Intelligent detection and activity evaluation method and system for chest CT image lesions of pulmonary tuberculosis

Abstract

This invention discloses a method and system for intelligent detection and activity assessment of pulmonary tuberculosis lesions on chest CT images. The method includes the following steps: acquiring chest CT image sequences of the same pulmonary tuberculosis patient at at least two follow-up time points to obtain suspected lesion areas at each time point; constructing a spatial topological mapping relationship between the suspected lesion areas and the lobes, segments, and bronchial bifurcation levels, and assigning a unique topological node identifier to each suspected lesion area; establishing a cross-time temporal correspondence relationship of lesions based on the topological node identifiers, satisfying topological consistency constraints; calculating a self-consistent decay index characterizing the temporal stability of lesions based on the temporal correspondence relationship; calculating the changes in the topological node identifiers of suspected lesion areas at different time points to obtain a topological evolution index characterizing the spatial invasiveness of lesions; and generating and outputting an auxiliary analysis report on lesion activity. This invention can improve the reliability of lesion tracking in multi-temporal CT image follow-up analysis.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and in particular to a method and system for intelligent detection and activity assessment of lesions on chest CT images of pulmonary tuberculosis. Background Technology

[0002] Pulmonary tuberculosis is a chronic infectious disease caused by Mycobacterium tuberculosis. Its lung lesions may manifest as persistent, recurrent, or disseminated over a long period of time. Clinically, comparing and analyzing a series of chest CT images of the same patient at different time points (such as before treatment, during treatment, and during follow-up) is a key means of assessing changes in lesions, judging disease activity, and evaluating treatment efficacy.

[0003] Currently, research on intelligent analysis of pulmonary tuberculosis lesions based on CT images mainly focuses on the detection, segmentation, and classification of lesions at a single time point. When multi-time-point follow-up image analysis is involved, the mainstream technical approach usually follows the framework of "independent detection first, then correlation and comparison." Specifically, suspected lesion areas are first detected independently on CT images at each time point; then, the correspondence between lesions at different time points is established mainly based on the degree of overlap of these areas in three-dimensional space (spatial overlap method), the similarity of image features (feature matching method), or a combination of both, to calculate the rate of change of features such as volume and density to assess their activity.

[0004] However, the inventors recognized that the aforementioned method has an inherent flaw in its core logic: it fails to model and constrain the "continuity of identity of the same lesion over time." Because the lungs undergo significant deformation during respiration, and patient positions and lung inflation vary between different scans, the spatial coordinates and morphological appearance of lesions at the same anatomical location may change considerably on images. In this situation, relying solely on spatial overlap or feature similarity for matching can easily misclassify changes in the image appearance of the same lesion caused by physiological movement or scanning differences as different lesions; conversely, it may also mistakenly associate different lesions that are spatially adjacent but anatomically discontinuous as the same lesion. This "mismatch" problem directly leads to errors in subsequent calculations of the rate of change, significantly reducing the reliability of activity assessment. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for intelligent detection and activity assessment of pulmonary tuberculosis lesions on chest CT images, comprising the following steps:

[0007] Acquire chest CT image sequences of the same pulmonary tuberculosis patient at at least two follow-up time points, and perform registration and lesion detection on the chest CT image sequences to obtain suspected lesion areas at each time point;

[0008] Based on the lung anatomy segmentation results of the chest CT images, a spatial topological mapping relationship is constructed between the suspected lesion area and the lung lobe, lung segment and bronchial bifurcation levels, and a unique topological node identifier is assigned to each suspected lesion area.

[0009] Based on the topological node identifiers, a cross-time temporal correspondence relationship of lesions is established that satisfies the topological consistency constraint;

[0010] Based on the aforementioned temporal correspondence, a self-consistent decay index characterizing the temporal stability of lesions is calculated.

[0011] The changes in topological node identifiers of the suspected lesion area at different time points are calculated to obtain a topological evolution index characterizing the spatial invasiveness of the lesion.

[0012] The self-consistency decay index, the topological evolution index, and the patient's associated clinical information are acquired and fused to generate and output a lesion activity auxiliary analysis report to assist clinical decision-making.

[0013] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, the spatial topological mapping relationship is achieved by constructing a hierarchical topological model of lung anatomy that includes lobe level, lung segment level and bronchial bifurcation level.

[0014] Each suspected lesion region is mapped and assigned to a unique topological node in the hierarchical topology model based on its spatial location.

[0015] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, the step of establishing a temporal correspondence relationship of lesions across time satisfying topological consistency constraints includes:

[0016] Construct topology constraint rules based on the topology node identifiers;

[0017] According to the topology constraint rules, a candidate matching object is determined only when suspected lesion areas at different time points are assigned to the same topology node or to a topology node with a preset adjacency relationship.

[0018] For the candidate matching objects, their morphological features, grayscale statistical features, and texture distribution features are extracted, and cross-time matching is performed based on the extracted features to establish a temporal correspondence.

[0019] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, wherein: the calculation of the self-consistency decay index is only performed on suspected lesion areas that have successfully established a temporal correspondence through the topological constraint rules;

[0020] The similarity is quantified by calculating the changes in the weighted multidimensional image feature vectors of the suspected lesion area at different time points.

[0021] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, wherein: when calculating the self-consistency attenuation index, different weight coefficients are assigned to morphological features, gray-scale statistical features, and texture distribution features, and the weight coefficients are adaptively adjusted according to the anatomical level of the suspected lesion area and the follow-up image performance;

[0022] The adaptive adjustment of the weighting coefficients follows these rules:

[0023] When the suspected lesion area is located at the lung segment or bronchial bifurcation level, the weighting coefficient of the texture distribution feature is increased;

[0024] When the suspected lesion area is located at the lobar level, the weighting coefficient of the morphological feature is increased;

[0025] When cavitation or calcification is observed in follow-up images, the weighting coefficient of the gray-scale statistical features of the corresponding suspected lesion area should be increased.

[0026] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, the topological evolution index is quantified by the following formula:

[0027] ;

[0028] in, A coefficient characterizing whether lesions change between lung lobe levels. A coefficient characterizing whether lesions change between lung segmental levels. This is a distance coefficient, the value of which is calculated based on the path length of the lesion center point moving in the bronchial bifurcation hierarchy topology and obtained after normalization. , , Preset weights related to the migration direction, and satisfying < < This serves as a warning that migration towards the distal bronchi carries a higher risk of spatial invasion.

