CN-121640519-B - Preoperative evaluation-oriented perivascular inflammation image group identification method and system for pancreatic system membranous region

Abstract

The invention relates to the technical field of medical image analysis, in particular to a pancreas membranous region perivascular inflammation image group identification method and system for preoperative evaluation. The method comprises the steps of obtaining a multi-stage reinforced abdominal image, generating a standardized image structure through data quality control, denoising registration, intensity normalization and voxel resampling, carrying out pancreas membranous region blood vessel candidate region positioning, blood vessel skeleton extraction and central line tracking, constructing an annular region of interest structure, carrying out blood vessel and surrounding tissue segmentation and mask generation, extracting traditional image histology characteristics and coupling characteristics reflecting interaction relation between the blood vessel and surrounding tissue, generating an effective characteristic set, and generating an inference report comprising an inflammation identification tag and an interpretable mapping through domain correction, identification model training and parameter solidification. The invention realizes the automatic and accurate quantitative identification of perivascular inflammation, and effectively improves the objectivity and accuracy of preoperative evaluation.

Inventors

• ZHANG BEIBEI
• LI TIANRAN
• ZHANG DONGLI

Assignees

• 中国人民解放军总医院第四医学中心

Dates

Publication Date

20260508

Application Date

20251225

Claims (4)

