CN-122025034-A - Respiratory tract normalization model construction method and system based on medical image

CN122025034ACN 122025034 ACN122025034 ACN 122025034ACN-122025034-A

Abstract

The application provides a respiratory tract normalization model construction method and a system based on medical images, which relate to the technical field of medical image modeling, and the application obtains a volume median model as a reference model by acquiring an airway segmentation result of a chest CT image of a target group; the method comprises the steps of taking a reference model as a target, performing affine registration on other models to form a preliminary registration model group, then selecting a sample subset from the preliminary registration model group to perform nonlinear deformation registration to obtain a fine registration model group, then superposing the fine registration model group and the reference model to generate an airway appearance probability map, calculating a space self-adaptive threshold map according to the space consistency of two airway frameworks, finally extracting voxels with probability values higher than a corresponding threshold to generate a normalized airway model, combining the fine registration model group to perform statistical shape modeling, extracting a main variation mode of the airway morphology of a target group, and providing a basic template and quantifying the main variation of the morphology for group comparison, disease identification and the like.

Inventors

XU XINXI
GUO WEIQI
ZHANG YUEHUA
FAN JINBO
LI PENGHUI
ZHAO XIUGUO
SU CHEN

Assignees

军事科学院系统工程研究院卫勤保障技术研究所

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. A respiratory tract normalization model construction method based on medical images is characterized by comprising the following steps: Acquiring airway segmentation results of chest CT images of a target group, obtaining a plurality of airway three-dimensional models, and selecting a median model from the airway three-dimensional models according to the volume as a reference model; Performing affine registration on the rest of the airway three-dimensional models by taking the reference model as a target to obtain a preliminary registration model group with spatial alignment; Selecting a sample subset from the preliminary registration model set, and performing nonlinear deformation registration on the sample subset to obtain a fine registration model set; Performing spatial superposition on the fine registration model set and a mask of the reference model to generate a probability map representing the occurrence probability of the airway, and calculating a spatial self-adaptive threshold map according to the spatial consistency of the fine registration model set and the airway skeleton of the reference model; combining the probability map with the threshold map, extracting voxels with probability values higher than the corresponding position self-adaptive threshold values, and generating a normalized respiratory tract model; and carrying out statistical shape modeling based on the normalized respiratory tract model and the fine registration model group so as to extract a main variation mode of the airway morphology of the target group.
2. The method of claim 1, wherein the selecting a subset of samples from the preliminary registration model set, performing non-linear deformation registration on the subset of samples to obtain a fine registration model set, comprises: Taking the volume value of the reference model as the center, and selecting a plurality of models from the preliminary registration model group to form a sample subset in a preset volume value range; setting a total time upper limit for the whole registration process of the sample subset, and executing nonlinear deformation registration with the reference model one by one on the models in the sample subset, wherein the nonlinear deformation registration process synchronously optimizes the spatial correspondence of the two parties in a bidirectional iteration mode; And in the one-by-one registration process, dynamically predicting the residual time required by completing the registration of all the residual models, stopping the subsequent registration if the residual time exceeds the preset proportion of the total time upper limit, and outputting all the models which have completed the registration at the moment as a fine registration model group.
3. The method of claim 2, wherein the process of nonlinear deformation registration further comprises: Identifying coarse airway regions and fine branch regions on each preliminary registration model based on the morphology of each preliminary registration model in the sample subset; applying strong smoothing to the coarse airway region and weak smoothing or no smoothing to the fine branch region; and respectively adopting different regularization constraint parameters and iteration adjustment strategies for the coarse airway region and the fine branch region.
4. The method of claim 1, wherein the computing a spatially adaptive thresholding map based on spatial correspondence of the fine registration model set to the airway skeleton of the reference model comprises: respectively extracting the airway skeleton of each model and the reference model in the fine registration model group to obtain a corresponding skeleton binary image; superposing all the skeleton binary diagrams in a unified three-dimensional space, and calculating the frequency of marking each space position as a skeleton point; Calculating an adaptive threshold for each spatial position based on the frequency, wherein the higher the frequency is, the lower the corresponding threshold is; the threshold map is composed of adaptive thresholds for all spatial locations.
5. The method of claim 1, wherein the combining the probability map with the threshold map extracts voxels having probability values above corresponding location-adaptive thresholds, generating a normalized respiratory model, comprising: in a unified three-dimensional space coordinate system, comparing the probability value of each space position in the probability map with the self-adaptive threshold value of the same space position in the threshold map; marking all spatial positions of which the probability values are larger than the adaptive threshold value; The set of all marked spatial locations is used as a normalized respiratory tract model.
6. The method of claim 1, wherein the statistically shape modeling based on the normalized airway model and the fine registration model set to extract a dominant variability pattern of airway morphology of the target population comprises: Performing unified space and image attribute standardization processing on each model in the normalized respiratory tract model and the fine registration model group; Preliminarily aligning the space positions of the standardized models; Iteratively adjusting the spatial positions of the corresponding particles on all models by an optimization algorithm to minimize an energy function distributed across the model particles, thereby establishing a stable spatial correspondence between anatomical structures of all models; And based on the spatial correspondence, performing principal component analysis on all the corresponding point positions, and extracting a main variation direction obtained by the principal component analysis as an airway morphology variation mode of a target group.
7. The method of claim 1, further comprising performing a three-dimensional structural feature quantification analysis on the normalized airway model, the quantification analysis comprising: Extracting an airway skeleton of the normalized airway model, calculating the total length of the airway skeleton of the model based on the airway skeleton, sampling on each point of the airway skeleton to obtain a corresponding airway local radius set, and calculating the average radius of the airway; identifying branch points and terminal points based on the airway skeleton, and respectively counting the total number of the branch points and the total number of the terminal points; And in the normalized respiratory tract model, identifying all voxels with the local diameter of the airway smaller than a preset threshold value, and calculating the proportion of the number of the voxels to the total number of the airway voxels of the model as a small branch proportion.
8. The method as recited in claim 1, further comprising: Carrying out multi-threshold slicing on the probability map to obtain a plurality of airway templates under different confidence levels, wherein each airway template is composed of voxels with probability values larger than corresponding thresholds; for each of the plurality of airway templates, its volume, skeleton length, number of bifurcation points, and number of end points are calculated.
9. The method of claim 1, further comprising verifying a representation of the normalized airway model: Obtaining geometrical parameters of a target population sample at a plurality of bifurcation levels, and calculating statistical confidence intervals of the geometrical parameters in the population; verifying whether corresponding geometric parameter values of the normalized respiratory tract model at the same bifurcation level fall within the statistical confidence interval.
10. A respiratory tract normalization model construction system based on medical imaging, comprising: The acquisition module is used for acquiring airway segmentation results of chest CT images of the target group, obtaining a plurality of airway three-dimensional models, and selecting a median model from the airway three-dimensional models according to the volume as a reference model; the registration module is used for carrying out affine registration on the other airway three-dimensional models by taking the reference model as a target so as to obtain a preliminary registration model group with spatial alignment; The registration module is further used for selecting a sample subset from the preliminary registration model set, and performing nonlinear deformation registration on the sample subset to obtain a fine registration model set; the generation module is used for carrying out spatial superposition on the fine registration model set and the mask of the reference model to generate a probability map representing the occurrence probability of the airway, and calculating a spatially self-adaptive threshold map according to the spatial consistency of the fine registration model set and the airway skeleton of the reference model; The generation module is also used for combining the probability map and the threshold map, extracting voxels with probability values higher than the corresponding position self-adaptive threshold, and generating a normalized respiratory tract model; And the extraction module is used for carrying out statistical shape modeling based on the normalized respiratory tract model and the fine registration model group so as to extract the main variation mode of the airway morphology of the target group.