[0029] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, the fusion and generation step achieves quantitative scoring of lesion activity through the following evaluation function:

[0030] ;or

[0031] ;

[0032] in, It is a self-consistent decay index; As a topological evolution indicator; For clinical information vectors; This is a clinical information correction function used to adjust assessment results based on the patient's specific clinical condition; and It is a modulation factor greater than zero, which is related to the specific level at which the topology shift occurs.

[0033] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, the generation of the lesion activity auxiliary analysis report further includes:

[0034] Based on the quantitative score of lesion activity The system compares the scores with multiple preset risk reference intervals and automatically extracts medical knowledge entries associated with the intervals in which the scores are located from a preset clinical medical knowledge base.

[0035] Quantitative scoring of the lesion activity The self-consistency decay index and the lesion evolution information represented by the topological evolution index, as well as the associated medical knowledge items, are integrated to generate a structured report text and a visual summary.

[0036] As a preferred embodiment of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images described in this invention, the method further includes the steps of report output and system integration.

[0037] The generated lesion activity auxiliary analysis report, which includes the structured report text and visual summary, is automatically packaged into a standardized clinical data exchange format package;

[0038] Through the medical information system integration interface, the standardized data packet is automatically pushed and integrated into the patient's electronic medical record system. At the same time, according to preset rules, a system notification message containing key conclusions of lesion activity quantitative scoring is sent to the attending physician's workstation.

[0039] This invention also provides an intelligent detection and activity assessment system for pulmonary tuberculosis lesions on chest CT images, applied to the above-mentioned assessment method, including:

[0040] The image data acquisition and processing module is configured to acquire chest CT image sequences of the same pulmonary tuberculosis patient at at least two follow-up time points, and to register and detect lesions in the sequences, and output the suspected lesion areas at each time point.

[0041] The topology mapping and identification module is configured to construct a spatial topology mapping relationship between the suspected lesion area and the lung lobe, lung segment and bronchial bifurcation levels based on the lung anatomical structure segmentation results, and assign a unique topology node identifier to each suspected lesion area.

[0042] The time-series correspondence establishment module is configured to establish a cross-time lesion time-series correspondence that satisfies topological consistency constraints based on the topological node identifier;

[0043] The indicator calculation module is configured to calculate a self-consistent decay index representing the temporal stability of lesions based on the time-series correspondence, and to calculate a topological evolution index representing spatial invasiveness based on changes in topological node identifiers.

[0044] The intelligent report generation and decision support module is configured to: integrate the self-consistency decay index, the topological evolution index and the patient's clinical information, and generate a structured lesion activity auxiliary analysis report based on the lesion activity quantitative score and associated clinical medical knowledge.

[0045] A clinical system integration interface is configured to output the auxiliary analysis report to the clinical diagnosis and treatment environment in a standardized format.

[0046] The beneficial effects of this invention are:

[0047] 1. This invention constructs a hierarchical topological model of the lung anatomy and assigns a unique topological node identifier to suspected lesion areas. Then, using the topological node identifier as a constraint, a cross-time sequence correspondence is established only when the lesions are in the same or a preset topological neighborhood. This can fundamentally reduce the risk of mismatch caused by respiratory displacement, scanning differences and local deformation from the perspective of anatomical structural continuity, thereby significantly improving the accuracy and reliability of lesion tracking in multi-temporal CT image follow-up analysis.

[0048] 2. This invention calculates the self-consistent decay index, which characterizes the temporal stability of lesions, and the topological evolution index, which characterizes their spatial invasiveness, respectively. It then integrates these two in conjunction with the patient's clinical information to create a model that achieves a simultaneous and quantitative characterization of the temporal evolution stability of pulmonary tuberculosis lesion activity and its spatial anatomical invasiveness. This overcomes the limitations of existing methods that rely on a single assessment dimension and provides a more comprehensive and objective quantitative assessment basis for clinical practice.

[0049] 3. This invention automatically generates an auxiliary analysis report that integrates quantitative scores, evolution information, visualization maps, and structured suggestions by associating the quantitative lesion activity score generated after fusion assessment with a pre-set clinical medical knowledge base. It can also be integrated with hospital information systems through standardized interfaces, realizing full-process automation and intelligence from image analysis to clinical decision support. While improving assessment efficiency and consistency, it also enhances the interpretability of results and the fit with clinical diagnosis and treatment processes. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0051] Figure 1 This is a flowchart illustrating the overall process of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images according to the present invention.

[0052] Figure 2 This is a flowchart of the topological constraint matching process for the intelligent detection and activity assessment method of pulmonary tuberculosis lesions on chest CT images according to the present invention.

[0053] Figure 3 This is a flowchart illustrating the report generation and system integration process of the intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images according to the present invention. Detailed Implementation

[0054] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0055] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0056] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0057] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0058] Example 1

[0059] Reference Figure 1-3 This is the first embodiment of the present invention, which provides a method for intelligent detection and activity assessment of pulmonary tuberculosis lesions on chest CT images, including the following steps:

[0060] S1: Obtain chest CT image sequences of the same pulmonary tuberculosis patient at at least two follow-up time points, and perform registration and lesion detection on the chest CT image sequences to obtain the suspected lesion areas at each time point.

[0061] Registration, in particular, aims to eliminate spatial deviations in anatomical structures caused by differences in scanning position and respiratory phase, requiring spatial alignment of CT image sequences from different time points. This invention can be implemented using mature image registration algorithms in the field. For example, rigid registration based on mutual information can be used to initially align the entire lung cavity, followed by a non-rigid registration algorithm to finely correct local lung tissue deformation.

[0062] In this process, lesion detection automatically identifies and segments suspected pulmonary tuberculosis lesion areas on CT images at various time points after registration. This operation can be accomplished using efficient lesion detection models in existing technologies, such as semantic segmentation models based on convolutional neural networks. These models are trained on image datasets labeled with pulmonary tuberculosis lesions and can output binarized masks or bounding boxes of suspected lesion areas.

[0063] Through the above operations, step S1 can obtain a set of spatiotemporally aligned standardized data containing preliminary identification of lesion targets, thereby providing standardized input data for subsequent core analysis processes.

[0064] S2: Based on the lung anatomy segmentation results of chest CT images, construct the spatial topological mapping relationship between the suspected lesion area and the lung lobe, lung segment and bronchial bifurcation level, and assign a unique topological node identifier to each suspected lesion area.

[0065] Specifically, the spatial topological mapping relationship is achieved by constructing a hierarchical topological model of lung anatomy that includes the lobe level, lung segment level, and bronchial bifurcation level.

