A region of interest evaluation method and apparatus

Abstract

This application provides a method and apparatus for region of interest (ROI) assessment, relating to the field of medical imaging technology. The method includes: acquiring chest images of a user at different times; performing the following for each chest image: extracting a lung region from the chest image, identifying anatomical landmarks in the lung region, determining the relative positions of different layers in the chest image based on the anatomical landmarks; determining the mapping relationship between different layers in different chest images based on the relative positions of the layers in each chest image; identifying a region of interest (ROI) within the lung region of each chest image; and generating a ROI assessment result based on the layer in which the ROI is located in the chest image and the mapping relationship between different layers in different chest images. This application can improve the accuracy of the assessment results.

Description

Technical Field

[0001] This application relates to the field of medical imaging technology, and in particular to a method and apparatus for evaluating regions of interest. Background Technology

[0002] Throughout the diagnosis and treatment of lung diseases, comparing chest images of users at different times (such as before and after treatment, and follow-up intervals) allows for dynamic observation and evaluation of changes in areas of interest such as lung lesions, nodules, and inflammatory areas. This is a key means of judging disease progression, evaluating treatment effectiveness, and guiding clinical decision-making.

[0003] Currently, clinical assessment of regions of interest across different time periods in chest imaging relies primarily on manual interpretation by radiologists. This involves visually comparing anatomical structures across images from different time periods, manually locating and confirming the same region of interest, and subjectively judging changes in its size, shape, and density to form the assessment result.

[0004] However, when users receive chest imaging at different times, the spatial location and slice index of the same anatomical structure often shift significantly in different imaging phases due to differences in respiratory depth, body positioning, and other imaging conditions. For example, the same lung nodule located on slice 120 in baseline CT may shift to slice 135 in follow-up CT. Existing methods rely on manual comparison and localization by physicians, which is not only time-consuming, labor-intensive, and inefficient, but also makes it difficult to guarantee the accuracy and repeatability of localization. The localization results often differ significantly between different physicians and at different times of image review, directly affecting the accuracy of subsequent assessments. Summary of the Invention

[0005] This application is made in view of at least one of the above-mentioned technical problems existing in the prior art, and the application can improve the accuracy of the evaluation results.

[0006] In a first aspect, embodiments of this application provide a method for evaluating a region of interest, including:

[0007] Obtain chest images of the user at different times; For each of the chest images, the following steps are performed: extracting the lung region from the chest image, identifying anatomical landmarks in the lung region, and determining the relative positions of each layer in the chest image based on the anatomical landmarks; Based on the relative positions of each layer in each of the aforementioned chest images, the mapping relationship between the layers of different chest images is determined; Identify the region of interest in the lung region of each of the chest images; Based on the layer in which the region of interest is located in the chest image and the mapping relationship between different layers of the chest image, a region of interest evaluation result is generated.

[0008] Secondly, embodiments of this application provide a region of interest evaluation apparatus, comprising: The acquisition module is configured to acquire chest images of the user at different times; The determination module is configured to perform the following for each of the chest images: extracting the lung region from the chest image, identifying anatomical landmarks in the lung region, and determining the relative positions of each layer in the chest image based on the anatomical landmarks; The generation module is configured to determine the mapping relationship between different layers of the chest images based on the relative positions of each layer in each chest image; identify the region of interest in the lung region of each chest image; and generate a region of interest evaluation result based on the layer in the chest image where the region of interest is located and the mapping relationship between different layers of the chest images.

[0009] This application provides a method and apparatus for region of interest (ROI) assessment. By using anatomical landmark localization and establishing mapping relationships, it converts absolute positions into relative positions, eliminating spatial inconsistencies caused by differences in scanning conditions. This makes images from different time periods comparable and improves the accuracy of assessment results. From acquiring multi-phase images to generating the final assessment result, no manual intervention is required. This solves the problems of reliance on manual labor, low efficiency, and high subjectivity in multi-phase lung cancer lesion follow-up assessment in clinical practice, making it particularly suitable for large-scale clinical follow-up and tumor treatment efficacy monitoring scenarios. Based on clear layer mapping relationships and ROI level localization, all assessment indicators, such as volume changes and density changes, are supported by quantified spatial benchmarks, avoiding assessment errors caused by layer mismatch. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating a region of interest assessment method provided in one embodiment of this application; Figure 2 This is a schematic diagram of a region of interest evaluation device provided in one embodiment of this application. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions of the embodiments of this application, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] like Figure 1 As shown in the figure, this application provides a method for evaluating regions of interest, including: Step 101: Obtain chest images of the user at different times.

[0014] This program acquires chest computed tomographic images (CT scans, MRI scans, etc.) of the same user at at least two different time points. The images are three-dimensional volumetric data of multi-slice 2D images. Images from different time points may exhibit differences in scanning parameters, such as slice thickness, slice spacing, scanning equipment, tube voltage / current, and patient status, such as respiratory depth, body position, and posture. The acquired images include metadata such as slice number, voxel spacing, and HU value calibration information.

[0015] Step 102: Perform the following for each chest image: Extract the lung region from the chest image, identify anatomical landmarks in the lung region, and determine the relative positions of each layer in the chest image based on the anatomical landmarks.

[0016] From chest images, the lung region, i.e., the lung parenchyma (including both lungs), is extracted, and non-lung tissues such as the thorax, sternum, heart, major blood vessels, tracheal wall, and pleura are removed. The lung region is extracted based on HU value thresholding and morphological processing.

[0017] Specifically, the HU value threshold range is set to [-1000, -400]. The chest image is binarized and segmented to obtain an initial lung mask. Closure operations are used to fill in small cavities within the lungs, and opening operations are used to remove isolated small connected components from the mask, such as sections of the trachea and small blood vessels. The minimum connected component size is ≥5000 voxels. Finally, the left and right lung parenchyma are separated using region growing to obtain separate masks for each lung. The mask is a binary image, with lung parenchyma regions represented by 1 and non-lung regions by 0. The mask corresponds one-to-one with the layer number and voxel position of the original chest image.

[0018] This embodiment eliminates interference from irrelevant tissues, reducing the computational load for subsequent anatomical landmark identification and region of interest identification.

[0019] Anatomical landmarks are feature points within the lung region that are anatomically unique, spatially stable, and easily automatically identifiable. Anatomical structures with variable locations and ambiguous boundaries are excluded. The lung apex and lung base are preferred as anatomical landmarks. The tracheal bifurcation point and the center of the left and right hilum can also be used as anatomical landmarks.

