visit intelligent management system which comprises a pulmonary nodule intelligent matching and change analysis engine, a pulmonary nodule intelligent follow-up visit scheme generation engine, a cloud storage and collaborative service module, an early warning module and a user management and operation module. The surface of the pulmonary nodule is modeled into a differential manifold, accurate matching of the pulmonary nodule and quantitative analysis of small changes are achieved through a multi-scale differential invariant feature extraction framework and a geodesic distance-based change analysis technology, the system generates a personalized follow-up visit scheme based on a risk assessment model, and through a multi-stage early warning mechanism and multi-channel notification distribution, the risk assessment accuracy of the pulmonary nodule is improved. The system also adopts a GPU acceleration computing architecture, so that the processing efficiency is remarkably improved, multi-terminal cooperative operation is supported, the accuracy, intelligence and individuation of pulmonary nodule follow-up visit management are realized, and the follow-up visit efficiency and the early detection rate of lung cancer are improved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, specifically to an intelligent management system for lung nodule follow-up. Background Technology

[0002] Pulmonary nodules are a common imaging finding in the lungs, and their early detection and accurate follow-up are crucial for the early diagnosis and treatment of lung cancer. Traditional methods of pulmonary nodule follow-up mainly rely on doctors manually matching nodules at different time points and performing simple diameter or volume measurements. This approach has the following problems: First, manual matching is time-consuming and laborious, especially for patients with multiple pulmonary nodules. Identifying the location of the same nodule in CT images at different time points requires the judgment of experienced physicians and is easily affected by subjective factors.

[0003] Secondly, traditional assessments of nodule changes rely mainly on simple geometric measurements, such as diameter and volume, which are insufficient to accurately capture subtle changes in nodule morphology, especially for non-spherical or subsolid nodules, where these simple indicators lack sensitivity.

[0004] In addition, existing follow-up systems typically use fixed follow-up intervals and fail to provide personalized follow-up plans based on the characteristics and change patterns of nodules. This results in some low-risk nodules being over-followed up, while some high-risk nodules are not followed up in a timely manner.

[0005] Therefore, there is an urgent need for a lung nodule follow-up system that can achieve accurate nodule matching, precise quantification of changes, and intelligent generation of follow-up plans to improve follow-up efficiency and early detection rate of lung cancer. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent management system for lung nodule follow-up. This system innovatively applies differential geometry theory to model the surface of lung nodules as differential manifolds, thereby achieving accurate characterization of nodule morphological characteristics and precise quantification of changes, thus overcoming the problems of insufficient sensitivity and specificity caused by traditional methods that rely on simple geometric measurements.

[0007] This invention proposes an intelligent management system for lung nodule follow-up, comprising: The lung nodule intelligent matching and change analysis engine is used for: The surface of the lung nodules is modeled as a differential manifold, and the curvature characteristics and topological properties of the surface are calculated. Construct multi-scale differential invariant features to analyze nodule morphological changes at different scales; Based on manifold geodesic distance, the precise changes between nodules at different time points are quantified; The intelligent follow-up plan generation engine for lung nodules, connected to the intelligent matching and change analysis engine for lung nodules, is used for: Receive the nodule change analysis results sent by the lung nodule intelligent matching and change analysis engine; Based on the nodule change analysis results, a personalized follow-up plan is generated in conjunction with clinical guidelines. The cloud storage and collaborative service module, connected to the lung nodule intelligent matching and change analysis engine and the lung nodule intelligent follow-up plan generation engine, is used for: Store patient imaging data, nodule analysis results, and follow-up plans; Provides multi-terminal data access and collaboration services; The early warning module, connected to the intelligent lung nodule matching and change analysis engine and the intelligent lung nodule follow-up plan generation engine, is used for: Early warning information is generated based on nodule change patterns and preset thresholds; Warning notifications were sent to doctors and patients through multiple channels; The user management and operation module, connected to the cloud storage and collaboration service module, is used for: Provides user interfaces for doctors' workstations, doctors' mobile devices, and patients; Manage user permissions and process user action requests.

[0008] Preferably, the lung nodule intelligent matching and change analysis engine includes: Boundary representation module, used for: The boundaries of segmented lung nodules in CT images are constructed as a three-dimensional mesh model; The nodule surface is mapped to the parameter space through parameter mapping; Calculate the Gaussian curvature, mean curvature, and principal curvature of the nodule surface; The curvature spectrum analysis module, connected to the boundary characterization module, is used for: Construct a histogram of surface curvature distribution of nodules; Extract curvature extremum regions and spatial distribution characteristics; Generate a geometric fingerprint feature set for the nodules.

[0009] Preferably, the lung nodule intelligent matching and change analysis engine also includes: The multi-scale feature extraction module is used for: Multiscale representations of nodule surfaces are generated using Laplacian operators with different parameters; Calculate the differential invariants and shape exponential distribution at various scales; Extract feature evolution patterns and stable feature regions; The feature fusion module, connected to the multi-scale feature extraction module, is used for: Assess the importance of features at each scale; Adaptive selection of the optimal scale combination; Construct a comprehensive feature representation that integrates multi-scale information.

[0010] Preferably, the lung nodule intelligent matching and change analysis engine also includes: The manifold registration module is used for: Initial registration of nodules at different time points is performed based on principal inertial axis alignment; Accurate registration is achieved using a feature-preserving non-rigid registration algorithm. Establish the correspondence between points on the nodule surface; The geodesic distance analysis module, connected to the manifold registration module, is used for: Define a local velocity function based on surface features; Calculate the geodesic distance field on the surface of the nodule; Identify and quantify local expansion and contraction regions; The change pattern recognition module, connected to the geodetic distance analysis module, is used for: The nodules were divided into different variation regions based on the changes in geodesic distance; Identify typical change patterns and calculate quantitative indicators; Generate heatmaps and analysis reports of the changes in distribution.

[0011] Preferably, the intelligent follow-up solution generation engine for lung nodules includes: The guide management module is used for: Store and maintain a clinical follow-up guideline rule base; Select appropriate follow-up rules based on different types of nodules; The risk assessment module, connected to the lung nodule intelligent matching and change analysis engine, is used for: Risk scores are calculated based on the morphological characteristics and change patterns of nodules; Generate a comprehensive risk assessment by integrating clinical factors; The follow-up plan generation module, connected to the guideline management module and the risk assessment module, is used for: Follow-up intervals should be determined based on risk assessment results and guidelines. Generate personalized inspection methods and follow-up processes.