[0066] Each suspected lesion area is mapped and assigned to a unique topological node in the hierarchical topology model based on its spatial location.

[0067] Step S2 above aims to assign a "identity address" with clear anatomical significance to each detected suspected lesion area, based on the inherent hierarchical structure of the lungs. Specifically:

[0068] In one specific implementation, the segmentation results of lung anatomy can be obtained by processing registered CT images using a pre-trained deep learning segmentation model (such as a segmentation network for lobes and segments), or extracted from auxiliary structured reports generated by imaging equipment. The segmentation results should include at least: lobar masks (containing five binary 3D matrices, such as RUL for the right upper lobe, RML for the right middle lobe, RLL for the right lower lobe, LUL for the left upper lobe, and LLL for the left lower lobe), segmental masks (containing approximately 18 binary 3D matrices), and information on the bronchial tree centerline and bifurcation points. These segmentation masks form the geometric basis for constructing a hierarchical topological model.

[0069] In one specific implementation, the construction of a hierarchical topological model of lung anatomy includes the following steps:

[0070] First, obtain the lobar mask, segmental mask, and bronchial tree information provided by the lung anatomy segmentation results. The bronchial tree information includes a series of sequentially connected three-dimensional coordinate points (centerline) and a list of labeled bifurcation point indices. For example, the centerline can be represented as a set of points. The fork point index indicates Which points are branch points (e.g.) , ).

[0071] Next, topology generation and node definition are performed. Topology generation involves using the aforementioned set of bronchial centerline points. Using the bifurcation point index as input, a continuous sequence of coordinate points is divided into different branch segments using a traversal algorithm (such as depth-first search), and a tree-like connected graph is constructed with the bifurcation points as boundaries. This graph is defined here as a bronchial topology graph, where each vertex represents a bifurcation point or terminal point, and each edge represents a segment of the bronchial centerline. Node definition: the entire lung is defined as the root node. Based on the lobar mask, 5 first-level nodes are defined, each node ID corresponding to the mask name (e.g., RUL). Based on the lung segment mask, approximately 18 second-level nodes are defined (e.g., RUL_S1). The spatial region associated with bifurcation points beyond a certain generation (e.g., level 3 and above) in the aforementioned generated bronchial topology graph is defined as a third-level node, whose ID can be derived from its lung segment and branch sequence (e.g., RUL_S1_B2, representing the second critical branch point of the right upper lobe apical segment).

[0072] Next, node spatial extent association is performed, that is, each node is associated with a specific spatial extent description for spatial querying. Specifically, for first- and second-level nodes, their spatial extent is directly associated with the three-dimensional axial bounding box formed by the minimum and maximum voxel indices of the corresponding lung lobe and lung segment masks in the left-right, front-back, and head-foot directions. For third-level nodes, their spatial extent can be associated with a spherical spatial region with a fixed physical radius (e.g., 5mm) centered on the coordinates of the bifurcation point, or determined by tracing the subsegment segmentation mask to which it belongs downstream.

[0073] Then, based on anatomical relationships, directed connections are established in the model data structure: from the root node to all five primary nodes (lung lobes); from each primary node to all secondary nodes (lung segments) anatomically contained within it. For example, RUL is connected to RUL_S1, RUL_S2, and RUL_S3; from each secondary node to all its downstream tertiary nodes (bronchial bifurcation points); at the bronchial topology level, based on the aforementioned generated tree-like connectivity graph, connections representing bronchial accessibility are established between tertiary nodes.

[0074] Ultimately, the model is instantiated as a graph data structure containing a list of nodes and a list of edges. Each node is an object with attributes including at least: node ID, node level, and spatial extent description. Each edge is an object with attributes including at least: parent node ID, child node ID, and relationship type.

[0075] Through this step, the geometric information extracted from the segmentation results is structured into a multi-layer topological map containing lung lobes, lung segments, and bronchial bifurcation points. This map serves as a spatial dictionary, making it possible to map any lesion coordinates to a unique topological identifier, thus laying a decisive foundation for establishing subsequent temporal correspondences.

[0076] In one specific implementation, mapping and assigning a unique topological node involves: calculating the three-dimensional geometric center point (centroid) of the lesion region; then, starting from the finest granularity (tertiary node, bronchial bifurcation level), progressively determining which tertiary node the centroid coordinates fall into. If a match is found, the mapping process terminates, and the lesion is identified as this tertiary node. If no tertiary node is matched, the process checks which secondary node's mask the centroid falls into. If a match is found, the lesion is mapped to the secondary node. If no match is found, the lesion is finally queried and mapped to the primary node where the centroid is located. Finally, based on the mapping result, a structured string identifier is generated. For example, a lesion mapped to the lung segment node "right upper lobe apex segment" can be identified as "RUL_S1"; if mapped to a finer bronchial bifurcation node, the identifier can be "RUL_S1_B2". This identifier is unique for the lesion in a single analysis and directly reflects its precise location in the anatomical structure.

[0077] It should be noted that, through step S2, each discrete radiographic lesion is assigned a unique topological node identifier based on a hierarchical topological model, with a clear anatomical affiliation. The establishment of this topological identifier provides an indispensable and computable logical basis for performing cross-time point lesion matching under topological consistency constraints in the subsequent step S3.

[0078] S3: Based on the topological node identifier, establish a cross-time lesion temporal correspondence that satisfies the topological consistency constraint.

[0079] Specifically, establishing a cross-temporal correspondence between lesions that satisfies topological consistency constraints includes:

[0080] Construct topology constraint rules based on topology node identifiers;

[0081] According to the topology constraint rules, a suspected lesion area at different time points is only considered a candidate matching object when it is assigned to the same topology node or to a topology node with a preset adjacency relationship.

[0082] For candidate matching objects, their morphological features, grayscale statistical features, and texture distribution features are extracted, and cross-time matching is performed based on the extracted features to establish a temporal correspondence.

[0083] The morphological features include, but are not limited to, three-dimensional volume, surface area, sphericity, and longest diameter. Gray-level statistical features include, but are not limited to, the mean, standard deviation, skewness, and kurtosis of CT values ​​within the region. Texture distribution features include, but are not limited to, contrast, correlation, energy, and homogeneity calculated based on the gray-level co-occurrence matrix (GLCM). These three features can sensitively capture key pathophysiological changes in pulmonary tuberculosis lesions during follow-up from different dimensions.

