CN-122023393-A - CT scanning phase intelligent identification method and system based on organ segmentation

CN122023393ACN 122023393 ACN122023393 ACN 122023393ACN-122023393-A

Abstract

The application relates to the technical field of medical image processing, in particular to an intelligent phase identification method and system for CT scanning based on organ segmentation, wherein the method comprises the steps of carrying out standardization processing on a CT enhanced scanning sequence to obtain standardized image data, inputting the standardized image data into a preset segmentation model, determining an organ mask dictionary, and obtaining a corresponding sampling strategy according to the organ mask dictionary; cutting the key organ structure according to the sampling strategy to obtain a corresponding region of interest, extracting numerical characteristics according to the regions of interest of all the key organ structures, carrying out characteristic interaction on the numerical characteristics after updating to construct a phase characteristic vector, and inputting the phase characteristic vector into a preset rule for judgment to obtain comprehensive multi-source information, wherein the comprehensive multi-source information represents a standardized phase with a phase label. The application can accurately identify the phase based on the image content only, and is not influenced by tag deletion or error.

Inventors

XIE YINAN
ZHANG HAORUI
Liang Luxia

Assignees

杭州英放生物科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260408

Claims (10)

1. The intelligent phase identification method for CT scanning based on organ segmentation is characterized by comprising the following steps: Receiving a CT enhanced scanning sequence, carrying out standardization processing on the CT enhanced scanning sequence to obtain standardized image data, inputting the standardized image data into a preset segmentation model, determining an organ mask dictionary, and obtaining a corresponding sampling strategy according to the organ mask dictionary; Cutting the key organ structures according to the sampling strategy to obtain corresponding regions of interest, extracting numerical characteristics according to the regions of interest of all the key organ structures, and carrying out characteristic interaction on the updated numerical characteristics to construct phase characteristic vectors; And inputting the phase characteristic vector into a preset rule for judgment to obtain comprehensive multi-source information, wherein the comprehensive multi-source information represents a standardized phase with a phase label.
2. The method for intelligently identifying phases of CT scan based on organ segmentation according to claim 1, wherein the key organ structure comprises liver organ, the organ mask dictionary comprises a liver segmentation mask, and the updated numerical features are subjected to feature interaction to construct phase feature vectors, wherein the numerical feature updating mode comprises the following steps: Acquiring corrosion parameters based on a sampling strategy determined by the liver segmentation mask, and performing shrinkage processing on the liver segmentation mask according to the corrosion parameters to acquire a corrosion liver segmentation mask; Resampling the region of interest according to the liver segmentation mask to obtain corresponding sampling voxels, wherein each sampling voxel is used for correspondingly extracting HU values; And removing sampling voxels with HU values larger than a preset threshold value in the corroded liver segmentation mask to update the numerical characteristics of the liver organ.
3. The method of intelligent identification of the phase of a CT scan based on organ segmentation according to claim 2, wherein resampling the region of interest according to the liver segmentation mask to obtain corresponding sampled voxels comprises the steps of: randomly generating a plurality of sampling areas for the region of interest, and judging whether sampling voxels in the sampling areas are larger than a preset voxel threshold value or not; If the sampling voxels of the sampling region are larger than a preset voxel threshold, judging that the sampling is successful, and removing the region of interest; And if the sampling voxels of the sampling region are not greater than the preset voxel threshold, judging that the sampling is invalid, and repeating the random sampling of the region of interest.
4. The method for intelligently identifying phases of CT scan based on organ segmentation according to claim 2, wherein the standardized image data is input to a preset segmentation model, and an organ mask dictionary is determined, comprising the steps of: loading standardized image data, and re-splicing voxels of the standardized image data to obtain each voxel to be processed; Obtaining organ categories corresponding to the voxels to be processed based on a preset segmentation model, and judging whether key organ structures are completely identified according to the detected organ categories; If yes, generating a mask extraction signal, and determining an organ mask dictionary corresponding to the key organ structure based on the mask extraction signal and the organ category; If not, an organ processing signal is generated, and the normalized image data is marked as a low quality image based on the organ processing signal, and a rough region of interest mask is generated based on the organ processing signal.