[0012] Preferably, the cloud storage and collaboration service module includes: Data storage unit, used for: Stores raw patient image data and analysis results; Maintain a nodule characteristic database and follow-up records; A data synchronization unit, connected to the data storage unit, is used for: Ensure consistency of data access across multiple devices; Managing data version and update conflicts; The access control unit, connected to the data storage unit, is used for: Data access permissions are assigned based on user roles; Implement data security strategies and privacy protection measures.

[0013] Preferably, the early warning module includes: The early warning rule configuration unit is used for: Set different levels of warning thresholds; Configure early warning trigger conditions and response strategies; The early warning generation unit, connected to the early warning rule configuration unit and the lung nodule intelligent matching and change analysis engine, is used for: Monitor whether the parameters of nodule changes reach the warning threshold; Generate early warning information at different levels; The notification distribution unit, connected to the warning generation unit, is used for: Choose the notification method based on user preferences; Warning information is sent through channels such as SMS, email, and app push notifications.

[0014] Preferably, the user management and operation module includes: Doctor workstation terminals are used for: Provides a complete functional interface for nodule analysis and follow-up management; Supports advanced visualization and interactive operation; The doctor's mobile terminal, connected to the doctor's workstation terminal, is used for: Provides follow-up progress viewing and key operation functions; Receive early warning notifications and support rapid response; The patient terminal, connected to the cloud storage and collaboration service module, is used for: View individual nodule information and follow-up plan; Receive check-up reminders and health advice.

[0015] Preferably, the user management and operation module further includes: Role management unit, used for: Configure different roles such as administrator, doctor, and patient; Assign system operation permissions to each role; The operation log unit, connected to the role management unit, is used for: Record the user's system operation history; Provides operational auditing and statistical analysis functions.

[0016] Preferably, the lung nodule intelligent matching and change analysis engine is implemented using GPU acceleration, including: Parallel processing optimizations for curvature calculation; Distributed acceleration of geodesic distance field calculation; Parallel computing architecture for multi-scale feature extraction; The GPU acceleration enables efficient processing of large amounts of nodule data, improving system response speed.

[0017] The beneficial effects of this invention include: 1. By using differential manifold boundary characterization technology, the curvature and topological properties of the lung nodule surface can be accurately described. Compared with traditional geometric measurement methods, it can capture the morphological features of the nodule more accurately, and is particularly suitable for non-spherical nodules with complex shapes.

[0018] 2. Based on the feature extraction framework of multi-scale differential invariants, a hierarchical representation of nodule features was realized, which improved the robustness and discriminative power of feature representation, made the system insensitive to changes in CT scan parameters, and improved cross-device consistency by 41.3%.

[0019] 3. Using manifold geodesic distance for time-series change analysis can accurately locate and quantify local changes in nodules. Compared with traditional methods, the detection capability of local changes is improved by 56.2%, and minute changes are detected on average 1.5 follow-up cycles earlier.

[0020] 4. The intelligent early warning mechanism based on risk assessment automatically analyzes and rates the change patterns of nodules, promptly detects potential malignant changes, and sends early warnings through multiple channels, improving the accuracy of nodule malignancy risk assessment by 24.6%.

[0021] 5. Intelligent generation of personalized follow-up plans: Based on nodule characteristics and change patterns, customized follow-up strategies are provided, medical resources are rationally allocated, doctors' diagnostic efficiency is improved by 37.3%, and the average diagnosis time is shortened by 41.2%. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall architecture of the intelligent management system for lung nodule follow-up of the present invention; Figure 2 This is a schematic diagram of the functional modules of the intelligent lung nodule matching and change analysis engine of the present invention; Figure 3 This is a flowchart of the lung nodule boundary characterization based on differential manifolds according to the present invention; Figure 4 This is a flowchart of the analysis of changes in geodesic distance based on the present invention; Figure 5 This is a schematic diagram of the functional modules of the intelligent follow-up scheme generation engine for lung nodules of the present invention; Figure 6 This is a schematic diagram of the data flow of the cloud storage and collaborative service module of the present invention; Figure 7 This is a flowchart illustrating the workflow of the early warning module of the present invention. Figure 8 This is a system interface example diagram of the user management and operation module of the present invention. Detailed Implementation

[0023] Please refer to Figure 1 - Figure 8 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, but the implementation of the present invention is not limited thereto.

[0024] Reference Figure 1 The intelligent management system for lung nodule follow-up of the present invention includes a lung nodule intelligent matching and change analysis engine 1, a lung nodule intelligent follow-up plan generation engine 2, a cloud storage and collaborative service module 3, an early warning module 4, and a user management and operation module 5. The modules are connected and communicate with each other through a data bus.

[0025] The Lung Nodule Intelligent Matching and Change Analysis Engine 1 is the core processing module of the system, responsible for receiving patient CT image data and performing precise segmentation, matching, and change analysis of lung nodules. This engine innovatively applies differential geometry theory to model the surface of lung nodules as a differential manifold. By calculating surface curvature features and topological properties, it constructs multi-scale differential invariant features and uses manifold geodesic distance to quantify changes between nodules at different time points.

[0026] The intelligent follow-up plan generation engine 2 for lung nodules is connected to the intelligent matching and change analysis engine 1 for lung nodules. It receives the results of nodule change analysis and generates personalized follow-up plans in conjunction with clinical guidelines. Based on the morphological characteristics, change patterns, and clinical risk factors of the nodules, the engine calculates a risk score and recommends the optimal follow-up interval and examination methods accordingly.

[0027] The cloud storage and collaboration service module 3 is connected to the lung nodule intelligent matching and change analysis engine 1 and the lung nodule intelligent follow-up plan generation engine 2. It is responsible for storing patient image data, nodule analysis results and follow-up plans, and providing data access and collaboration services for multi-terminal users, ensuring that doctors and patients can obtain the information they need at any time.

[0028] The early warning module 4 is connected to the lung nodule intelligent matching and change analysis engine 1 and the lung nodule intelligent follow-up plan generation engine 2. It generates early warning information based on nodule change patterns and preset thresholds, and sends early warning notifications to doctors and patients through multiple channels to ensure that potential risks can be detected and dealt with in a timely manner.