[0084] In one specific implementation, the topological constraint rule is a judgment function used to determine whether the topological node identifiers (denoted as ID_T1 and ID_T2) of two lesions from different time points satisfy the anatomical spatial continuity condition. Its output is a Boolean value (true or false). The specific rules are as follows:

[0085] For identical topological nodes: if ID_T1 and ID_T2 are exactly the same, then the strongest constraint is satisfied.

[0086] Preset adjacency relationships: If nodes are different, adjacency is determined based on the hierarchical topology model constructed by S2, including but not limited to the following cases:

[0087] Adjacency within the bronchial tree: At the bronchial bifurcation level (third-level nodes), two nodes are considered adjacent if they are directly connected by an edge in the bronchial topology (i.e., they belong to the upstream and downstream bifurcation points of the same bronchial segment). For example, nodes RUL_S1_B2 and RUL_S1_B3.

[0088] Spatial adjacency within a lung segment: If two lesions are mapped to different tertiary nodes under the same lung segment, and the three-dimensional Euclidean distance of their spatial range (such as the centroid) is less than a preset threshold (such as 10 mm), they are determined to be adjacent.

[0089] Anatomical adjacency across lung segments: If two lesions are mapped to different secondary nodes (lung segments), a predefined "lung segment adjacency table" needs to be consulted for determination. This table is pre-established based on anatomical knowledge; for example, the apical segment (RUL_S1) and posterior segment (RUL_S2) of the right upper lobe are adjacent lung segments. If the secondary nodes corresponding to ID_T1 and ID_T2 are marked as adjacent in this table, then the adjacency relationship is determined to be satisfied.

[0090] Traverse all lesion pairs at different time points and apply the above topological constraint rules for filtering. Only when the "same topological node" or "preset adjacency relationship" is satisfied, the lesion pair is retained as a candidate matching object.

[0091] In one specific implementation, cross-time matching based on the extracted features includes:

[0092] For the set of candidate matching objects filtered by topological constraints, the similarity of feature vectors (such as cosine similarity) between each pair of candidate objects is calculated. Then, an optimal matching algorithm (such as the Hungarian algorithm) is used to find the unique matching object with the most similar features for each lesion at another time point among the candidate pairs that satisfy the topological constraints. A similarity threshold can be set; if the similarity of the optimal match is lower than this threshold, the match is considered to have failed, and the lesion has no corresponding relationship at that time point.

[0093] The successfully established matching pairs constitute a temporal correspondence between lesions across time. This relationship ensures that the associated lesions are not only similar in image features, but more importantly, they have a reasonable position of continuity or adjacency in anatomical structure.

[0094] It should be noted that step S3, through a two-stage strategy of "topological constraint first, feature matching later," transforms lesion tracking from a global search to local identification. This significantly reduces the risk of misclassifying lesions in different locations as the same lesion and avoids matching drift caused by image artifacts or positional differences, thus providing a reliable data foundation for subsequent calculation of true and reliable lesion evolution indicators.

[0095] S4: Based on the temporal correspondence, calculate the self-consistent decay index that characterizes the temporal stability of lesions.

[0096] Specifically, the self-consistency decay index is calculated only for suspected lesion regions that have successfully established temporal correspondence through topological constraint rules; it is quantified by calculating the similarity changes between the weighted multidimensional image feature vectors of suspected lesion regions at different time points.

[0097] Furthermore, when calculating the self-consistency decay index, different weight coefficients are assigned to morphological features, gray-scale statistical features, and texture distribution features, and the weight coefficients are adaptively adjusted according to the anatomical level (lobe, segment, or bronchial bifurcation level) of the suspected lesion area and the follow-up imaging manifestations.

[0098] The adaptive adjustment of the weighting coefficients follows these rules:

[0099] When the suspected lesion area is located at the lung segment or bronchial bifurcation level, increase the weight coefficient of the texture distribution feature;

[0100] When the suspected lesion area is located at the lobar level, the weighting coefficient of morphological features is increased;

[0101] When cavitation or calcification is observed in follow-up images, the weighting coefficient of the gray-scale statistical features of the corresponding suspected lesion area is increased.

[0102] The adaptively adjusted weight coefficients will be used to construct a weighted feature vector, and then to calculate a weighted similarity that better reflects the specific anatomical and pathological background, so as to finally quantify the self-consistency decay index.

[0103] In one specific implementation, the process of calculating the weighted similarity is as follows: For each having The lesions with time-series correspondence at each time point are first identified at each time point. Based on the morphology, grayscale, and texture adjustments extracted in step S3, original feature sub-vectors are constructed, denoted as: morphological feature sub-vector. Gray-scale statistical feature vectors Texture distribution feature sub-vectors Subsequently, based on the current anatomical level and imaging manifestations of the lesion, the aforementioned adaptive adjustment rules were applied to determine the weighting coefficients. , , (satisfy + + =1). Using these weights, construct the weighted feature vector for that time point. This weighting is achieved by concatenating the square roots of each original feature vector with its corresponding weight, and its mathematical expression is as follows:

[0104] ;

[0105] in, To indicate the square root operation, use the superscript. This represents the transpose of a vector. The use of square root weighting ensures that the weights directly influence the similarity results in a predetermined linear proportion during subsequent cosine similarity calculations, thereby guaranteeing the correct effectiveness of the weighting mechanism in similarity measurement.

[0106] Finally, based on the weighted feature vector Calculate the effect of the lesion on all adjacent time points. Weighted feature vector similarity between In a preferred embodiment, cosine similarity is used for calculation.

[0107] In one specific implementation, the self-consistency decay index is quantified based on the aforementioned weighted similarity. This index characterizes the trend and degree to which the weighted feature similarity decreases over time. Its mathematical expression is defined as: ,in, The total number of time points for which a temporal correspondence was established for this lesion; This represents the summation operation, which sums all values ​​for the lesion. The weighted similarity calculated for each adjacent time point is summed; the physical meaning of this index is the average inconsistency of the weighted features over time. The closer the value is to 0, the more consistent the imaging characteristics of the lesion are at different time points, and the higher the temporal stability. The larger the value, the greater the fluctuation of the feature over time, and the worse the stability. For example: a lesion at three time points ( The weighted similarity of (=3) is , Then the self-consistency decay index This result indicates that the lesion maintained high consistency of characteristics and good stability during the follow-up period.

[0108] In one specific implementation, the adaptive adjustment process of the weighting coefficients is as follows:

[0109] Weight coefficient definition and initialization: Let the weight coefficient triples be... These correspond to the weights of morphological features, grayscale statistical features, and texture distribution features, respectively. The initial default weights are set to equal values, i.e. Satisfying the normalization condition + + =1.