[0020] Lung apex: In the lung mask, find the layer with lung parenchyma voxels at the top along the human body's longitudinal axis (z-axis). The centroid of the lung parenchyma at this layer is the coordinate of the lung apex.

[0021] Lung base: In the lung mask, find the layer with lung parenchyma voxels at the bottom along the human body's longitudinal axis (z-axis). The centroid of the lung parenchyma at this layer is the coordinate of the lung base.

[0022] This embodiment uses the lung apex and lung base as anatomical landmarks, leveraging the stability of their anatomical positions to establish a unified anatomical reference coordinate system for multi-phase imaging. This addresses the problems of traditional registration methods, such as the lack of a fixed benchmark and susceptibility to patient positioning and respiration. This embodiment employs an automated identification method based on lung masks to replace manual annotation of anatomical landmarks, achieving automated and standardized landmark identification and avoiding the subjectivity and repetitive errors of manual annotation.

[0023] The absolute numbers of each layer of the chest image are converted into relative positions within the interval [0,1] to standardize the anatomical position. position [i] = (i - Lung apex ) / (Lung base - Lung apex ), where i is the absolute number of the i-th layer of the chest image, Lung apex Number the layer containing the lung apex, Lung base Let i be the layer number of the lung base, and position[i] be the relative position of the i-th layer.

[0024] By converting the strata of multi-phase images from anatomically insignificant numerical numbers to anatomically significant relative positions, the problem of inconsistent absolute layer numbers for the same anatomical location in multi-phase images is solved in traditional registration methods. For example, the hilum may be layer 120 in one phase and layer 135 in another. The relative positions are limited to the [0,1] interval, achieving a unified standard for position measurement across multi-phase images and avoiding positional comparison failures caused by differences in the total number of image layers.

[0025] Step 103: Based on the relative positions of each layer in each chest image, determine the mapping relationship between the layers of different chest images.

[0026] The mapping relationship can include a forward mapping from the first chest imaging layer number to the second chest imaging layer number, as well as a reverse mapping from the second chest imaging layer number to the first chest imaging layer number, to accommodate the bidirectional query requirements for subsequent cross-period tracking of regions of interest. The mapping relationship can be stored in the form of key-value pairs.

[0027] Step 104: Identify the region of interest in the lung region of each chest image.

[0028] Regions of interest (ROIs) can be areas of interest such as lung cancer lesions or lung nodules.

[0029] Specifically, Regions of Interest (ROIs) can be identified by combining the density, morphological, and spatial characteristics of lung tissue. For example, within a lung mask, voxel regions with HU values ​​deviating from normal lung tissue ([-1000, -400]) can be screened, such as ground-glass nodules [-700, -400] and solid nodules >-400. Morphological analysis of these density-abnormal regions can be performed to screen for areas with nodular morphology, excluding sheet-like or cord-like normal lung textures or inflammation. The three-dimensional boundaries, centroid, volume, and other morphological features of the regions can be extracted. False positive regions that are too small (e.g., <5mm³), have blurred boundaries, or are connected to blood vessels / trachea are removed to obtain the final ROI.

[0030] Step 105: Generate region of interest assessment results based on the layer in which the region of interest is located in the chest image and the mapping relationship between different layers of the chest image.

[0031] For multi-ROI scenarios, a global optimal matching algorithm, such as the Hungarian algorithm, is used to achieve a one-to-one correspondence between multiple ROIs, while marking newly added ROIs (those that only exist in later images) and disappeared ROIs (those that only exist in earlier images).

[0032] For the same tracked ROI, a multi-dimensional quantitative comparison is performed based on its morphology, density, and spatial characteristics across multiple periods to obtain the region of interest assessment results. Specific dimensions may include: Spatial changes: changes in the coverage area and shifts in the center of gravity; Changes in morphological characteristics: rate of volume change, rate of change of longest diameter / shortest diameter, and changes in sphericity / edge regularity; Changes in density characteristics: changes in average HU value and changes in the proportion of solid / ground glass components.

[0033] This application's embodiments utilize a technical path of anatomical landmark localization followed by the establishment of mapping relationships. This transforms absolute positions into relative positions, eliminating spatial inconsistencies caused by differences in scanning conditions, making images from different periods comparable, and improving the accuracy of evaluation results. From acquiring multi-phase images to generating the final evaluation result, no manual intervention is required. This solves the problems of reliance on manual labor, low efficiency, and high subjectivity in multi-phase lung cancer lesion follow-up evaluation in clinical practice, making it particularly suitable for large-scale clinical follow-up and tumor treatment efficacy monitoring scenarios. Based on clear layer mapping relationships and region-of-interest level localization, all evaluation indicators, such as volume changes and density changes, are supported by quantified spatial benchmarks, avoiding evaluation errors caused by layer mismatch.

[0034] In one embodiment of this application, identifying anatomical landmarks in the lung region includes: The lung region is divided into the left lung and the right lung; Identify the apex and base of the left lung and the apex and base of the right lung, respectively.

[0035] Using the mediastinal midline as the core dividing line, which is the natural anatomical boundary between the left and right lungs in chest imaging without anatomical ambiguity, and can be precisely located using the voxel coordinate system of chest images, the relative positions of each layer within the left and right lung regions are calculated based on the left lung apex-left lung base and right lung apex-right lung base, respectively.

[0036] The left and right lungs of the human body have natural differences in anatomical shape. If the left and right lungs are identified as a whole, the apex and base of the lungs are prone to the problem that the anatomical landmarks are biased towards the larger lung lobes. Lung identification completely avoids this bias, so that the anatomical landmarks of each lung are accurately matched with its own anatomical features.

[0037] In one embodiment of this application, determining the mapping relationship between layers of different chest images based on the relative positions of layers in various chest images includes: Based on the relative positions of each layer in the first chest image and the relative positions of each layer in the second chest image, a first mapping relationship is determined between each layer in the first chest image and each layer in the second chest image. For each layer in the first chest image, the following steps are performed: Based on the first mapping relationship, in each layer of the second chest image, determine multiple candidate layers corresponding to the current layer in the first chest image; determine the affine transformation matrix corresponding to each candidate layer; perform spatial transformation on the candidate layers based on the affine transformation matrix; calculate the similarity between the transformed candidate layers and the current layer; and based on the similarity between the current layer and each transformed candidate layer, determine the second mapping relationship between each layer in the first chest image and each layer in the second chest image.