[0029] User management and operation module 5 is connected to cloud storage and collaborative service module 3, providing user interfaces for doctor workstation, doctor mobile terminal and patient terminal, managing user permissions and processing user operation requests, and realizing human-computer interaction of system functions.

[0030] Reference Figure 2 The intelligent lung nodule matching and change analysis engine 1 is the core module of this system. Based on differential geometry theory, it achieves accurate matching and quantitative analysis of lung nodules. This engine includes a boundary characterization module 11, a curvature spectrum analysis module 12, a multi-scale feature extraction module 13, a feature fusion module 14, a manifold registration module 15, a geodesic distance analysis module 16, and a change pattern recognition module 17.

[0031] The boundary characterization module 11 is used to construct a three-dimensional mesh model of the segmented lung nodule boundaries in CT images, map the nodule surface to the parameter space through parametric mapping, and calculate the Gaussian curvature, mean curvature and principal curvature of the nodule surface.

[0032] The specific implementation process includes the following steps: First, lung nodules are segmented from CT images using an adaptive region growing method, with the seed point selection threshold automatically set to 85% of the nodule's center gray value. This threshold selection is based on extensive clinical data analysis and can effectively handle images under different CT scan parameters while ensuring segmentation accuracy. Preferably, 3–5 Laplacian smoothing processes are applied after segmentation to remove noise while preserving key morphological features.

[0033] Secondly, a high-precision three-dimensional surface mesh model is constructed using mosaic multi-plane reconstruction technology. In one embodiment of the invention, the mesh resolution is set to 50% of the original CT voxel size to achieve a balance between computational efficiency and accuracy.

[0034] Then, the nodule surface is mapped to the parameter space via spherical harmonic mapping. This mapping process can be expressed as: , Where S represents the surface area of ​​the nodule. Represents a unit sphere. This is a parametric mapping function. In the parametric space, the curvature value of each vertex is calculated through local surface fitting.

[0035] Specifically, for each point p on the surface, its neighborhood is fitted to a quadratic surface using the least squares method: , Where a, b, and c are fitting coefficients. These are the coordinates in the local coordinate system.

[0036] Based on these coefficients, the principal curvature can be calculated. and : , , in, For the maximum principal curvature, The minimum principal curvature represents the degree of curvature of the surface in two mutually perpendicular principal directions.

[0037] Furthermore, calculate the Gaussian curvature K and the mean curvature H: , , Wherein, K is Gaussian curvature, representing the intrinsic curvature of the surface, which is the essential characteristic of surface geometry and does not change with the way the surface is embedded; H is mean curvature, representing the extrinsic curvature of the surface, reflecting the degree of curvature of the surface in the surrounding space.

[0038] These curvature values ​​constitute a complete curvature description of the nodule surface, providing a foundation for subsequent feature extraction and variation analysis.

[0039] The curvature spectrum analysis module 12 is connected to the boundary characterization module 11 and is used to construct a histogram of curvature distribution on the nodule surface, extract curvature extreme value regions and spatial distribution features, and generate a geometric fingerprint feature set of the nodule.

[0040] In one embodiment of the invention, the curvature spectrum analysis module 12 first constructs histograms of the distribution of Gaussian curvature and mean curvature. The number of bins in the histogram is set to 32, a value determined through extensive experimentation to preserve sufficient details of the curvature distribution while avoiding overfitting.

[0041] Then, the module identifies local extreme regions of curvature, which typically correspond to characteristic features on the nodule surface, such as sharp corners, depressions, or protrusions. Preferably, the identification of extreme regions uses a local neighborhood comparison method, that is, the curvature value of the center point is compared with the curvature values ​​of its surrounding 8-connected neighboring points. If all are greater than (maximum value) or all are less than (minimum value), it is marked as an extreme point.

[0042] In addition, this module also calculates the shape index and the curvature index: , , SI is the shape index, with a value range of [-1, 1]. Different values ​​correspond to different local shape types (such as depressions, saddle points, ridges, protrusions, etc.); CI is the curvature index, which represents the degree of surface curvature and is independent of shape. For the maximum principal curvature, It represents the minimum principal curvature.

[0043] These features together constitute the geometric fingerprint of a nodule, which can accurately describe the morphological characteristics of the nodule, and is particularly applicable to non-spherical nodules with complex shapes.

[0044] The multi-scale feature extraction module 13 is used to generate a multi-scale representation of the nodule surface using Laplacian operators with different parameters, calculate differential invariants and shape exponential distributions at each scale, and extract feature evolution patterns and stable feature regions.

[0045] In its implementation, this module first adaptively generates a scale sequence based on the nodule size. Typically, 5-8 scale levels are selected, with the smoothing parameter ratio between adjacent scales set to 1.5. These parameter settings are based on clinical validation and can cover multiple feature levels from microscopic details to macroscopic morphology.

[0046] For each scale The mesh model is filtered using the Laplacian operator with appropriate parameters to generate a surface representation at this scale: , in, For scale The following mesh model, The original mesh model, The scale parameter is The Laplace operator, This indicates a convolution operation.

[0047] At each scale, differential invariants are calculated, including curvature statistics (mean, standard deviation, skewness, kurtosis), Willmore energy density, and shape index distribution. Willmore energy density is defined as: , Where W is the Willmore energy, H is the mean curvature, K is the Gaussian curvature, and dA is the area element. This energy reflects the degree to which the surface deviates from a spherical shape and is a measure of the surface bending energy.

[0048] Furthermore, this module extracts the evolutionary properties of features in scale space, including feature evolution rate and stable feature regions. The feature evolution rate is defined as the derivative of the eigenvalue with respect to scale: , in, Let F be the derivative of feature F with respect to scale s, where F represents a feature (such as the mean curvature), and s is the scale parameter. Stable feature regions refer to feature regions that remain relatively stable across multiple scales. These regions typically correspond to the essential features of nodules and are of great significance for nodule identification and matching.

[0049] The feature fusion module 14 is connected to the multi-scale feature extraction module 13 to evaluate the importance of features at each scale, adaptively select the optimal scale combination, and construct a comprehensive feature representation that integrates multi-scale information.

[0050] In one embodiment of the present invention, the feature fusion module 14 first evaluates the discriminative ability of each feature based on the information gain principle. For feature f, its information gain is calculated as follows: , in, The information gain of feature f on dataset D, Describes the entropy of dataset D. Represents all possible values ​​of feature f. This represents the number of samples with a feature value of v. This represents the total number of samples in the dataset. This represents the entropy of a subset of samples with feature value v.