[0110] Examples of weight adjustment based on anatomical hierarchy, such as:

[0111] When lesions are located at the lung segment or bronchial bifurcation level, the weight of texture distribution features should be increased according to the above rules. A specific numerical adjustment scheme is as follows: =0.25, =0.25, =0.5.

[0112] At this point, the texture feature weights The weights were significantly increased to 0.5, while the weights for morphology and grayscale features were correspondingly reduced to highlight the importance of microstructural changes in the region.

[0113] When the lesion is located at the lobar level, the weight of morphological features should be increased according to the rules. A specific numerical adjustment scheme is as follows: =0.5, =0.25, =0.25.

[0114] At this point, the morphological feature weights The value was increased to 0.5 to emphasize the dominant role of overall lesion size and contour evolution in the assessment of this macro-region.

[0115] An example of weighting adjustments based on follow-up image performance, such as:

[0116] When cavitation is observed in follow-up images, i.e., when a low-density cavity region is detected within the lesion, the weight of the grayscale statistical feature should be increased according to the rules. A specific numerical adjustment scheme is as follows: based on the existing weights, increase the grayscale feature weight to 0.6 and renormalize the remaining weights. For example, if the weights before adjustment were ( =0.3, =0.3, =0.4), then the adjusted value is: =0.2, =0.6, =0.2, this adjustment is designed to focus on significant changes in density distribution in order to capture pathological processes such as necrosis and liquefaction.

[0117] When calcification is observed in the lesions during follow-up imaging, specifically when high-density calcifications are detected within the lesions, the weight of the grayscale statistical features is also increased. A specific numerical adjustment scheme is as follows: =0.2, =0.6, =0.2. This adjustment makes the calculation of the self-consistent decay index more sensitive to mineralization processes that increase density.

[0118] Examples of weight adjustments under composite rules, such as:

[0119] When the lesion is located at the lobar level (condition A) and calcification is present (condition B), the system can first apply the anatomical level rules to obtain the basic weights. =0.5, =0.25, =0.25); then, image representation rules are applied to significantly increase the weight of grayscale features while maintaining the relative advantage of morphological feature weights. A feasible merging scheme is as follows: =0.4, =0.5, =0.1. This weight reflects both the focus on macroscopic morphological changes (due to its location in the lung lobe) and the focus on density-specific changes (due to the presence of calcification).

[0120] The adjusted weights mentioned above ( , , Substitute this into the formula for constructing the weighted eigenvector. In, and further used to calculate weighted similarity. and the final self-consistency decay index .

[0121] It should be noted that step S4 is only calculated for lesions that are successfully matched through topological constraints, ensuring that the data basis for subsequent analysis has anatomical rationality. Through the weight adaptive adjustment mechanism, the contribution of different features in the similarity measurement is dynamically optimized, so that the self-consistency decay index calculated in the end is not just a mathematical similarity change rate, but a quantitative stability measure that deeply integrates medical prior knowledge and can sensitively and specifically reflect the evolution of the pathophysiological state inside the lesion over time. This provides a solid, reliable and interpretable time-dimensional quantitative input for the final comprehensive assessment of lesion activity.

[0122] S5: Calculate the changes in topological node identifiers of suspected lesion areas at different time points to obtain topological evolution indicators characterizing the spatial invasiveness of lesions.

[0123] Specifically, the topological evolution index is quantified using the following formula:

[0124] ;

[0125] in, A coefficient characterizing whether lesions change between lung lobe levels. A coefficient characterizing whether lesions change between lung segmental levels. This is a distance coefficient, whose value is calculated based on the shortest topological path length of the lesion center point in the bronchial tree topology and obtained after normalization. , , Preset weights related to the migration direction, and satisfying < < This is to demonstrate that migration to the distal bronchi carries a higher risk of spatial invasion.

[0126] Among them, setting < < The medical basis for this is as follows: In the course of pulmonary tuberculosis, the spread of lesions along the same bronchial tree to the distal (peripheral) (corresponding to a high B value) is a typical manifestation of high activity and strong invasiveness, and carries the highest risk; the spread across lung segments within the same lung lobe (corresponding to S=1) is the next most common; although "skipping" across lung lobes (corresponding to L=1) may occur, it is sometimes related to lymphatic drainage or hematogenous dissemination, and its specificity as a direct invasiveness indicator is relatively low.

[0127] In a preferred embodiment, one available weight is set to = 0.2, = 0.3, = 0.5. This setting ensures that the bronchial migration distance (B) contributes the most to the final metric.

[0128] In one specific implementation, the lobar level variation coefficient This is a binary variable representing whether a lesion skips across lobes. It is determined by comparing the lobar identifiers of the lesion at the initial and final time points; if they are the same, then it is set as... If they are different, then assume .

[0129] In one specific implementation, the lung segment level variation coefficient This is a binary variable representing whether a lesion has migrated across lung segments within the same lung lobe. It is determined by comparing the lung segment identifiers of the lesion at the initial and final time points; if they are the same, then it is set as... If they are different, then assume .

[0130] In one specific implementation, the distance coefficient The value calculation steps are as follows:

[0131] Determine endpoints: Obtain the lesion at the initial time point. and final time point The finest-grained topological node mapped (usually a tertiary node, i.e., a bronchial bifurcation point). Let its node IDs be... and .

[0132] Calculate the topological path length: In the bronchial tree topology (three-level node connection graph) constructed in step S2 above, calculate... and The number of edges on the shortest path between two bifurcations (i.e., the number of bronchial segments traversed to move from one bifurcation point to another). Denoteed as... .

[0133] Normalization: To eliminate differences in individual bronchial tree size, the original path length was normalized to... Interval. A preferred implementation is to divide it by a preset maximum representative path length for the whole lung. (For example, the maximum depth or diameter of the bronchial tree obtained through statistics on a sample dataset). Then the normalized distance coefficient... for: , The closer the value is to 1, the farther the bronchial tree has migrated.

[0134] The above calculations , , Values ​​and presets , , Substitute the weights into the formula: The topological evolution index was calculated. Value. Due to , ,and + + =1 (weights are usually normalized), therefore .when A value closer to 0 indicates a highly stable lesion location with no signs of spatial invasion; when... The closer the value is to 1, the more significant the lesion has migrated over a long distance to the distal bronchus, indicating an extremely high risk of spatial invasion.