1. 1. A preoperatively-assessed perivascular inflammatory imaging group identification method for pancreatic system membranous regions, comprising: Acquiring a multi-stage enhanced abdominal image, clinical information fields and a preprocessing configuration structure, and executing data quality control and acquisition parameter recording, denoising registration and intensity normalization and voxel resampling processing to generate a standardized image structure; Based on a standardized image structure, performing pancreatic system membranous region blood vessel candidate region positioning, blood vessel skeleton extraction and central line tracking, and annular region of interest construction treatment to obtain an annular region of interest structure, wherein the annular region of interest structure comprises an annular shell voxel set pointer, an annular region of interest index table pointer, a section normal sequence pointer, an inside and outside radius parameter table pointer, a sampling step length parameter table pointer and a blood vessel central line structure pointer; based on the annular region of interest structure, performing segmentation and mask generation of blood vessels and surrounding tissues, extraction of image histology characteristics and coupling characteristic construction processing, and generating an effective characteristic set structure; Based on the effective feature set structure, performing domain correction processing, recognition model training and model parameter curing processing to generate an inference report structure; The process of performing the localization of the candidate region of the blood vessel of the pancreatic system membrane region further comprises: the method comprises the steps of firstly extracting abdomen body cavity boundary and spine reference layer field from a standardized image structure, then constructing a three-dimensional space limiting frame as a pancreas system membrane region range structure according to relative position constraint of the pancreas system membrane region in the body cavity, wherein the relative position constraint comprises a longitudinal range close to the mesenteric root running direction, a front-rear range close to the pancreas rear and a rejection range avoiding an air region outside the body cavity and a cortical bone high-density region, further extracting each time phase image from the standardized image structure, selecting a time phase combination after enhancement as a candidate screening source according to the standardized image index table, executing tubular structure response calculation on each time phase, wherein the tubular structure response is multi-scale filtering output aiming at an elongated communication form, the implementation process is that local direction consistency and boundary gradient stability are calculated in different scales, a response body is formed, then executing threshold segmentation on the response body to generate a candidate voxel set, executing a communication region analysis and outputting a candidate communication region table under the constraint of the pancreas system membrane region range structure, wherein the candidate voxel set comprises a communication region identifier, a voxel number, a main shaft direction, an average enhancement strength and a bifurcation degree mark, and a trigger condition are simultaneously, if the number of the communication region is less than the trigger rule is set in a certain condition, and the trigger is set when the rule is switched in the predetermined state, and the communication region is arranged in the predetermined state, and the rule is read; The process of extracting the vascular skeleton and tracking the central line further comprises the steps of performing hole filling and fracture bridging processing on a candidate voxel set, wherein the hole filling identifies voxel holes through local connectivity detection and performs neighborhood filling, fracture bridging is completed through short-distance end point pairing and form closing operation, then the voxel set after filling is subjected to thinning processing to generate the vascular skeleton voxel set, boundary voxels are iteratively stripped and the topology of end points and bifurcation points is kept unchanged through the thinning processing, a skeleton connected graph is constructed for the vascular skeleton voxel set, the skeleton connected graph consists of node sets and edge sets, the central line tracking processing comprises selecting a trunk candidate connected domain according to a main shaft direction field in a candidate connected domain table, selecting an inlet end point in the trunk candidate connected domain and performing path traversal to generate a central line coordinate sequence, the path traversal adopts a gradual expansion strategy comprising the connected graph, and simultaneously branches and records each bifurcation point according to a bifurcation point number table to form a central line quality mark table, the central line quality mark table comprises path continuity, bifurcation stability and loop detection mark fields, and the central line quality mark table displays the secondary bridging degree of a quality mark triggering candidate area when the path continuity is insufficient; The construction process of the annular region of interest further comprises the steps of extracting a central line coordinate sequence from a blood vessel central line structure and carrying out central line re-parameterization, wherein the central line re-parameterization generates a sampling point sequence by uniformly sampling according to arc length, sampling step length is taken from a preprocessing configuration structure and is written into a sampling step length parameter table, a section normal sequence is constructed at each sampling point, the section normal sequence generates a local tangential direction through direction difference of adjacent sampling points and is combined with a direction cosine field of a voxel grid description table to carry out coordinate consistency, an inner radius parameter table and an outer radius parameter table are generated according to an inner radius rule and an outer radius rule in the preprocessing configuration structure, the inner radius rule comprises a fixed radius scheme and a segmentation radius scheme, the fixed radius scheme gives different radius fields to different paragraphs according to local blood vessel scale estimation fields of a communication domain where the central line is located, an annular section voxel set is generated at each sampling point and is accumulated along the central line to form an annular shell voxel set, and simultaneously, and the annular shell set and the central line coordinate sequence are established to be associated with the central line coordinate to generate an annular region of interest index table; The process for executing segmentation and mask generation of blood vessels and surrounding tissues further comprises the steps of extracting an annular shell voxel set and an annular region of interest index table from the annular region of interest structure, calling corresponding voxel blocks in the standardized image structure according to sampling point marks and section marks of the annular region of interest index table to form a segmentation input block sequence, generating a segmentation path record table, separating a blood vessel region from a surrounding tissue region by using three types of paths of threshold segmentation, region growing and learning model reasoning, setting an intensity threshold value by using the threshold segmentation path to contain an enhancement intensity field and a candidate region index field of the standardized image structure, generating an initial blood vessel voxel set, performing boundary shrinkage and expansion correction according to a section normal sequence along a normal direction, selecting a seed point at an endpoint position of a vascular mask candidate, expanding according to a connected region similarity criterion, inputting a segmentation input block sequence into a training completed segmentation model by using the learning model path, outputting a probability map, generating a blood vessel voxel set according to a probability threshold value and a connected constraint, writing a segmentation value into a segmentation map, and a quality map, and triggering a quality map label when the segmentation map is written into a quality map label, and a quality map label is written in the mask label, and a quality map label is simultaneously triggered when the quality label is marked and a mask label is in the mask label is in a mask label; the image group chemical characteristic extraction and coupling characteristic construction process further comprises the steps of extracting an intensity statistical characteristic from a standardized image structure according to a vascular mask range, wherein the intensity statistical characteristic comprises a mean value, a branch point and a kurtosis type field, extracting texture characteristics from the vascular mask and surrounding tissue masks, the texture characteristics are generated through grey level discretization and neighborhood co-occurrence statistics, a bin rule and a neighborhood definition rule of the grey level discretization write in a characteristic extraction parameter table, the extracted morphological characteristics comprise surface reconstruction of a mask voxel set to obtain a volume, a surface area, a fine length and a bifurcation complexity field, the coupling characteristic construction process comprises the establishment of an interface neighborhood zone on the outer wall of the vascular mask, the interface neighborhood zone is formed by an intersection voxel set of an expansion voxel set of the vascular mask and a surrounding tissue mask, the thickness of the interface neighborhood zone is constrained by an inner radius parameter table, an outer radius parameter table and a sampling step length parameter table, the coupling characteristic table comprises a texture differential index, a gradient co-occurrence index and an interface morphology index, the texture differential index is generated through the difference value of the texture statistics in the vascular mask and the interface neighborhood zone texture statistics, the gradient co-occurrence index is generated through the gradient distribution in the vascular mask boundary direction and the boundary distribution gradient distribution and the boundary distribution is generated through the gradient contact with the surrounding tissue mask boundary distribution and the gradient distribution is generated; the effective feature set structure comprises a model entering feature list, a stability evaluation list, a screening rule record list, a selection rule record list, a feature screening log structure and an associated feature matrix index.
2. 2. The method of claim 1, wherein performing data quality control and acquisition parameter recording further comprises: The data quality control processing comprises missing segment checking, artifact marking and sequence integrity checking, wherein the missing segment checking identifies a continuous slice gap by comparing slice sequence fields in an image sequence index table with actual loaded slice counts and writes a missing range field in a data quality control marking table, the artifact marking is marked by performing coarse screening before intensity outlier suppression on each time phase slice and combining with a stripe, metal artifact and a motion blurred texture abnormal mode and writes an artifact type field in the data quality control marking table, the sequence integrity checking is used for generating a sequence integrity checking result field by checking a time phase identification set and a time phase set specified in a preprocessing configuration structure, the acquisition parameter recording processing comprises recording equipment model, layer thickness, reconstruction setting and enhanced contrast agent injection setting, wherein the equipment model is from image metadata, the layer thickness is from a slice interval field, the reconstruction setting is from a reconstruction core and a matrix field, the enhanced contrast agent injection setting is from an injection rate and dose field or an intra-hospital check single field, and the missing marking field is written in the acquisition parameter recording table when the field is missing and the preprocessing log structure is synchronously written.
3. 3. The method of claim 1, wherein the process of voxel resampling processing further comprises: The voxel resampling processing comprises the steps of reading a target voxel spacing field and a target grid size field, selecting a resampling interpolation mode and generating a resampling parameter record table, writing the resampling parameter record table into the interpolation mode field, the target voxel spacing field and the target grid size field, synchronously writing the resampling parameter record table into a preprocessing log structure, and generating a voxel grid description table at the same time, and generating a standardized image structure, wherein the standardized image structure comprises a standardized image data pointer, a standardized image index table, a voxel grid description table pointer, a normalization parameter record table pointer, a resampling parameter record table pointer and a preprocessing log structure pointer field.
4. 4. A preoperatively-assessed perivascular inflammatory imaging group identification system for the membranous region of the pancreas, applied to the method according to any of the preceding claims 1-3, comprising: the standardized image acquisition and preprocessing unit is used for executing data quality control and acquisition parameter recording, denoising registration and intensity normalization and voxel resampling processing to generate a standardized image structure; The blood vessel center line positioning and annular region of interest constructing unit is used for executing pancreatic system membranous region blood vessel candidate region positioning, blood vessel skeleton extraction and center line tracking, and annular region of interest constructing processing to obtain an annular region of interest structure; The segmentation and mask generation unit is used for executing segmentation of blood vessels and surrounding tissues and generating a blood vessel mask and a surrounding tissue mask to form a segmentation mask structure; The image histology feature extraction and coupling feature construction unit is used for extracting intensity features, texture features and morphological features to generate an image histology feature table; The feature stability screening and feature selecting unit is used for reading the resampling parameter record table, the normalization parameter record table, the registration parameter record table and the acquisition parameter record table to generate a disturbance configuration set; and the domain correction model training reasoning and version management unit is used for executing domain correction processing to obtain a domain correction characteristic structure and generating an alignment mapping record table.