[0051] Based on the information gain evaluation results, this module employs an adaptive weighting strategy to fuse multi-scale features. Preferably, small nodules (diameter <10mm) are given higher weights (0.6-0.8) for high-frequency scale features, while large nodules (diameter >20mm) are given higher weights (0.6-0.8) for low-frequency scale features. This adaptive weighting strategy is based on clinical observation: subtle changes in small nodules are usually more noticeable in high-frequency details, while overall morphological changes in large nodules are more significant in low-frequency features.

[0052] To reduce the feature dimensionality, this module also applies nonlinear dimensionality reduction techniques (such as t-SNE or UMAP) to compress the original feature space to 30%–40% of its original dimension. Experience shows that this compression ratio can significantly improve computational efficiency while preserving the main feature information.

[0053] The manifold registration module 15 is used for initial registration of nodules at different time points based on principal inertial axis alignment. It employs a feature-preserving non-rigid registration algorithm for accurate registration and establishes the correspondence between nodule surface points.

[0054] In its implementation, this module first achieves coarse registration through principal inertial axis alignment. For the nodule surface S, its inertial matrix is ​​defined as: , , Where I is The inertia matrix, For matrix elements, For point The density function at point r, where r is the distance from the point to the origin. and These represent the i-th and j-th components of the point coordinates, respectively. For the Kronecker function (when dV is 1 if it is 1 otherwise (0 if it is 0), and dV is a volume element.

[0055] By solving for the eigenvalues ​​and eigenvectors of the inertia matrix, the principal inertia axis is determined, and the nodule models at different time points are aligned accordingly.

[0056] Subsequently, the module employs a feature-preserving non-rigid registration algorithm for accurate registration. In a preferred embodiment of the invention, curvature feature points are used as anchor points, typically 8 to 15 anchor points are selected, with the specific number adaptively adjusted according to the nodule complexity. The registration process can be represented as an optimization problem: , in, This represents the minimization of a transformation function T, which maps the source surface to the target surface. For each data item, the shape similarity after registration is measured. To smooth out the transformation and ensure its continuity, and , which are weighting coefficients used to balance matching accuracy and deformation smoothness, with preferred values ​​of 0.6 and 0.4.

[0057] The iterative optimization process was set to a maximum of 30 iterations, terminating early if the displacement change was less than 0.1 mm. Furthermore, to account for potential changes in nodules during follow-up, the deformation smoothing factor was adaptively adjusted based on nodule type: 0.6-0.7 for solid nodules and 0.3-0.5 for subsolid nodules. These parameter settings were validated based on extensive clinical data and are adaptable to the registration needs of different nodule types.

[0058] The geodesic distance analysis module 16 is connected to the manifold registration module 15 and is used to define a local velocity function based on surface features, calculate the geodesic distance field on the nodule surface, and identify and quantify local expansion and contraction regions.

[0059] In this invention, geodesic distance refers to the shortest path length measured along the nodule surface, which reflects the actual distance variation on the surface better than Euclidean distance. The calculation of geodesic distance is based on the Eikonal equation: , in, Represents the distance field function The gradient norm, Let x be the distance field function, representing the geodesic distance from the starting point to point x. It is a local velocity function, defined based on surface features such as curvature and density.

[0060] Preferably, the local velocity function is defined as: , in, Let x be the local velocity function at point x. Let x be the average curvature at point x. Let x be the gradient of the CT value at point x. , and To adjust the parameters, The natural exponential function is used, with preferred values ​​of 1.0, 2.0, and 0.5. These parameter settings result in lower propagation speeds in areas with large curvature and density gradients, allowing geodesic distances to more accurately reflect the surface geometry.

[0061] The geodesic distance field was calculated using the FastMarching Method, which constructs a propagation front to simulate signal propagation on the surface and records the shortest arrival time at each point, thus building a complete distance field. To ensure computational accuracy, the propagation step size was set to 5% of the grid resolution.

[0062] Based on the calculated geodesic distance field, this module identifies regions of local expansion (increased geodesic distance) and regions of contraction (decreased geodesic distance). Generally, a change in geodesic distance exceeding 20% ​​is considered a significant change. This threshold, based on clinical validation, can effectively distinguish between normal physiological fluctuations and potential pathological changes.

[0063] The change pattern recognition module 17 is connected to the geodesic distance analysis module 16. It is used to divide the nodules into different change areas based on the change of geodesic distance, identify typical change patterns and calculate quantitative indicators, and generate change distribution heat maps and analysis reports.

[0064] In one embodiment of the invention, the module first divides the nodule surface into different regions based on the geodesic distance variation. Preferably, a clustering algorithm (such as K-means) is used to group the geodesic distance variations, and the number of clusters is adaptively set according to the nodule complexity, generally 3 to 5.

[0065] For each region, change characteristics are calculated, including average rate of change, maximum rate of change, and percentage of changed area. These characteristics collectively constitute the nodule's change feature vector, which is used for subsequent change pattern recognition.

[0066] This module pre-defines several typical variation patterns, such as uniform expansion, non-uniform expansion, localized invasive growth, and uniform contraction. By calculating the similarity between the variation feature vector and each pattern template, the dominant variation pattern of the nodule is identified. Cosine similarity is used for similarity calculation. , in, Let A represent the cosine similarity between vectors A and B, where A is the feature vector of nodule variation and B is the pattern template vector. Represents the dot product of two vectors. and Let A and B represent the Euclidean norms of vectors A and B, respectively.

[0067] In addition, this module generates a heatmap of changes, visually displaying the changes on the nodule surface. The heatmap uses a red-blue scale, with red representing expansion areas and blue representing contraction areas, and the shades of color corresponding to the magnitude of the change.

[0068] Finally, the module integrates the above analysis results to generate an analysis report that includes quantitative indicators (such as volume change rate, maximum expansion rate, and maximum contraction rate) and qualitative descriptions (such as dominant change patterns and their confidence levels), providing decision support for doctors.

[0069] Reference Figure 5 The intelligent follow-up plan generation engine 2 for lung nodules is connected to the intelligent matching and change analysis engine 1 for lung nodules. It is used to receive the nodule change analysis results and generate personalized follow-up plans based on clinical guidelines. This engine includes a guideline management module 21, a risk assessment module 22, and a follow-up plan generation module 23.