[0135] It should be noted that step S5 decouples and quantifies the complex spatial changes of the lesion into two computable dimensions: cross-level jumps (L, S) and continuous migration along the channel (B), and assigns differentiated risk weights to migrations in different directions. < < A spatial invasiveness quantitative model conforming to the dissemination pattern of pulmonary tuberculosis was constructed. This topological evolution index The self-consistent decay index calculated in step S4 They complement each other and together provide objective and quantitative spatial dimensions for the final comprehensive assessment of activities.

[0136] S6: Acquire and integrate self-consistency decay indicators, topological evolution indicators, and patient-related clinical information to generate and output a lesion activity auxiliary analysis report to assist clinical decision-making.

[0137] Specifically, the fusion and generation steps achieve quantitative scoring of lesion activity through the following evaluation functions:

[0138] ;or

[0139] ;

[0140] This assessment function is used to calculate the final quantitative score of lesion activity. This demonstrates that temporal instability is fundamental, spatial invasiveness amplifies the risks, and clinical status... The assessment logic for individualized correction. Furthermore, this concept is implemented through two different mathematical forms, demonstrating its universality.

[0141] First formula This reflects a linear amplification relationship. Its advantages are simple calculation and intuitive understanding. It directly represents the slope of the amplification. When When the value is not too large, this linear relationship is reasonable and stable.

[0142] The second formula This reflects an exponential amplification relationship. Its advantage is that it is useful when considering spatial invasiveness indicators. At higher levels, this can produce a stronger risk amplification effect, which is more in line with the clinical understanding that "once bronchial dissemination occurs, the risk increases sharply." The rate of exponential growth was controlled.

[0143] Those skilled in the art can choose a function form that better reflects the actual risk quantification relationship based on the analysis of historical data or the results of clinical validation.

[0144] In the above, It is a self-consistent decay index; This is a topological evolution indicator.

[0145] and A modulation factor greater than zero, used for amplification. Impact on the final score. Its value is related to the specific level at which the migration occurred in step S5, reflecting the risk differences between different levels of migration. For example:

[0146] If migration only occurs at the lobar level ( ),but or ;

[0147] If migration occurs at the lung segment level ( ),but or ;

[0148] If the migration involves the bronchial level ( >0), then or .

[0149] The clinical information vector is a multidimensional vector used to characterize the non-radiographic patient status associated with assessment of pulmonary tuberculosis activity. Its data can be automatically extracted from hospital information systems or electronic medical records and typically includes: treatment status (e.g., untreated, under treatment, treatment completed), duration of illness, laboratory test results, symptom scores, etc.

[0150] This is a clinical information correction function used to adjust assessment results based on the patient's specific clinical condition, i.e., to fine-tune the score. Correction function The expression is:

[0151] ;

[0152] in, This is an indicator function (1 if treatment is in progress, 0 otherwise). Similarly (phlegm yin is 1, otherwise it is 0).

[0153] coefficient and A small negative moderating value (e.g., -0.1) indicates that treatment or sputum smear clearance will slightly reduce the activity score, reflecting a positive effect of treatment. (Coefficient) and Systematic biases in activity scores across different clinical states can be observed on historical datasets, based on imaging indicators, and calibrated using simple linear regression or expert experience. For example, if data shows that patients undergoing treatment exhibit an average systematic downward trend in activity scores of approximately 10%, then a threshold can be set. .

[0154] It should be noted that the beneficial effect of the above quantitative scoring formula is that it uses an interpretable mathematical model to represent the intrinsic temporal instability of imaging indicators ( ) of lesions. Topological indices characterizing its spatial invasion behavior ) and individualized clinical status information of patients ( By organically integrating these methods, a comprehensive score that combines objectivity, quantification, pathological mechanism correlation, and clinical operability is ultimately output, thus realizing a paradigm shift in the assessment of pulmonary tuberculosis lesion activity from single image interpretation to multi-dimensional, quantifiable, and intelligent auxiliary evaluation.

[0155] Specifically, the generation of the lesion activity auxiliary analysis report also includes:

[0156] Based on quantification of lesion activity score It compares the data with multiple preset risk reference intervals and automatically extracts medical knowledge items associated with the interval in which the score is located from a preset clinical medical knowledge base.

[0157] Quantitative scoring of lesion activity The lesion evolution information represented by the self-consistency decay index and the topological evolution index, as well as the associated medical knowledge items, are integrated to generate a structured report text and a visual summary.

[0158] In one specific implementation, for ease of clinical interpretation, a preset can be made. The scoring range and activity level, for example: stable: Light activity: High activity: .

[0159] In one specific implementation, medical knowledge item extraction includes: organizing a pre-built clinical medical knowledge base using scoring intervals as key indexes. When the calculated... After scoring, the system automatically determines the interval to which the lesion belongs and uses the interval value (e.g., highly active) as the query key to retrieve pre-associated, standardized text entries from the knowledge base. For example, for a highly active interval, the retrieved knowledge entry might be: "Indicates high lesion activity; clinicians are advised to make a comprehensive judgment based on biological evidence such as sputum culture and molecular testing, and to assess the adequacy of the current treatment plan."

[0160] In one specific implementation, the structured report text and visual summary are automatically integrated by the system. The system has a pre-built report generation template that defines the layout and placeholders for sections such as core conclusions, key evidence, evolutionary summaries, supporting recommendations, and clinical data snapshots. The integration process and content are as follows:

[0161] Key conclusion: The calculated quantitative score of lesion activity The corresponding classification (such as high activity level) is automatically filled into the template.

[0162] Key evidence:

[0163] Self-consistency decay curve: with the follow-up time point as the horizontal axis and the weighted similarity calculated in step S4 as the vertical axis. Using the vertical axis, generate a line chart to visually display the trend of feature consistency decay over time.

[0164] Lesion topology migration path diagram: On the schematic diagram of the lung hierarchical topology model constructed in step S2, the topology node identifiers of the lesion at different time points (such as RUL_S1→ RUL_S1_B2) are superimposed as highlighted nodes and connecting lines to clearly show its spatial migration path in the anatomical structure.

[0165] Evolution Summary: Based on the self-consistency decay index, topological evolution index, and comparison of image features at each time point, the system automatically generates a structured text summary, such as: the lesion volume increased by about 15% during the follow-up period, the average density decreased, the internal texture heterogeneity increased, and it migrated from the apical segment of the right upper lobe (RUL_S1) to the distal bronchus (RUL_S1_B2).

[0166] Additional suggestion: Fill in this section with the standard entries retrieved from the clinical medical knowledge base.