[0038] For each slice in the first chest image, among all valid lung slices in the second chest image, the slice with the smallest relative positional ratio deviation from the current slice is selected as the initial matching slice. The first chest image can be a baseline CT, and the second chest image can be a follow-up CT.

[0039] Centered on the initial matching layer of the current layer in the first mapping relationship, a candidate layer window with a fixed range is defined. The window size can be ±5 to 10 layers, that is, 5 to 10 layers above and below the initial matching layer. Candidate layers are selected only from the effective lung layers of the second chest image, and layers without lung parenchyma are removed to avoid errors caused by invalid layers participating in registration.

[0040] The effective regions of two-dimensional slices from the current layer of the first chest image and each candidate layer of the second chest image, i.e., the areas covered by the lung mask, are extracted. Feature point detection is performed only on the effective regions, eliminating interference from non-lung tissues. Based on SIFT (Scale-Invariant Feature Transform) / ORB (Oriented FAST and Rotated BRIEF), feature points are detected in the effective regions of the current layer slice and candidate layer slices respectively, obtaining the two-dimensional coordinates of the feature points and their corresponding feature descriptors. Based on Euclidean distance, similarity matching is performed on the two sets of feature descriptors, eliminating outliers with excessive matching errors and retaining inlier feature pairs. RANSAC (Random Sample Consensus) is used to solve the two-dimensional affine transformation matrix based on the filtered inlier feature pairs. This affine transformation matrix can correct for rigid spatial deformation between the two slices caused by patient position and scanning angle. The affine transformation matrix contains 6 degrees of freedom: translation, rotation, scaling, and shearing.

[0041] An affine transformation is performed on the effective regions of the two-dimensional slices of the candidate layer of the second chest image, while non-effective regions are excluded from the transformation to ensure computational efficiency. According to the affine transformation matrix of the corresponding candidate layer, coordinate mapping and grayscale interpolation are performed on each pixel of the candidate layer slice. The transformed candidate layer slice maintains the same pixel size and coordinate system as the current layer slice of the first chest image, achieving geometric alignment between the two slices and eliminating layer misalignment caused by rigid deformation.

[0042] The similarity between the transformed candidate layer and the current layer is calculated, focusing only on the intersection of the lung masks of the two slices, eliminating non-overlapping regions to ensure the validity of the results. Similarity is calculated based on normalized cross-correlation or mutual information.

[0043] Normalized cross-correlation: The calculated value ranges from [-1, 1]. The closer the value is to 1, the higher the similarity between the two slices.

[0044] Mutual information: The larger the calculated value, the stronger the correlation between the gray values ​​of the two slices and the higher the similarity.

[0045] Sort the similarity values ​​of all candidate layers corresponding to the current layer, and select the candidate layer with the highest similarity value as the exact matching layer for the current layer.

[0046] Set a minimum similarity threshold. If the highest similarity value is lower than the threshold, mark the layer as having no effective match to avoid subsequent errors caused by low-precision matching. Repeat the above registration steps for all valid lung layers of the first chest image to finally generate the second mapping relationship.

[0047] Coarse registration based solely on relative position can only achieve a rough match at the anatomical position level, and cannot correct rigid spatial deformations caused by patient rotation, translation, and scanning angle deviations. In this embodiment, however, the candidate layer is geometrically corrected through affine transformation, which completely eliminates the above-mentioned rigid deformation errors, enabling pixel-level geometric spatial alignment of the two slices and reducing pixel-level errors in layer registration.

[0048] In one embodiment of this application, determining the affine transformation matrix corresponding to each candidate layer includes: Execute for each candidate layer: Detect key points in the current layer and key points in candidate layers; Determine the similarity between each key point in the current layer and each key point in the candidate layer; Based on similarity, determine the matching relationship between key points in the current layer and key points in the candidate layer; Calculate the affine transformation matrix based on the matching relationship.

[0049] The key points to be detected are pixels in the slice that have abrupt changes in gray value, edge features, and corner features (such as lung texture intersections, lesion edge points, and interlobar fissure feature points). These points have spatial uniqueness and feature stability, and are the core carriers of feature matching in image registration.

[0050] Output keypoint sets for the current layer and candidate layers respectively. Each keypoint contains two-dimensional spatial coordinates and feature descriptors.

[0051] Based on the type of feature descriptor, an appropriate algorithm is selected to achieve quantified similarity scoring: For floating-point feature descriptors, Euclidean distance (the smaller the distance, the higher the similarity) or cosine similarity (the closer the value is to 1, the higher the similarity) is preferred.

[0052] If the feature descriptor is binary, the Hamming distance method is preferred (the smaller the distance, the higher the similarity).

[0053] For each keypoint in the current layer, a similarity calculation must be performed with all keypoints in the candidate layers. For a single keypoint in the current layer, the keypoint with the highest similarity value among all keypoints in the candidate layers is selected as the unique matching object. A similarity threshold is set (e.g., Euclidean distance ≤ preset value, cosine similarity ≥ 0.7, Hamming distance ≤ preset value). If the similarity value of a keypoint pair does not reach the threshold, the matching pair is directly discarded and marked as an invalid match. The final matching relationship is a one-to-one bidirectional matching pair, that is, a keypoint in the current layer corresponds to only one keypoint in the candidate layer, and a keypoint in the candidate layer corresponds to only one keypoint in the current layer.

[0054] This application embodiment achieves robust and accurate solving of the affine transformation matrix through key point detection → key point similarity calculation → key point matching → matrix solving, solving the problems of low matrix accuracy and poor spatial correction effect caused by chaotic feature point matching and external point interference in traditional transformation matrix solving.

[0055] In one embodiment of this application, a region of interest (ROI) evaluation result is generated based on the layer in a chest image where the ROI is located and the mapping relationship between different layers of different chest images, including: Based on the second mapping relationship, determine the reference layer in the second chest image corresponding to the target layer where the first region of interest is located in the first chest image; The search range is determined based on the reference layer; Identify the second region of interest within the search area; Calculate the overall similarity between the first region of interest and the second region of interest; Based on the comprehensive similarity between the first region of interest and the second region of interest, an evaluation result of the region of interest is generated.

[0056] The target layer of the first ROI is the layer containing its centroid or the layer with the largest cross-section in the first chest image. This is the core feature layer of the ROI, avoiding mapping bias caused by using non-core layers as a reference. Through the positive mapping of the second mapping relationship (first chest image → second chest image), the unique and precise layer corresponding to the target layer in the second chest image is directly queried; this layer is the reference layer. Using the reference layer as the central layer, combined with the conventional slice thickness of chest CT (0.625~1.25mm) and minor residual registration errors, a symmetrical three-dimensional layer search range is defined to avoid ROI omissions due to minor registration errors.