Description

Respiratory tract normalization model construction method and system based on medical image Technical Field The application relates to the technical field of medical image modeling, in particular to a respiratory tract normalization model construction method and system based on medical images. Background The respiratory tract normalization model construction based on the medical image is a key technology in the field of medical image processing, can provide basic support for respiratory tract group morphology analysis, early disease identification and AI model training, and has wide application prospect. The current common respiratory tract normalization correlation method is mainly based on statistical modeling of measurement parameters, namely data fitting is carried out by extracting a few morphological indexes of the airway, so that a reference model for comparison analysis is constructed, and the method is suitable for preliminary airway morphological research. The method is difficult to comprehensively reflect the three-dimensional morphological characteristics of the airway, and is in face of the conditions of large airway morphology difference and non-uniform image coordinate system and posture of different individuals, and the uniform alignment of the group-level structure cannot be realized, so that the built model lacks stability and representativeness. Therefore, the technical problem of poor standardization effect of the airway structure under the group scale exists in the prior art. Disclosure of Invention The application aims to provide a respiratory tract normalization model construction method and system based on medical images, which are used for solving the problem of poor normalization effect of airway structures under group scale in the prior art. In order to solve the technical problems, in a first aspect, the present application provides a respiratory tract normalization model construction method based on medical images, including: Acquiring airway segmentation results of chest CT images of a target group, obtaining a plurality of airway three-dimensional models, and selecting a median model from the airway three-dimensional models according to the volume as a reference model; Performing affine registration on the rest of the airway three-dimensional models by taking the reference model as a target to obtain a preliminary registration model group with spatial alignment; Selecting a sample subset from the preliminary registration model set, and performing nonlinear deformation registration on the sample subset to obtain a fine registration model set; Performing spatial superposition on the fine registration model set and a mask of the reference model to generate a probability map representing the occurrence probability of the airway, and calculating a spatial self-adaptive threshold map according to the spatial consistency of the fine registration model set and the airway skeleton of the reference model; combining the probability map with the threshold map, extracting voxels with probability values higher than the corresponding position self-adaptive threshold values, and generating a normalized respiratory tract model; and carrying out statistical shape modeling based on the normalized respiratory tract model and the fine registration model group so as to extract a main variation mode of the airway morphology of the target group. Optionally, the selecting a sample subset from the preliminary registration model set, and performing nonlinear deformation registration on the sample subset to obtain a fine registration model set, including: Taking the volume value of the reference model as the center, and selecting a plurality of models from the preliminary registration model group to form a sample subset in a preset volume value range; setting a total time upper limit for the whole registration process of the sample subset, and executing nonlinear deformation registration with the reference model one by one on the models in the sample subset, wherein the nonlinear deformation registration process synchronously optimizes the spatial correspondence of the two parties in a bidirectional iteration mode; And in the one-by-one registration process, dynamically predicting the residual time required by completing the registration of all the residual models, stopping the subsequent registration if the residual time exceeds the preset proportion of the total time upper limit, and outputting all the models which have completed the registration at the moment as a fine registration model group. Optionally, the process of nonlinear deformation registration further includes: Identifying coarse airway regions and fine branch regions on each preliminary registration model based on the morphology of each preliminary registration model in the sample subset; applying strong smoothing to the coarse airway region and weak smoothing or no smoothing to the fine branch region; and respectively adopting different