[0167] Clinical data snapshot: to be used to calculate the correction function Clinical information vector Key fields (such as currently under treatment, recent sputum smear: negative) are presented in a list format.

[0168] It should be noted that the above report generation process is based on the calculated quantitative score of lesion activity. Automatic triggering. The system calls the report engine to automatically populate and render the scoring results, various intermediate indicators, visual charts and data, and knowledge base retrieval results according to the rules of the preset template. Finally, it outputs a complete structured document, such as a PDF, containing quantitative scores, qualitative analysis, visual evidence and clinical recommendations, completing the fully automatic conversion from image analysis to decision support information.

[0169] It should be noted that step S6 uses a lesion activity quantification scoring function to combine multiple independent quantitative indicators generated in the previous steps ( , ) and patient-specific information ( An organic and interpretable mathematical fusion was performed to generate a comprehensive score that is both objective and clinically relevant. Furthermore, this abstract score can be transformed into a structured clinical report containing decision support information.

[0170] In summary, this invention constructs a hierarchical topological model of the lung anatomy and assigns a unique topological node identifier to suspected lesion areas. Using this topological node identifier as a constraint, a cross-temporal correspondence is established only when lesions are in the same or a pre-defined topological neighborhood. This fundamentally reduces the risk of mismatches caused by respiratory displacement, scanning differences, and local deformation, significantly improving the accuracy and reliability of lesion tracking in multi-temporal CT image follow-up analysis. Furthermore, this invention calculates a self-consistent decay index characterizing the temporal stability of lesions and a topological evolution index characterizing their spatial invasiveness, respectively, and integrates these two with patient clinical information to achieve a simultaneous and quantitative characterization of the temporal evolution stability and spatial anatomical invasiveness of pulmonary tuberculosis lesion activity. This overcomes the limitation of existing methods that rely on a single assessment dimension, providing a more comprehensive and objective quantitative assessment basis for clinical practice. This invention automatically generates an auxiliary analysis report that integrates quantitative scores, evolution information, visualization maps, and structured suggestions by linking the quantitative lesion activity score generated after fusion assessment with a pre-built clinical medical knowledge base. It can also be integrated with hospital information systems through standardized interfaces, realizing full-process automation and intelligence from image analysis to clinical decision support. While improving assessment efficiency and consistency, it also enhances the interpretability of results and the fit with clinical diagnosis and treatment processes.

[0171] Example 2, as Figure 3 This is a second embodiment of the present invention, which differs from the first embodiment in that the method further includes the step of report output and system integration.

[0172] The generated lesion activity auxiliary analysis report, which includes structured report text and visual summaries, is automatically packaged into a standardized clinical data exchange format package;

[0173] Through the medical information system integration interface, standardized data packets are automatically pushed and integrated into the patient's electronic medical record system. At the same time, according to preset rules, a system notification message containing key conclusions of lesion activity quantitative scoring is sent to the attending physician's workstation.

[0174] In one specific implementation, the data is automatically packaged into a standardized clinical data exchange format package, specifically including:

[0175] Format standard selection: Encapsulate the format according to the HL7 Clinical Documentation Architecture (CDA) standard or the Diagnostic Report resource specification in FHIR. These two are widely supported interoperability standards in current healthcare information systems.

[0176] Content mapping: Maps the core components of the report to corresponding fields in a standard format.

[0177] Quantitative scoring And the activity level is mapped to the conclusion section of DiagnosticReport.conclusionCode or CDA.

[0178] The structured report text is mapped to the text content of DiagnosticReport.presentedForm or CDA.

[0179] The visual summary is presented as a chart when generating a PDF report. When packaged into a standard data package, the corresponding image data is embedded into the multimedia entity section of the document in Base64 encoding.

[0180] Metadata such as patient identifier, examination time, and attending physician is populated in the document header.

[0181] In one specific implementation, according to preset rules, a system notification message containing key conclusions of the quantitative score of lesion activity is sent to the attending physician's workstation, including:

[0182] Triggering rules: The rules are based on the quantitative score of lesion activity. And its level settings. For example, it can be preset: when A notification is triggered immediately when the value is ≥0.6 (i.e., high activity); when... (For light activity) it only triggers on the first occurrence.

[0183] Message Generation: The system automatically generates a concise notification message with a template such as:

Early Warning of Active Tuberculosis Lesions

[0184] Sending channels: The message is pushed to the attending physician's workstation portal message center and hospital office system via the hospital's internal communication middleware, or sent to their linked mobile phone via a secure SMS gateway. The notification may include a deep link that directly jumps to the report details page.

[0185] It should be noted that, through the above-described implementation methods, the output of this invention is no longer an isolated file or interface, but rather becomes an active data node within the hospital's information ecosystem. It automates the entire process of evaluation results, from generation and standardization to transmission and clinical reach, ensuring that critical diagnostic and treatment information is delivered to the front lines of clinical care in a standardized, secure, and timely manner. This plays a supporting role in shortening decision-making cycles and improving diagnostic and treatment efficiency.

[0186] Example 3 is the third embodiment of the present invention. This embodiment provides an intelligent detection and activity assessment system for pulmonary tuberculosis lesions on chest CT images, applied to the above-mentioned intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images, including:

[0187] The image data acquisition and processing module is configured to acquire chest CT image sequences of the same pulmonary tuberculosis patient at at least two follow-up time points, and to register and detect lesions in the sequences, and output the suspected lesion areas at each time point.

[0188] The topology mapping and identification module is configured to construct a spatial topology mapping relationship between suspected lesion areas and the lung lobe, lung segment and bronchial bifurcation levels based on the lung anatomy segmentation results, and assign a unique topology node identifier to each suspected lesion area.

[0189] The temporal correspondence establishment module is configured to establish cross-time lesion temporal correspondences that satisfy topological consistency constraints based on topological node identifiers;

[0190] The indicator calculation module is configured to calculate a self-consistent decay index representing the temporal stability of lesions based on the temporal correspondence, and to calculate a topological evolution index representing spatial invasiveness based on the change of topological node identifiers.

[0191] The intelligent report generation and decision support module is configured to: integrate self-consistency decay indicators, topological evolution indicators and patient clinical information, and generate a structured auxiliary analysis report on lesion activity based on the quantitative score of lesion activity and related clinical medical knowledge.

[0192] A clinical system integration interface, configured to output auxiliary analysis reports to the clinical diagnosis and treatment environment in a standardized format.