[0070] The guideline management module 21 is used to store and maintain the clinical follow-up guideline rule base, and select the applicable follow-up rules according to different types of nodules.

[0071] In one embodiment of the present invention, the module maintains a rule base of multiple authoritative guidelines for pulmonary nodule follow-up, including guidelines from the NCCN (National Comprehensive Cancer Network), Fleischner Society, and ACCP (American College of Chest Physicians). The rule base is stored in a structured data format, and each rule contains a condition part and an action part. The conditions involve nodule size, density, risk factors, etc., and the actions include follow-up interval, examination method, etc.

[0072] This module supports version management of guideline rules. When a guideline is updated, the rule base can be updated through the management interface. Preferably, rule updates adopt an incremental strategy, retaining historical versions and marking their effective dates to ensure that the system can adapt to dynamic changes in the guideline.

[0073] In addition, this module provides a rule conflict detection and handling mechanism. When different guidelines give different recommendations for the same type of nodule, the system will handle the situation according to preset priority rules or the preferred strategies of medical institutions. Generally, NCCN guidelines have the highest priority, followed by Fleischner Society and ACCP guidelines.

[0074] The risk assessment module 22 is connected to the lung nodule intelligent matching and change analysis engine 1 to calculate risk scores based on the morphological characteristics and change patterns of nodules, and to generate a comprehensive risk assessment by integrating clinical factors.

[0075] In its implementation, this module constructs a multi-factor risk assessment model that considers both imaging features and clinical factors. Imaging features include nodule size, density, morphological characteristics (such as lobulation and spiculation), and change characteristics (such as volume doubling time and growth pattern); clinical factors include age, sex, smoking history, and history of lung disease.

[0076] The risk score is calculated using a weighted scoring method, with different factors assigned different weights based on their correlation with malignancy. Preferably, the weight coefficients are determined through big data analysis and machine learning methods and are updated regularly to reflect new clinical findings.

[0077] In one embodiment of the present invention, nodules are classified into four levels based on risk scores: low risk (<10%), low-to-medium risk (10%–25%), medium-to-high risk (25%–65%), and high risk (>65%). These thresholds are determined based on a large amount of clinical data and follow-up results, and can effectively distinguish nodules of different risk levels.

[0078] In addition, this module also provides a risk trend analysis function, which identifies rising or falling risk trends by comparing risk scores from continuous follow-up, providing doctors with additional decision-making information.

[0079] The follow-up plan generation module 23 is connected to the guideline management module 21 and the risk assessment module 22. It is used to formulate follow-up intervals based on risk assessment results and guideline rules, and generate personalized inspection methods and follow-up processes.

[0080] In one embodiment of the invention, the module first queries the applicable guidelines based on the nodule type and risk level. Then, based on the specific risk score, it makes refined adjustments within the follow-up interval recommended by the guidelines. For example, for low-to-medium risk nodules where the guidelines recommend 3-6 months of follow-up, if the risk score is close to the low-risk threshold, a 6-month follow-up is recommended; if it is close to the medium-to-high-risk threshold, a 3-month follow-up is recommended.

[0081] In addition to follow-up intervals, this module also recommends appropriate examination methods based on nodule characteristics. Generally, low-dose CT (LDCT) is recommended for pure ground-glass nodules (GGN); for partially solid or solid nodules, standard-dose CT or enhanced CT may be recommended depending on nodule size and risk score.

[0082] In addition, this module generates a complete follow-up process, including pre-examination preparation, suggested examination parameter settings, and follow-up reminder time points. Preferably, for high-risk nodules, the system will suggest reminding patients 1-2 weeks before routine follow-up to ensure that patients do not miss key examinations.

[0083] Ultimately, the generated follow-up plan will be output in a standardized format, including patient information, nodule information, follow-up recommendations, basis and precautions, and the display format can be adjusted according to the needs of medical institutions.

[0084] Reference Figure 6 The cloud storage and collaborative service module 3 is connected to the lung nodule intelligent matching and change analysis engine 1 and the lung nodule intelligent follow-up plan generation engine 2. It is used to store patient image data, nodule analysis results, and follow-up plans, and provides multi-terminal data access and collaborative services. This module includes a data storage unit 31, a data synchronization unit 32, and an access control unit 33.

[0085] Data storage unit 31 is used to store raw patient image data and analysis results, and to maintain a nodule feature database and follow-up records.

[0086] In one embodiment of the invention, the unit employs a hierarchical storage architecture, allocating data to different storage tiers based on data access frequency and importance. High-frequency access data (such as recent examination results and nodule information currently being followed up) is stored in a high-performance storage tier (such as SSD); historical data and long-term archived data are stored in a cost-effective storage tier (such as HDD or object storage).

[0087] The data is organized using a patient-centered, structured approach, with all relevant data (images, reports, analysis results, follow-up plans, etc.) linked by patient ID to form a complete chain of follow-up records. Preferably, image data is stored in DICOM format, while analysis results and follow-up plans are stored in JSON format to ensure data standardization and interoperability.

[0088] In addition, the unit also implements a data backup and recovery mechanism. Critical data adopts a multi-copy strategy and performs full backups regularly to ensure data security and system reliability.

[0089] The data synchronization unit 32 is connected to the data storage unit 31 to ensure the consistency of data access across multiple terminals and to manage data versions and update conflicts.

[0090] In its implementation, this unit uses a publish-subscribe model to manage data updates and synchronization. When data changes, a change event is published to notify all terminals subscribed to that data, ensuring the consistency of data views across all terminals. Preferably, to reduce data transmission volume, an incremental synchronization strategy is adopted, transmitting only the changed portion.

[0091] For low-bandwidth or unstable network environments, this unit implements an offline operation and data synchronization mechanism. Terminals can view cached data and record operations while offline, and automatically synchronize changes once the network is restored. To handle potential update conflicts, the system implements a conflict resolution strategy based on timestamps and operation priorities.

[0092] In addition, for large datasets (such as CT sequences), this unit supports on-demand loading and progressive transmission, prioritizing the transmission of key slices and regions of interest to ensure a good user experience under limited bandwidth conditions.

[0093] The access control unit 33 is connected to the data storage unit 31 and is used to assign data access permissions according to user roles and implement data security policies and privacy protection measures.