[0057] The optimal search range is ±3 to 5 layers, which can be adaptively adjusted based on clinical scanning parameters (the smaller the slice thickness, the smaller the range can be). This ensures coverage of the possible location of the first ROI in the second chest image while avoiding increased computation and false matches due to an excessively large search range. The search range not only includes the slice range but is also limited to the effective lung region in the second chest image within the same lobe as the first ROI. For example, if the first ROI is located in the left upper lobe, the search is limited to ±3 to 5 layers of the reference slice for the left upper lobe in the second chest image, further eliminating irrelevant areas and improving search efficiency. For each extracted second ROI, its core feature information is extracted, including 3D centroid coordinates, volume, average HU value, morphological feature vector, and the proportion of solid / ground-glass opacity, providing a data foundation for subsequent comprehensive similarity calculations.

[0058] Using the core feature information of the first ROI and the core feature information of each second ROI as the calculation objects, a multi-dimensional feature similarity metric calculation is performed to avoid misjudgment caused by single-dimensional matching.

[0059] Set a comprehensive similarity threshold and combine it with similarity ranking to achieve accurate matching: If only one second ROI has a comprehensive similarity greater than or equal to the threshold, it is determined that the same ROI has been successfully matched. If the combined similarity of multiple second ROIs is greater than or equal to the threshold, the second ROI with the highest similarity value is selected as the matching object; If the overall similarity of all second ROIs is less than the threshold, the first ROI is marked as vanished / completely alleviated. If a valid ROI is found outside the search area in the second chest image, it is marked as a new ROI / new metastasis.

[0060] For successfully matched interphase ROIs, quantitative change indicators are calculated based on their core characteristics, including volume change rate, diameter change rate, change in mean HU value, change in the proportion of solid components, and change in sphericity. These quantitative change indicators are then categorized by clinical diagnostic criteria (such as RECIST 1.1 criteria and lung nodule follow-up guidelines) to assess their clinical significance, such as partial remission, disease stability, disease progression, increase / decrease in solid components, and morphological regularization / irregularization.

[0061] By using local search, the search range of ROI is narrowed, avoiding the interference of false positive ROIs caused by searching the entire lung and all layers, improving the accuracy of ROI cross-period tracking, and significantly reducing the amount of computation.

[0062] In one embodiment of this application, calculating the comprehensive similarity between the first region of interest and the second region of interest includes: Based on the centroid of the first region of interest and the centroid of the second region of interest, calculate the spatial similarity between the first region of interest and the second region of interest; Based on the volumes of the first region of interest and the second region of interest, calculate the volume similarity between the first region of interest and the second region of interest; Obtain the shape feature vectors of the first region of interest and the second region of interest; Based on the morphological feature vectors of the first region of interest and the second region of interest, the morphological similarity between the first region of interest and the second region of interest is determined. The density similarity between the first region of interest and the second region of interest is determined based on the Henle unit values ​​of the first region of interest and the second region of interest. The overall similarity is calculated based on spatial similarity, volume similarity, morphological similarity, and density similarity.

[0063] Similarity calculation covers four core dimensions: spatial location, volume, morphology, and density, all of which are key clinical dimensions for assessing changes in lung ROI. In practical applications, the dimensions used can be adjusted based on specific needs.

[0064] Spatial location similarity: The centroid of the first ROI is mapped to the second chest image space, the Euclidean distance with the centroid of the second ROI is calculated, and the similarity value is normalized to 0~1 by the Gaussian kernel function; Volume similarity: Calculate the ratio of the volumes of two ROIs and use it directly as a similarity value of 0 to 1; Morphological similarity: Extract the morphological feature vectors (sphericity, elongation, flatness, compactness) of two ROIs, calculate the cosine similarity between the vectors, and the closer the value is to 1, the more similar the morphology. Density similarity: Combining the difference in average HU value and the difference in the proportion of solid / ground glass components (1 - absolute difference), the density similarity value is weighted and fused to a range of 0 to 1.

[0065] A weighted linear fusion strategy was adopted, with weights set based on clinical experience and the importance of each dimension to ROI matching. Spatial location was the core prior criterion (highest weight), volume was the core clinical assessment indicator (second highest weight), and morphology and density were auxiliary matching indicators (weights decreasing sequentially), while satisfying the weight normalization constraint (all weights sum to 1). The overall similarity was calculated as: w1 × spatial similarity + w2 × volume similarity + w3 × morphological similarity + w4 × density similarity, where w1 = 0.4, w2 = 0.3, w3 = 0.2, and w4 = 0.1.

[0066] This application's embodiments explicitly decompose the comprehensive similarity into four dimensions: space, volume, shape, and density. Standardized and unique calculation rules and formulas are established for each dimension, completely avoiding the problems of arbitrary dimensions and ambiguous rules in traditional calculations. This makes the similarity calculation highly standardized and reproducible, ensuring consistent results across different devices and operators. Through multi-dimensional feature fusion, the four core features of the ROI—spatial location, size, shape, and density—are taken into account, effectively preventing matching failures caused by changes in a single dimension feature.

[0067] In one embodiment of this application, a region of interest (ROI) evaluation result is generated based on the comprehensive similarity between a first ROI and a second ROI, including: A similarity matrix is ​​constructed based on the comprehensive similarity between multiple first regions of interest and multiple second regions of interest; the elements in the similarity matrix are used to represent the comprehensive similarity between the first regions of interest and the second regions of interest. Construct a cost matrix based on the similarity matrix; Based on the cost matrix, determine the matching relationship between the first region of interest and the second region of interest, and / or the unmatched first region of interest, and / or the unmatched second region of interest; If there is a matching relationship between the first region of interest and the second region of interest, a region of interest evaluation result is generated.

[0068] Suppose there are M first ROIs in the first chest image and N second ROIs in the second chest image. Then the similarity matrix is ​​an M×N two-dimensional matrix, where each row corresponds to a first ROI (numbered R1, R2, ..., R...). M The matrix columns correspond to the second ROI (numbered S1, S2, ..., S). N The element Sim(i,j) in the i-th row and j-th column of the similarity matrix represents the combined similarity between the i-th first ROI and the j-th second ROI.