[0193] It should be noted that the above system can be deployed as a software service in a hospital's private medical cloud or data center. Each module can be deployed on the same server or distributed through a microservice architecture, enabling high-performance data communication via the internal network. The system securely and in a standardized manner interfaces with the hospital's existing PACS, HIS, EMR, and other systems through the image data acquisition and processing module and clinical system integration interface, forming an intelligent auxiliary analysis node embedded in the clinical workflow.

[0194] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent detection and activity assessment of pulmonary tuberculosis chest CT image lesions, characterized in that, Includes the following steps: Acquire chest CT image sequences of the same pulmonary tuberculosis patient at at least two follow-up time points, and perform registration and lesion detection on the chest CT image sequences to obtain suspected lesion areas at each time point; Based on the lung anatomy segmentation results of the chest CT images, a spatial topological mapping relationship is constructed between the suspected lesion area and the lung lobe, lung segment and bronchial bifurcation levels, and a unique topological node identifier is assigned to each suspected lesion area. Based on the topological node identifiers, a cross-time temporal correspondence relationship of lesions is established that satisfies the topological consistency constraint; Based on the aforementioned temporal correspondence, a self-consistent decay index characterizing the temporal stability of lesions is calculated. The changes in topological node identifiers of the suspected lesion area at different time points are calculated to obtain a topological evolution index characterizing the spatial invasiveness of the lesion. The self-consistency decay index, the topological evolution index, and the patient's associated clinical information are acquired and fused to generate and output a lesion activity auxiliary analysis report to assist clinical decision-making.

2. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 1, characterized in that: The spatial topological mapping relationship is achieved by constructing a hierarchical topological model of lung anatomy that includes the lobe level, lung segment level, and bronchial bifurcation level. Each suspected lesion area is mapped and assigned to a unique topological node in the hierarchical topology model according to its spatial location.

3. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 1, characterized in that: The establishment of the cross-temporal lesion temporal correspondence relationship that satisfies the topological consistency constraint includes: Construct topology constraint rules based on the topology node identifiers; According to the topology constraint rules, a candidate matching object is determined only when suspected lesion areas at different time points are assigned to the same topology node or to a topology node with a preset adjacency relationship. For the candidate matching objects, their morphological features, grayscale statistical features, and texture distribution features are extracted, and cross-time matching is performed based on the extracted features to establish a temporal correspondence.

4. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 3, characterized in that: The calculation of the self-consistency decay index is only performed on suspected lesion areas that have successfully established a temporal correspondence through the topological constraint rules; The similarity is quantified by calculating the changes in the weighted multidimensional image feature vectors of the suspected lesion area at different time points.

5. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 4, characterized in that: When calculating the self-consistency decay index, different weight coefficients are assigned to morphological features, gray-scale statistical features, and texture distribution features, and the weight coefficients are adaptively adjusted according to the anatomical level of the suspected lesion area and the follow-up imaging performance. The adaptive adjustment of the weighting coefficients follows these rules: When the suspected lesion area is located at the lung segment or bronchial bifurcation level, the weighting coefficient of the texture distribution feature is increased; When the suspected lesion area is located at the lobar level, the weighting coefficient of the morphological feature is increased; When cavitation or calcification is observed in follow-up images, the weighting coefficient of the gray-scale statistical features of the corresponding suspected lesion area should be increased.

6. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 1, characterized in that: The topology evolution metric is quantified using the following formula: ； in, A coefficient characterizing whether lesions change between lung lobe levels. A coefficient characterizing whether lesions change between lung segmental levels. This is a distance coefficient, the value of which is calculated based on the path length of the lesion center point moving in the bronchial bifurcation hierarchy topology and obtained after normalization. , , Preset weights related to the migration direction, and satisfying < < This is to demonstrate that migration to the distal bronchi carries a higher risk of spatial invasion.

7. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 6, characterized in that: The fusion and generation steps achieve quantitative scoring of lesion activity through the following evaluation function: ;or ； in, It is a self-consistent decay index; As a topological evolution indicator; For clinical information vectors; This is a clinical information correction function used to adjust assessment results based on the patient's specific clinical condition; and It is a modulation factor greater than zero, which is related to the specific level at which the topology shift occurs.

8. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 7, characterized in that: The generation of the auxiliary analysis report on lesion activity also includes: Based on the quantitative score of lesion activity The system compares the scores with multiple preset risk reference intervals and automatically extracts medical knowledge entries associated with the intervals in which the scores are located from a preset clinical medical knowledge base. Quantitative scoring of the lesion activity The self-consistency decay index and the lesion evolution information represented by the topological evolution index, as well as the associated medical knowledge items, are integrated to generate a structured report text and a visual summary.

9. The intelligent detection and activity assessment method for pulmonary tuberculosis lesions on chest CT images as described in claim 8, characterized in that: The method also includes steps for report output and system integration: The generated lesion activity auxiliary analysis report, which includes the structured report text and visual summary, is automatically packaged into a standardized clinical data exchange format package; Through the medical information system integration interface, the standardized data packet is automatically pushed and integrated into the patient's electronic medical record system. At the same time, according to preset rules, a system notification message containing key conclusions of lesion activity quantitative scoring is sent to the attending physician's workstation.

10. A system for intelligent detection and activity assessment of pulmonary tuberculosis lesions on chest CT images, applied to the method for intelligent detection and activity assessment of pulmonary tuberculosis lesions on chest CT images as described in any one of claims 1-9, characterized in that, include: The image data acquisition and processing module is configured to acquire chest CT image sequences of the same pulmonary tuberculosis patient at at least two follow-up time points, and to register and detect lesions in the sequences, and output the suspected lesion areas at each time point. The topology mapping and identification module is configured to construct a spatial topology mapping relationship between the suspected lesion area and the lung lobe, lung segment and bronchial bifurcation levels based on the lung anatomical structure segmentation results, and assign a unique topology node identifier to each suspected lesion area. The time-series correspondence establishment module is configured to establish a cross-time lesion time-series correspondence that satisfies topological consistency constraints based on the topological node identifier; The indicator calculation module is configured to calculate a self-consistent decay index representing the temporal stability of lesions based on the time-series correspondence, and to calculate a topological evolution index representing spatial invasiveness based on changes in topological node identifiers. The intelligent report generation and decision support module is configured to: integrate the self-consistency decay index, the topological evolution index and the patient's clinical information, and generate a structured lesion activity auxiliary analysis report based on the lesion activity quantitative score and associated clinical medical knowledge; A clinical system integration interface is configured to output the auxiliary analysis report to the clinical diagnosis and treatment environment in a standardized format.

Images