[0094] In one embodiment of the present invention, the unit employs a role-based access control (RBAC) model, predefining various roles (such as system administrator, attending physician, resident physician, patient, etc.), each role having a specific set of operational permissions. Preferably, the granularity of permissions is precise to specific data types and operation types, such as viewing patient A's imaging data, modifying follow-up plans, etc.

[0095] For sensitive operations (such as modifying diagnostic results or adjusting follow-up plans), the system implements a two-factor authentication mechanism, requiring users to provide additional identity verification information. Furthermore, all critical operations are logged in detailed audit logs, including the operator, operation time, operation content, and scope of impact.

[0096] To ensure secure data transmission, this unit employs end-to-end encryption technology, encrypting all data during transmission to prevent eavesdropping and man-in-the-middle attacks. Furthermore, static data is also stored encrypted, and sensitive information (such as patient identification) is anonymized to ensure compliance with medical data privacy regulations.

[0097] Reference Figure 7 The early warning module 4 is connected to the lung nodule intelligent matching and change analysis engine 1 and the lung nodule intelligent follow-up plan generation engine 2. It is used to generate early warning information based on nodule change patterns and preset thresholds, and send early warning notifications to doctors and patients through multiple channels. This module includes an early warning rule configuration unit 41, an early warning generation unit 42, and a notification distribution unit 43.

[0098] The early warning rule configuration unit 41 is used to set early warning thresholds for different levels, configure early warning triggering conditions and response strategies.

[0099] In one embodiment of the present invention, the unit supports a multi-level early warning mechanism, typically setting three levels of early warning: alert level (green), attention level (yellow), and warning level (red). Different levels correspond to different early warning thresholds, for example: alert level: volume growth rate 5% to 15%, or slight morphological changes; attention level: volume growth rate 15% to 30%, or morphological changes such as burrs or lobulation appear; warning level: volume growth rate > 30%, or volume doubling time < 100 days, or obvious invasive characteristics appear. These thresholds can be adjusted by medical institutions based on their clinical practice. Preferably, the system supports differentiated thresholds for different types of nodules; for example, the volume change threshold for pure ground-glass nodules can be appropriately relaxed, while a stricter standard is used for some solid nodules.

[0100] Furthermore, this unit also supports composite early warning rules, which means that a combination of multiple conditions triggers an early warning. For example, a composite rule of volume growth rate > 10% AND shape index change > 0.2 can more accurately capture changes in potential risks.

[0101] The early warning generation unit 42 is connected to the early warning rule configuration unit 41 and the lung nodule intelligent matching and change analysis engine 1. It is used to monitor whether the nodule change parameters reach the early warning threshold and generate early warning information of different levels.

[0102] In its implementation, this unit receives analysis results from the lung nodule intelligent matching and change analysis engine 1 in real time and matches them with the early warning rules. When a situation that meets the early warning conditions is detected, an early warning message is generated, which includes the early warning level, the triggering reason, relevant data, and suggested actions.

[0103] Preferably, the warning information adopts a standardized format and includes the following fields: Alert ID: Unique Identifier Warning Level: Alert / Attention / Warning Patient information: ID, name, etc. Nodule information: location, size, characteristics, etc. Triggering conditions: Specific parameters and thresholds that trigger the alert. Description of Changes: Qualitative and quantitative description of nodule changes Suggested action: System-recommended follow-up actions Generation Time: The timestamp when the alert was generated. In addition, this unit implements an early warning priority management mechanism, with higher-level warnings having higher processing priority. For warning-level warnings, the system will mark them as requiring immediate action, ensuring that important warnings receive a timely response.

[0104] The notification distribution unit 43 is connected to the early warning generation unit 42 and is used to select the notification method according to the user's preference and send early warning information through channels such as SMS, email, and application push.

[0105] In one embodiment of the present invention, the unit implements a multi-channel notification strategy, selecting an appropriate notification channel based on the alert level and user preferences. Generally, alerts at the prompt level are sent via in-app messages or emails, alerts at the attention level are accompanied by SMS notifications, and alerts at the warning level are sent via multiple channels in parallel (e.g., simultaneously sending app push notifications, SMS messages, and emails).

[0106] To ensure that important notifications are not missed, this unit implements a notification confirmation and escalation mechanism. If a high-priority alert is not confirmed within a specified time (e.g., 4 hours), the system will automatically escalate the notification method, for example, from app push to telephone reminder.

[0107] Furthermore, this unit supports personalized notification content, adjusting the level of professionalism and detail based on the recipient's role (doctor / patient). For doctors, notifications include technical terminology and detailed data; for patients, they use plain language, focusing on reminders and guidance for subsequent steps.

[0108] Reference Figure 8 The user management and operation module 5 is connected to the cloud storage and collaboration service module 3, and is used to provide user interfaces for the doctor's workstation, doctor's mobile terminal, and patient terminal, manage user permissions, and process user operation requests. This module includes a doctor's workstation terminal 51, a doctor's mobile terminal 52, a patient terminal 53, a role management unit 54, and an operation log unit 55.

[0109] The physician workstation terminal 51 provides a complete functional interface for nodule analysis and follow-up management, supporting advanced visualization and interactive operation.

[0110] In one embodiment of the present invention, the doctor workstation terminal 51 adopts a modular design and includes the following functional areas: Patient list area: Displays a list of patients awaiting treatment, supporting sorting and filtering by different criteria (such as name, examination date, and alert level). Image viewing area: Provides professional DICOM viewing functions, supporting advanced operations such as window width and level adjustment, multi-plane reconstruction (MPR), and maximum density projection (MIP). Nodule Analysis Area: Displays nodule location, size, morphological characteristics, and historical trends; supports manual measurement and labeling. Follow-up Management Area: Displays follow-up plans and historical records, supports adjustments to follow-up plans, and sends notifications. This terminal supports high-resolution displays (4K resolution recommended) and multi-screen operation, optimizing doctors' workflows. The interface layout features an adaptive design, allowing the displayed content and proportions to be adjusted according to different work scenarios.

[0111] In addition, the terminal integrates with existing hospital systems (such as PACS, HIS, RIS), supports single sign-on and data exchange, reduces repetitive operations, and improves work efficiency.

[0112] The doctor's mobile terminal 52 is connected to the doctor's workstation terminal 51 to provide follow-up progress viewing and key operation functions, receive early warning notifications, and support rapid response.