[0069] The element Cost(i,j) in the i-th row and j-th column of the cost matrix is ​​1. Sim(i,j) is normalized through a simple linear transformation. The element values ​​of the cost matrix are ∈[0,1]. The closer the value is to 0, the lower the matching cost and the higher the matching degree between the two ROIs.

[0070] The cost matrix and similarity matrix are M×N matrices of the same dimension, with rows and columns corresponding one-to-one with the first ROI and the second ROI, respectively, ensuring that the mapping relationship between the matrices is unique and traceable.

[0071] Using the cost matrix as input, the algorithm follows the standardized steps of row reduction, column reduction, and finding augmenting paths in the Hungarian algorithm to find the globally optimal matching scheme that satisfies that each first ROI matches at most one second ROI and each second ROI matches at most one first ROI. The goal is to minimize the sum of costs for all matching pairs.

[0072] A dual judgment is made by combining a cost threshold (preferably Cost≤0.4, corresponding to a comprehensive similarity Sim≥0.6): If the cost value of the matching pair solved by the algorithm is less than or equal to the threshold, it is determined to be a valid matching relationship, that is, the first ROI and the second ROI in this group are the same ROI across different periods; If the cost of a matching pair obtained by the algorithm is greater than the threshold, it is determined to be an invalid match.

[0073] Based on valid matching relationships, ROIs that did not achieve valid matching are classified and labeled: A first ROI that does not form a valid match with any second ROI is marked as an unmatched first ROI, corresponding to the clinical scenario of lesion disappearance / complete remission; Secondary ROIs that do not form a valid match with any primary ROI are marked as unmatched secondary ROIs, corresponding to the clinical scenario of new lesions / new metastases / new nodules.

[0074] When a valid matching relationship exists, an independent evaluation result is generated for each successfully matched ROI pair; when no valid matching relationship exists, only a qualitative evaluation result is output for all outpatient lesions disappearing or all lesions becoming new.

[0075] For each successfully matched first and second ROI, an evaluation result is generated based on its multi-dimensional feature quantification change indicators. Feature indicators may include: volume change rate, average HU value change, solid component ratio change, sphericity / morphological feature change, spatial centroid offset distance, etc. All indicators are quantified values.

[0076] Ultimately, standardized and structured assessment reports can be generated, which can include visualizations of multi-ROI matching relationships and lesion change trend charts to meet the needs of clinicians in judging the overall condition of multiple lesions. All results are traceable and reproducible.

[0077] In one embodiment of this application, generating a region of interest assessment result includes: Determine the differences between the first region of interest and the second region of interest in multiple dimensions; these multiple dimensions include multiple aspects such as volume, density, and morphology. A comprehensive score is determined based on differences across multiple dimensions; the comprehensive score is used to characterize the degree of change of the second region of interest compared to the first region of interest.

[0078] Volume dimension difference (ΔV): The core characteristic is the change in lesion size, calculated by the formula ΔV = (V²) / (V²) V1) / V1×100%; where V1 is the volume of the first ROI and V2 is the volume of the second ROI; positive values ​​indicate an increase in volume, negative values ​​indicate a decrease in volume, 0 indicates no change, and the larger the absolute value, the more significant the volume change. Density dimension difference (ΔD): The core characteristic is the change in lesion density and composition. The proportion of solid components can be used as the core quantitative indicator. The calculation formula is ΔD = (Solid2) / (ΔD) = ΔD / ΔD. Solid1) / Solid1×100%; where Solid1 is the percentage of solid component in the first ROI, and Solid2 is the percentage of solid component in the second ROI; positive values ​​indicate an increase in solidity, negative values ​​indicate an increase in glassization, and 0 indicates no change in composition; the average HU value difference can also be added as a supplement, the formula is ΔHU=(HU2) / (Solid1×100%). HU1) / HU1×100%.

[0079] Morphological dimension difference (ΔS): The core characteristic is the regularity of lesion morphological changes, based on the shape feature vector similarity transformation of claim 6, and the calculation formula is ΔS=1. S shape Among them, S shape This represents the morphological similarity value between the first and second ROIs (normalized to [0,1]); a value closer to 0 indicates no morphological change, while a value closer to 1 indicates a more significant morphological change; differences can also be calculated separately for single morphological features such as sphericity and compactness, such as ΔS. p =(S p2 S p1 ) / S p1 ×100% (positive values ​​indicate a shape that is closer to a sphere, while negative values ​​indicate a more irregular shape).

[0080] The overall score is a quantitative value that characterizes the degree of overall change of the second ROI relative to the first ROI. The score range is preferably normalized to [0, 100] points. The lower the score, the smaller the overall change of the lesion, and the higher the score, the more significant the overall change of the lesion. Alternatively, it can be divided into [-50, 50] points according to clinical needs (positive score indicates progression / solidification / irregularization, negative score indicates remission / ground-glass opacity / regularization).

[0081] Based on the differences in different dimensions, scores are calculated for each dimension, and then a comprehensive score is calculated based on the scores for each dimension. For example, a volume score is calculated based on the differences in the volume dimension, a density score is calculated based on the differences in the density dimension, a morphology score is calculated based on the differences in the morphology dimension, and a comprehensive score is calculated based on the volume score, density score, and morphology score.

[0082] This application establishes standardized percentage difference calculation rules for the three core dimensions of volume, density, and morphology, and clarifies the numerical representation of the direction of change, making the dimensional changes of different lesions and different patients directly quantifiable and comparable, solving the problem that traditional assessments of lesion enlargement and shrinkage lack specific numerical values ​​and morphological changes cannot be quantified.

[0083] In one embodiment of this application, the difference in volume dimension includes any one or more of the following: volume change rate, radial change rate, and volume doubling time.

[0084] Calculate the longest diameter: Extract the 3D contour point set of the ROI and calculate the maximum Euclidean distance between all point pairs. Calculate the rate of change of diameter: Rate of change of diameter = (Longest diameter in phase II - Longest diameter in phase I) / Longest diameter in phase I × 100%. According to RECIST criteria, a diameter reduction of more than 30% indicates partial remission, and an increase of more than 20% indicates disease progression.

[0085] If the ROI volume increases, the volume doubling time (VDT) is calculated as: VDT = (time interval × log2) / log(second-phase ROI volume / first-phase ROI volume). VDT is a key indicator for assessing tumor growth rate; typically, the VDT for malignant nodules is less than 400 days.

[0086] Volume score can be determined based on any one or more of the following: volume change rate, radial change rate, and volume doubling time.

[0087] This application's embodiments achieve comprehensive quantification of ROI volume changes through the expansion of three types of indicators, making the assessment of volume dimension differences more comprehensive and refined.