Description

Preoperative evaluation-oriented perivascular inflammation image group identification method and system for pancreatic system membranous region Technical Field The invention relates to the technical field of medical image analysis, in particular to a pancreas membranous region perivascular inflammation image group identification method and system for preoperative evaluation. Background In the field of medical image analysis, the existing scheme for identifying perivascular inflammation of pancreatic system membranous regions for preoperative evaluation is usually based on manual interpretation and empirical interpretation of multi-stage enhanced abdominal images, or the method comprises the steps of completing candidate region positioning and feature extraction in a local region and giving out a judging result, and has the limitations that the standardized image structure forming process is not uniform, the acquisition parameter recording table and the preprocessing configuration structure are missing or incomplete, the annular region of interest structure is difficult to stably construct and the like. The existing method is mostly dependent on subjective limitation and threshold adjustment of operators on the peripheral range of the blood vessel, the segmentation of the blood vessel and the peripheral tissues and the mask generation are easily influenced by imaging noise, registration errors and voxel resampling differences, the problems that the boundary between the blood vessel mask and the peripheral tissue mask is inconsistent, the mask quality mark is difficult to review and the like easily occur under the change of equipment and scanning conditions reflected by an acquisition parameter recording table, and the stable realization of feature matrix construction and effective feature set structure generation is difficult to meet. Aiming at the joint processing of image histology feature extraction and coupling feature construction, the prior art generally lacks the unified definition of an interface neighborhood zone of a vascular mask and a surrounding tissue mask and the unified management of a feature naming dictionary, and the association among a feature missing mark table, a feature extraction parameter table and a preprocessing log structure is incomplete, so that feature fields of homologous data are difficult to align in different batches of processing. Aiming at feature stability screening and feature selection operation, the prior art generally lacks consistency statistics and structured records of a screening rule record table and a selection rule record table under the controlled disturbance condition, and is difficult to form a recheckable model entering feature list. Aiming at continuous links generated by domain correction processing, recognition model training, model parameter solidification, case reasoning and interpretability mapping, the prior art has inconsistency in the aspect of the associated registration of an alignment mapping record table, a model version structure and a reasoning report structure, the update of a preprocessing configuration structure lacks traceable support, and the consistency process of acquisition, alignment, judgment, recording and update is difficult to form in a preoperative evaluation flow, so that the review and cross-batch comparison of the reasoning report structure are influenced. Disclosure of Invention In order to solve the technical problems, the invention provides a preoperatively-estimated perivascular inflammation image group identification method for pancreatic system membranous regions, which comprises the following steps: Acquiring a multi-stage enhanced abdominal image, clinical information fields and a preprocessing configuration structure, and executing data quality control and acquisition parameter recording, denoising registration and intensity normalization and voxel resampling processing to generate a standardized image structure; Based on a standardized image structure, performing pancreatic system membranous region blood vessel candidate region positioning, blood vessel skeleton extraction and central line tracking, and annular region of interest construction treatment to obtain an annular region of interest structure, wherein the annular region of interest structure comprises an annular shell voxel set pointer, an annular region of interest index table pointer, a section normal sequence pointer, an inside and outside radius parameter table pointer, a sampling step length parameter table pointer and a blood vessel central line structure pointer; based on the annular region of interest structure, performing segmentation and mask generation of blood vessels and surrounding tissues, extraction of image histology characteristics and coupling characteristic construction processing, and generating an effective characteristic set structure; based on the effective feature set structure, domain correction processing, recognition model train