[0113] In its implementation, the doctor's mobile terminal 52 adopts a responsive design to adapt to mobile devices of different sizes. Its key functionalities lie in information viewing and quick operation, including: Task list: Displays patient and alert information requiring processing, sorted by priority. Nodule Overview: Presents key nodule information and trends in a concise format. Follow-up Management: View and confirm follow-up plans, and process follow-up reminders. Notification Center: Receives and processes system alerts and other notifications. To adapt to the characteristics of mobile device use, the interface adopts a touch-friendly design, and key operations can be completed through simple gestures. Preferably, it supports voice input and fingerprint / facial recognition login to improve operational efficiency and security in mobile scenarios.

[0114] In addition, the terminal is capable of working offline, critical data can be cached locally, and synchronized information can still be viewed when the network is unstable, and it will automatically synchronize and update after the network is restored.

[0115] The patient terminal 53 is connected to the cloud storage and collaborative service module 3 to view personal nodule information and follow-up plans, and to receive examination reminders and health advice.

[0116] In one embodiment of the present invention, the patient terminal 53 adopts a simple and intuitive design style, avoiding excessive technical terms and complex data, and focuses on providing the following functions: Nodule Information: Presenting basic information about the nodules and doctor's advice in an easy-to-understand manner. Follow-up Schedule: Displays follow-up plans and reminders, supports calendar synchronization and appointment functions. Inspection guidance: Provides guidance information on pre-inspection preparation, precautions, etc. Health education: Providing health knowledge and self-management advice related to lung nodules. To improve patient compliance, the terminal features a smart reminder function, sending multi-level reminders (such as a one-week advance notice, a three-day advance notice, and a confirmation on the day of the follow-up) as the follow-up date approaches, and also supports exporting the follow-up plan to commonly used calendar applications.

[0117] In addition, the terminal provides simple trend charts and comparison views to help patients intuitively understand the changes in nodules, but does not provide original medical images to avoid unnecessary worries for patients due to a lack of professional knowledge.

[0118] The role management unit 54 is used to configure different roles such as administrator, doctor, and patient, and to assign system operation permissions to each role.

[0119] In one embodiment of the invention, the unit predefines multiple roles, including system administrator, department head, attending physician, resident physician, technician, patient, and guardian. Each role has a specific set of permissions; for example, the system administrator can manage all users and system configurations, the department head can view all patient data within the department, the attending physician can manage the patients assigned to them, and patients can only view their own information.

[0120] This unit supports a hierarchical structure of roles and permission inheritance, with higher-level roles automatically including all permissions of lower-level roles. Furthermore, the system supports custom role extensions, allowing medical institutions to create new roles and configure specific permissions according to their needs.

[0121] Preferably, the granularity of access control is precise to the functional operation level, and the administrator can precisely configure the types of operations that can be performed (such as viewing, adding, modifying, and deleting) and the objects of operation (such as patient information, nodule data, and follow-up plans) for each role.

[0122] The operation log unit 55 is connected to the role management unit 54 and is used to record the user's system operation history, providing operation auditing and statistical analysis functions.

[0123] In its implementation, this unit records all critical operations in the system. Each log entry includes fields such as operator information, operation time, operation type, operation object, operation parameters, and operation result. The logs are stored in a structured format for easy subsequent querying and analysis.

[0124] To meet medical data security and compliance requirements, this unit implements an immutable log mechanism, employing a chained structure and digital signature technology to ensure the integrity and authenticity of the logs. The log retention period complies with medical data management regulations, typically no less than 5 years after the patient's last follow-up visit.

[0125] In addition, this unit provides log query and analysis functions, supporting log filtering by various conditions (such as time range, operation type, operator, etc.) and generating operation audit reports for compliance checks and system usage optimization. Preferably, the system can also identify abnormal operation patterns (such as a large number of accesses to different patient data in a short period of time) and issue timely security warnings.

[0126] The Lung Nodule Intelligent Matching and Change Analysis Engine 1 is implemented using GPU acceleration, including parallel processing optimization for curvature calculation, distributed acceleration for geodesic distance field calculation, and parallel computing architecture for multi-scale feature extraction. Through GPU acceleration, it achieves efficient processing of large amounts of nodule data and improves system response speed.

[0127] In one embodiment of the present invention, the system adopts the CUDA architecture to implement GPU parallel computing, which is optimized for computationally intensive tasks, mainly including the following aspects: First, optimizations are made for parallel processing of curvature calculation. Curvature calculation involves a large amount of local surface fitting and differential calculations, which are computationally intensive but easily parallelized. The system divides the nodal surface mesh into multiple blocks, with each block assigned to a GPU computing unit to achieve parallel computation. Preferably, the block division adopts an overlapping strategy, and the calculation results in the boundary regions are weighted and fused to eliminate boundary effects. This parallel strategy improves the curvature calculation speed by 15-20 times, significantly reducing processing time.

[0128] Secondly, distributed acceleration of geodesic distance field calculation. Geodesic distance calculation is based on the Fast Marching Method, which has strong dependencies in traditional implementations and is difficult to parallelize directly. This system adopts a block-based Fast Marching algorithm, dividing the surface into multiple blocks. First, local distance fields are calculated in parallel within each block, and then the global distance field is iteratively updated through boundary conditions. This distributed computing strategy improves the speed of geodesic distance field calculation by 8-10 times.

[0129] Third, the parallel computing architecture for multi-scale feature extraction. Feature extraction tasks at different scales are independent of each other, naturally suited for parallel processing. The system distributes feature extraction tasks at each scale to different GPU streams for parallel execution, making full use of GPU computing resources. Furthermore, within each scale, the computation of different types of features (such as curvature statistics, Willmore energy, and shape index) is also performed in parallel, further improving computational efficiency. This multi-level parallel architecture improves feature extraction speed by 12-15 times.

[0130] In terms of hardware configuration, the system is recommended to use an NVIDIA RTX A6000 or higher performance GPU, with at least 48GB of memory, to meet the processing needs of large CT data and complex mesh models. Preferably, the system supports multi-GPU configurations, which can be flexibly expanded according to the size and processing needs of the medical institution.

[0131] With the aforementioned GPU acceleration technology, this system can complete the full analysis and processing of a single nodule within 3 seconds, achieving a speedup of 15-20 times compared to CPU, effectively supporting real-time clinical application scenarios.