[0088] Calculate the mean Henlein unit (HJ) value for all voxels within the ROI: Mean Henlein Units = ΣHJ value / number of voxels. Calculate the change in mean Henlein Units: ΔHJ = Phase II mean Henlein Units - Phase I mean Henlein Units. An increase in mean Henlein Units may indicate increased solid components, calcification, or hemorrhage; a decrease in mean Henlein Units may indicate necrosis, liquefaction, or cavity formation.

[0089] ROI voxels are classified based on the Henle unit threshold: Real components: Henle units > -400 Ground glass composition: -700 < Heinz units ≤ -400 Emphysema / Air Composition: Henlein Units ≤ -700 Calculate the volume percentage of each component: Solid component ratio = Volume of solid component / Total volume.

[0090] Calculate the change in composition: ΔReal ratio = Real ratio of the second period - Real ratio of the first period.

[0091] An increased solid ratio suggests enhanced tumor invasiveness, while a decreased solid ratio suggests effective treatment.

[0092] Based on Henlein units of ROI voxels, histograms can be constructed, and the density distribution changes can be calculated based on the Bhattacharyya distance between the histograms corresponding to two chest images.

[0093] The density score is calculated based on one or more of the above-mentioned variations in average Henness units, real ratio, and density distribution.

[0094] This application embodiment quantifies the changes in average Henle units, component proportion, and density distribution histogram, and combines the density score with multi-indicator fusion to achieve accurate, comprehensive, and clinically oriented assessment of ROI density changes. It not only makes up for the limitations of traditional single density index assessment, but also transforms the raw CT density data into a quantitative basis that meets the diagnostic and treatment needs of lung lesions (lung cancer / lung nodules), providing standardized technical support for the clinical interpretation, efficacy judgment, and benign-malignant differentiation of cross-stage lesion density changes.

[0095] Calculate the sphericity of the ROI. The sphericity value ranges from 0 to 1, with a larger value indicating that it is closer to a sphere.

[0096] Calculate the change in sphericity: Δsphericity = Sphericity in the second period - Sphericity in the first period.

[0097] Decreased sphericity indicates irregularity in shape, which may be a sign of malignancy; increased sphericity indicates that the shape is becoming more regular, which may be a sign of effective treatment.

[0098] Based on the Sobel filter, the average gradient intensity of the pixels at the ROI boundary is calculated. The higher the gradient value, the clearer the boundary.

[0099] Based on the detection of burr-like protrusions by contour curvature, the number of burrs is counted.

[0100] Perform a Fourier transform on the ROI contour coordinates and extract the standard deviation of the high-frequency components (3rd to 10th harmonics). The larger the standard deviation, the higher the degree of segmentation.

[0101] Calculate the surface area-to-volume ratio of the ROI to determine the change in the surface area-to-volume ratio of the ROI between two periods. An increase in the ratio indicates that the ROI has a more complex shape and more irregular boundaries.

[0102] Based on one or more of the above-mentioned changes in surface area to volume ratio, standard deviation of high-frequency components, number of burrs, average gradient intensity, and sphericity, a morphological score is calculated.

[0103] This application's embodiments achieve precise quantification, pathological correlation, and clinically oriented assessment of ROI morphological changes through quantitative analysis of indicator changes and morphological scoring by fusion of multiple indicators. This overcomes the shortcomings of traditional subjective and singular morphological assessments, transforming the visual characteristics of lung lesion morphology into standardized and calculable quantitative indicators, and providing objective technical support for the identification of malignant signs of lesion morphological evolution across different periods and the determination of treatment efficacy.

[0104] In one embodiment of this application, the patient was diagnosed with non-small cell lung cancer (NSCLC) with a lesion in the upper right lung and received four cycles of chemotherapy. Chest CT scans were performed before treatment (January 10, 2025) and after treatment (July 10, 2025) to assess the efficacy.

[0105] data: - Pre-treatment CT (CT1): 512×512×320 slices, slice thickness 1.0mm, spacing 0.7×0.7mm - Post-treatment CT (CT2): 512×512×305 slices, slice thickness 1.25mm, spacing 0.8×0.8mm - Scan interval: 180 days The evaluation process is as follows: I. Image Preprocessing and Lesion Detection 1. Load DICOM image and extract metadata. 2. HU value standardization 3. Lung dissection: CT1: Slice 50 at the apex of the lung, slice 270 at the base of the lung. CT2: Slice 45 at the apex of the lung, slice 260 at the base of the lung. 4. Lesion detection: CT scan detected one lesion: - Location: Upper right lung, centroid coordinates (256, 180, 120) - Volume: 1580 mm³ - Longest diameter: 21.5 mm CT scan detected one lesion: - Location: Upper right lung, centroid coordinates (260, 175, 115) - Volume: 980 mm³ - Longest diameter: 17.2 mm II. CT Registration 1. Calculate the relative position: CT1, layer 120: (120-50) / (270-50) = 0.318 2. Locate the corresponding layer in CT2: mapped_layer = 45 + 0.318×(260-45) = 113.4≈113 3. Set the search range: [108, 118] 4. Perform affine registration on the candidate layers and calculate the similarity: Layer 108: NCC=0.72 Layer 109: NCC=0.75 ... Layer 115: NCC=0.91 ←Highest ... 5. Determine the registration slice: CT1 slice 120. CT2, layer 115 6. Repeat the above process for all layers. III. Lesion Tracking 1. Candidate screening: The lesion's center of gravity was on slice 120 in CT1, and near slice 115 in CT2. The search range is [110, 120]. The lesion center of CT2 is on slice 115, so it is included in the candidate list. 2. Similarity calculation: Spatial distance: Centroid mapping: (256, 180, 120) → (258, 178, 115) Actual center of gravity: (260, 175, 115) Distance = 3.6 mm S_spatial = exp(-3.6² / (2×20²)) = 0.98 Volume similarity: Volume_ratio = 980 / 1580 = 0.62 S_volume = 0.62 Morphological similarity: Shape_1 = [0.75, 0.82, 0.68, ...] # Sphericity, etc. Shape_2 = [0.78, 0.80, 0.70, ...] S_shape = 0.91 Density similarity: ΔHU = |-380 - (-420)| = 40 S_density = exp(-40 / 100) = 0.67 Overall similarity: S_total = 0.4×0.98 + 0.3×0.62 + 0.2×0.91 + 0.1×0.67 = 0.392 + 0.186 + 0.182 + 0.067 = 0.827 3. Matching results: Similarity 0.827 > threshold 0.7 → Match successful IV. Change Assessment Geometric feature changes: - Volume change: 980 - 1580 = -600 mm³ - Volume change rate: -600 / 1580 × 100% = -38.0% - VDT: N / A (Volume Reduction) - Variation in diameter: 17.2 - 21.5 = -4.3 mm - Rate of change of diameter: -4.3 / 21.5 × 100% = -20.0% - RECIST classification: -20.0% < -30%? No → SD (close to PR boundary) Density characteristics change: - Average HU change: -380 - (-420) = +40 HU - Change in the proportion of actual transactions: 0.58 - 0.52 = +0.06 Explanation: Shrinking lesions with increased density may indicate necrosis absorption. Morphological changes: - Sphericity change: 0.78 - 0.75 = +0.03 - Number of edge burrs: 5 → 3 (reduced) - Explanation: The shape tends to be regular, and the boundaries become clearer. Overall assessment: - Significantly reduced volume (-38%) - Improved shape (increased sphericity, reduced burrs) - Density changes (solidification) may be due to post-necrosis fibrosis. - Conclusion: Partial remission (PR), continued chemotherapy recommended. Output report: Lung Cancer Treatment Efficacy Evaluation Report Patient Information: XXX Diagnosis: Non-small cell lung cancer (NSCLC) Treatment regimen: 4 cycles of chemotherapy (pemetrexed + cisplatin) Image comparison: - Baseline scan: 2025-01-10 - Follow-up scan: 2025-07-10 - Time interval: 180 days Lesion tracking results: Lesion number: 1 Location: Right upper lung Match quality: 0.827 (High) Baseline characteristics: -Volume: 1580 mm³ - Longest diameter: 21.5 mm -Average HU: -420 HU - Substance percentage: 52% Follow-up characteristics: -Volume: 980 mm³ - Longest diameter: 17.2 mm - Average HU: 380 HU - Actual percentage: 58% Change Analysis: -Volume change: -600 mm³ (-38.0%) ↓ -Diameter variation: -4.3 mm (-20.0%) ↓ - RECIST assessment: SD (close to PR) -Density change: +40 HU (realization) - Improved shape: Sphericity increased, burrs decreased Overall assessment conclusion: The lesion volume has shrunk significantly and the shape has become more regular, indicating a good response to chemotherapy.