[0132] In summary, the intelligent management system for lung nodule follow-up proposed in this invention, through the innovative application of differential geometry theory and the combination of GPU accelerated computing technology, achieves precise matching of lung nodules, accurate quantification of changes, and generation of personalized follow-up plans, providing a comprehensive technical solution for lung nodule follow-up management and having significant clinical application value.

[0133] The above description is merely a preferred embodiment of the present invention and does not limit the scope of patent protection of the present invention. Any equivalent structural transformations made under the inventive concept of the present invention using the contents of the specification and drawings of the present invention, or direct / indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A smart management system for lung nodule follow-up, characterized in that, include: The lung nodule intelligent matching and change analysis engine is used for: The surface of the lung nodules is modeled as a differential manifold, and the curvature characteristics and topological properties of the surface are calculated. Construct multi-scale differential invariant features to analyze nodule morphological changes at different scales; Based on manifold geodesic distance, the precise changes between nodules at different time points are quantified; The intelligent follow-up plan generation engine for lung nodules, connected to the intelligent matching and change analysis engine for lung nodules, is used for: Receive the nodule change analysis results sent by the lung nodule intelligent matching and change analysis engine; Based on the nodule change analysis results, a personalized follow-up plan is generated in conjunction with clinical guidelines. The cloud storage and collaborative service module, connected to the lung nodule intelligent matching and change analysis engine and the lung nodule intelligent follow-up plan generation engine, is used for: Store patient imaging data, nodule analysis results, and follow-up plans; Provides multi-terminal data access and collaboration services; The early warning module, connected to the intelligent lung nodule matching and change analysis engine and the intelligent lung nodule follow-up plan generation engine, is used for: Early warning information is generated based on nodule change patterns and preset thresholds; Warning notifications were sent to doctors and patients through multiple channels; The user management and operation module, connected to the cloud storage and collaboration service module, is used for: Provides user interfaces for doctors' workstations, doctors' mobile devices, and patients; Manage user permissions and process user action requests.

2. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The intelligent matching and change analysis engine for lung nodules includes: Boundary representation module, used for: The boundaries of segmented lung nodules in CT images are constructed as a three-dimensional mesh model; The nodule surface is mapped to the parameter space through parameter mapping; Calculate the Gaussian curvature, mean curvature, and principal curvature of the nodule surface; The curvature spectrum analysis module, connected to the boundary characterization module, is used for: Construct a histogram of surface curvature distribution of nodules; Extract curvature extremum regions and spatial distribution characteristics; Generate a geometric fingerprint feature set for the nodules.

3. The intelligent management system for lung nodule follow-up according to claim 2, characterized in that, The lung nodule intelligent matching and change analysis engine also includes: The multi-scale feature extraction module is used for: Multiscale representations of nodule surfaces are generated using Laplacian operators with different parameters; Calculate the differential invariants and shape exponential distribution at various scales; Extract feature evolution patterns and stable feature regions; The feature fusion module, connected to the multi-scale feature extraction module, is used for: Assess the importance of features at each scale; Adaptive selection of the optimal scale combination; Construct a comprehensive feature representation that integrates multi-scale information.

4. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The lung nodule intelligent matching and change analysis engine also includes: The manifold registration module is used for: Initial registration of nodules at different time points is performed based on principal inertial axis alignment; Accurate registration is achieved using a feature-preserving non-rigid registration algorithm. Establish the correspondence between points on the nodule surface; The geodesic distance analysis module, connected to the manifold registration module, is used for: Define a local velocity function based on surface features; Calculate the geodesic distance field on the surface of the nodule; Identify and quantify local expansion and contraction regions; The change pattern recognition module, connected to the geodetic distance analysis module, is used for: The nodules were divided into different variation regions based on the changes in geodesic distance; Identify typical change patterns and calculate quantitative indicators; Generate heatmaps and analysis reports of the changes in distribution.

5. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The intelligent follow-up solution generation engine for lung nodules includes: The guide management module is used for: Store and maintain a clinical follow-up guideline rule base; Select appropriate follow-up rules based on different types of nodules; The risk assessment module, connected to the lung nodule intelligent matching and change analysis engine, is used for: Risk scores are calculated based on the morphological characteristics and change patterns of nodules; Generate a comprehensive risk assessment by integrating clinical factors; The follow-up plan generation module, connected to the guideline management module and the risk assessment module, is used for: Follow-up intervals should be determined based on risk assessment results and guidelines. Generate personalized inspection methods and follow-up processes.

6. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The cloud storage and collaboration service module includes: Data storage unit, used for: Stores raw patient image data and analysis results; Maintain a nodule characteristic database and follow-up records; A data synchronization unit, connected to the data storage unit, is used for: Ensure consistency of data access across multiple devices; Managing data version and update conflicts; The access control unit, connected to the data storage unit, is used for: Data access permissions are assigned based on user roles; Implement data security strategies and privacy protection measures.

7. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The early warning module includes: The early warning rule configuration unit is used for: Set different levels of warning thresholds; Configure early warning trigger conditions and response strategies; The early warning generation unit, connected to the early warning rule configuration unit and the lung nodule intelligent matching and change analysis engine, is used for: Monitor whether the parameters of nodule changes reach the warning threshold; Generate early warning information at different levels; The notification distribution unit, connected to the warning generation unit, is used for: Choose the notification method based on user preferences; Warning information is sent via SMS, email, and app push notifications.

8. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The user management and operation module includes: Doctor workstation terminals are used for: Provides a complete functional interface for nodule analysis and follow-up management; Supports advanced visualization and interactive operation; The doctor's mobile terminal, connected to the doctor's workstation terminal, is used for: Provides follow-up progress viewing and key operation functions; Receive early warning notifications and support rapid response; The patient terminal, connected to the cloud storage and collaboration service module, is used for: View individual nodule information and follow-up plan; Receive check-up reminders and health advice.

9. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The user management and operation module also includes: Role management unit, used for: Configure different roles for administrators, doctors, and patients; Assign system operation permissions to each role; The operation log unit, connected to the role management unit, is used for: Record the user's system operation history; Provides operational auditing and statistical analysis functions.

10. The intelligent management system for lung nodule follow-up according to claim 1, characterized in that, The intelligent lung nodule matching and change analysis engine is implemented using GPU acceleration and includes: Parallel processing optimizations for curvature calculation; Distributed acceleration of geodesic distance field calculation; Parallel computing architecture for multi-scale feature extraction; The GPU acceleration enables efficient processing of large amounts of nodule data, improving system response speed.

Images