[0106] Although the RECIST criteria classify it as SD, considering multiple dimensions of indicators, it is close to the PR criteria.

[0107] Increased density may indicate post-necrosis fibrosis, which is a sign that the treatment is effective.

[0108] Clinical recommendations: 1. Continue the current chemotherapy regimen. 2. It is recommended to have a follow-up CT scan in 2-3 months to assess any further changes. 3. If the shrinkage continues, surgical treatment may be considered. like Figure 2 As shown, this application embodiment provides a region of interest evaluation apparatus, including: The acquisition module 201 is configured to acquire chest images of the user at different times; The determination module 202 is configured to perform the following for each chest image: extract the lung region from the chest image, identify anatomical landmarks in the lung region, and determine the relative position of each layer in the chest image based on the anatomical landmarks; The generation module 203 is configured to determine the mapping relationship between different chest images based on the relative positions of each layer in each chest image; identify the region of interest in the lung region of each chest image; and generate the region of interest evaluation result based on the layer in which the region of interest is located in the chest image and the mapping relationship between different chest images.

[0109] In one embodiment of this application, the determining module 202 is configured to divide the lung region into the left lung and the right lung; and to determine the apex and base of the left lung and the apex and base of the right lung, respectively.

[0110] In one embodiment of this application, the generation module 203 is configured to determine a first mapping relationship between each layer in the first chest image and each layer in the second chest image based on the relative positions of each layer in the first chest image and each layer in the second chest image; and to perform the following for each layer in the first chest image: based on the first mapping relationship, determine multiple candidate layers corresponding to the current layer in the first chest image; determine the affine transformation matrix corresponding to each candidate layer; perform spatial transformation on the candidate layers based on the affine transformation matrix; calculate the similarity between the transformed candidate layers and the current layer; and determine a second mapping relationship between each layer in the first chest image and each layer in the second chest image based on the similarity between the current layer and each transformed candidate layer.

[0111] In one embodiment of this application, the generation module 203 is configured to perform the following for each candidate layer: detect key points in the current layer and key points in the candidate layer; determine the similarity between each key point in the current layer and each key point in the candidate layer; determine the matching relationship between the key points in the current layer and the key points in the candidate layer based on the similarity; and calculate the affine transformation matrix based on the matching relationship.

[0112] In one embodiment of this application, the generation module 203 is configured to, based on a second mapping relationship, determine the reference layer in the second chest image corresponding to the target layer in the first chest image where the first region of interest is located; determine the search range based on the reference layer; determine the second region of interest located within the search range; calculate the comprehensive similarity between the first region of interest and the second region of interest; and generate a region of interest evaluation result based on the comprehensive similarity between the first region of interest and the second region of interest.

[0113] In one embodiment of this application, the generation module 203 is configured to: calculate the spatial similarity between the first region of interest (ROI) and the second region of interest (ROI) based on the centroids of the first and second ROIs; calculate the volume similarity between the first and second ROIs based on their volumes; obtain the shape feature vectors of the first and second ROIs; determine the morphological similarity between the first and second ROIs based on their morphological feature vectors; determine the density similarity between the first and second ROIs based on their Henlein unit values; and calculate the comprehensive similarity based on the spatial similarity, volume similarity, morphological similarity, and density similarity.

[0114] In one embodiment of this application, the generation module 203 is configured to construct a similarity matrix based on the comprehensive similarity between multiple first regions of interest and multiple second regions of interest; wherein, the elements in the similarity matrix are used to characterize the comprehensive similarity between the first regions of interest and the second regions of interest; a cost matrix is ​​constructed based on the similarity matrix; based on the cost matrix, the matching relationship between the first regions of interest and the second regions of interest, and / or the unmatched first regions of interest, and / or the unmatched second regions of interest are determined; and if there is a matching relationship between the first regions of interest and the second regions of interest, an evaluation result of the regions of interest is generated.

[0115] In one embodiment of this application, the generation module 203 is configured to determine the differences between a first region of interest and a second region of interest in multiple dimensions; wherein the multiple dimensions include multiple of volume, density and morphology; and to determine a comprehensive score based on the differences in multiple dimensions; wherein the comprehensive score is used to characterize the degree of change of the second region of interest compared to the first region of interest.

[0116] This application provides an electronic device, including: One or more processors; Storage device for storing one or more programs. When one or more programs are executed by one or more processors, the one or more processors implement the methods as described in any of the above embodiments.

[0117] This application provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any of the above embodiments.

[0118] This application provides a computer program product that, when executed by a processor, implements the method as described in any of the above embodiments.

[0119] It should be understood that in the embodiments of this application, the processor may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0120] It should also be understood that the memory mentioned in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (Read-Only Memory). Only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM).

[0121] It should be noted that when the processor is a general-purpose processor, DSP, ASIC, FPGA, or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, the memory (storage module) is integrated into the processor.

[0122] It should be noted that the memories described herein are intended to include, but are not limited to, these and any other suitable types of memories.

[0123] In addition to the data bus, this bus may also include a power bus, a control bus, and a status signal bus. However, for clarity, all buses are labeled "bus" in the diagram.

[0124] It should also be understood that the first, second, third, fourth and various numerical designations used herein are merely for descriptive convenience and are not intended to limit the scope of this application.

[0125] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0126] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The steps of the method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0127] In the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0128] Those skilled in the art will recognize that the various illustrative logical blocks (ILBs) and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0129] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0130] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0131] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0132] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0133] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for evaluating regions of interest, characterized in that, include: Obtain chest images of the user at different times; For each of the chest images, the following steps are performed: extracting the lung region from the chest image, identifying anatomical landmarks in the lung region, and determining the relative positions of each layer in the chest image based on the anatomical landmarks; Based on the relative positions of each layer in each of the aforementioned chest images, the mapping relationship between the layers of different chest images is determined; Identify the region of interest in the lung region of each of the chest images; Based on the layer in which the region of interest is located in the chest image and the mapping relationship between different layers of the chest image, a region of interest evaluation result is generated.

2. The method as described in claim 1, characterized in that, Identifying anatomical landmarks in the lung region includes: The lung region is divided into the left lung and the right lung; The apex and base of the left lung, and the apex and base of the right lung, are determined respectively.

3. The method as described in claim 1, characterized in that, Based on the relative positions of each layer in each of the aforementioned chest images, the mapping relationship between the layers of different chest images is determined, including: Based on the relative positions of each layer in the first chest image and the relative positions of each layer in the second chest image, a first mapping relationship between each layer in the first chest image and each layer in the second chest image is determined. For each layer in the first chest image, the following steps are performed: based on the first mapping relationship, in each layer of the second chest image, determine multiple candidate layers corresponding to the current layer in the first chest image; determine the affine transformation matrix corresponding to each candidate layer; perform a spatial transformation on the candidate layers based on the affine transformation matrix; calculate the similarity between the transformed candidate layers and the current layer; and based on the similarity between the current layer and each transformed candidate layer, determine a second mapping relationship between each layer in the first chest image and each layer in the second chest image.

4. The method as described in claim 3, characterized in that, Determining the affine transformation matrix corresponding to each of the candidate layers includes: Execute for each candidate layer: Detect key points in the current layer and key points in the candidate layer; Determine the similarity between each key point in the current layer and each key point in the candidate layer; Based on the similarity, the matching relationship between the key points in the current layer and the key points in the candidate layer is determined; Based on the matching relationship, the affine transformation matrix is ​​calculated.

5. The method as described in claim 3, characterized in that, Based on the layer in which the region of interest is located in the chest image and the mapping relationship between different layers of the chest image, a region of interest evaluation result is generated, including: Based on the second mapping relationship, the reference layer in the second chest image corresponding to the target layer where the first region of interest is located in the first chest image is determined. The search range is determined based on the reference layer; Identify a second region of interest located within the search range; Calculate the overall similarity between the first region of interest and the second region of interest; Based on the comprehensive similarity between the first region of interest and the second region of interest, a region of interest evaluation result is generated.

6. The method as described in claim 5, characterized in that, Calculating the overall similarity between the first region of interest and the second region of interest includes: Based on the centroids of the first region of interest and the second region of interest, calculate the spatial similarity between the first region of interest and the second region of interest. Based on the volume of the first region of interest and the volume of the second region of interest, calculate the volume similarity between the first region of interest and the second region of interest; Obtain the shape feature vectors of the first region of interest and the second region of interest; Based on the morphological feature vectors of the first region of interest and the second region of interest, the morphological similarity between the first region of interest and the second region of interest is determined. Based on the Henlein unit values ​​of the first region of interest and the second region of interest, the density similarity between the first region of interest and the second region of interest is determined; The comprehensive similarity is calculated based on the spatial similarity, the volume similarity, the morphological similarity, and the density similarity.

7. The method as described in claim 5, characterized in that, Based on the comprehensive similarity between the first region of interest and the second region of interest, a region of interest evaluation result is generated, including: A similarity matrix is ​​constructed based on the comprehensive similarity between multiple first regions of interest and multiple second regions of interest; wherein, the elements in the similarity matrix are used to characterize the comprehensive similarity between the first regions of interest and the second regions of interest; Based on the similarity matrix, construct the cost matrix; Based on the cost matrix, determine the matching relationship between the first region of interest and the second region of interest, and / or the unmatched first region of interest, and / or the unmatched second region of interest; If there is a matching relationship between the first region of interest and the second region of interest, a region of interest evaluation result is generated.

8. The method as described in claim 7, characterized in that, Generate region of interest assessment results, including: The differences between the first region of interest and the second region of interest are determined in multiple dimensions; wherein, the multiple dimensions include multiple of volume, density and morphology; A comprehensive score is determined based on the differences in the multiple dimensions; wherein the comprehensive score is used to characterize the degree of change of the second region of interest compared to the first region of interest.

9. The method as described in claim 7, characterized in that, in, Differences in volume dimensions include any one or more of the following: rate of change of volume, rate of change of diameter, and time of volume doubling.

10. A region of interest assessment device, characterized in that, include: The acquisition module is configured to acquire chest images of the user at different times; The determination module is configured to perform the following for each of the chest images: extracting the lung region from the chest image, identifying anatomical landmarks in the lung region, and determining the relative positions of each layer in the chest image based on the anatomical landmarks; The generation module is configured to determine the mapping relationship between different layers of the chest images based on the relative positions of each layer in each chest image; identify the region of interest in the lung region of each chest image; and generate a region of interest evaluation result based on the layer in the chest image where the region of interest is located and the mapping relationship between different layers of the chest images.

Images