5. The method for intelligently identifying phases of CT scan based on organ segmentation according to claim 1, wherein the integrated multi-source information includes a phase tag and a confidence parameter, the preset rule judgment includes a rule engine and a nonlinear discrimination model, and the phase feature vector is input to the preset rule judgment to obtain the integrated multi-source information, comprising the steps of: Preliminary judgment is carried out on the phase feature vector based on the rule engine so as to obtain a phase identification type corresponding to the key organ and a corresponding rule confidence coefficient, and whether the phase identification type needs to be updated is judged according to the rule confidence coefficient; Generating a model calling signal when the phase identification type is judged to need to be updated, and calling a nonlinear discrimination model based on the model calling signal to calculate corresponding training fusion parameters; Acquiring metadata analysis time sequence parameters according to the standardized image data, and carrying out weighted fusion on the rule confidence coefficient, the training fusion parameters and the metadata analysis time sequence parameters according to preset weights so as to acquire a phase label and the confidence coefficient parameters; And when the phase identification type is judged to be not required to be updated, the phase identification type is directly identified to obtain a phase label.
6. The method for intelligently identifying phases of CT scan based on organ segmentation according to claim 5, wherein the metadata analysis timing parameters include metadata timing confidence, and further comprising the steps of, before fusing the rule confidence, training fusion parameters, and metadata analysis timing parameters according to preset weights to obtain confidence parameters: comparing the metadata time sequence confidence with a preset time sequence threshold value, and judging whether the metadata time sequence confidence is larger than the preset time sequence threshold value or not; If yes, generating a result output signal, taking the training fusion parameter as a decision result based on the result output signal, and outputting a phase label and a confidence coefficient parameter according to the training fusion parameter.
7. The method for intelligently identifying phases of CT scan based on organ segmentation according to claim 5, further comprising the steps of, after weighting and fusing the rule confidence, training fusion parameters, and metadata analysis timing parameters according to preset weights: comparing the final confidence coefficient with a preset fusion threshold value according to the fused final confidence coefficient, and judging whether the final confidence coefficient is larger than the preset fusion threshold value or not; If yes, determining a phase label and a confidence coefficient parameter based on the final confidence coefficient; If not, generating an artificial review signal, and marking the CT enhanced scanning sequence based on the artificial review signal.
8. The method for intelligent identification of phase of CT scan based on organ segmentation according to claim 5, wherein invoking a nonlinear discrimination model based on the model invoking signal to calculate corresponding training fusion parameters comprises the steps of: acquiring an inspection type, and determining whether to start auxiliary judgment based on the inspection type; If the auxiliary judgment is started, optimizing the phase prediction based on the time sequence association, and judging whether the phase prediction sequence belongs to a reasonable sequence or not; if the phase prediction sequence belongs to a reasonable sequence, the confidence coefficient is improved; if the phase prediction order does not belong to the reasonable order, the confidence is corrected.
9. The intelligent phase identification method for CT scan based on organ segmentation according to claim 8, wherein the step of performing a preset rule judgment based on the phase identification feature to obtain comprehensive multi-source information comprises the steps of: a special organ identification strategy is determined based on the examination type and the phase identification feature is updated based on the special organ identification strategy.
10. A phase intelligent identification system for CT scanning based on organ segmentation, wherein the phase intelligent identification method for CT scanning based on organ segmentation according to any one of claims 1 to 9 is performed, comprising: The preprocessing module is used for receiving the CT enhanced scanning sequence and carrying out standardized processing on the CT enhanced scanning sequence so as to obtain standardized image data; the organ segmentation extraction module inputs the standardized image data into a preset segmentation model, determines an organ mask dictionary, and acquires a corresponding sampling strategy according to the organ mask dictionary; the phase characteristic processing module is used for extracting numerical characteristics based on the interested areas according to all key organ structures, and carrying out characteristic interaction on the updated numerical characteristics so as to construct a phase characteristic vector; And the prediction fusion module inputs the phase characteristic vector to a preset rule judgment to acquire comprehensive multi-source information, wherein the comprehensive multi-source information characterizes a standardized phase with a phase label.

Description

CT scanning phase intelligent identification method and system based on organ segmentation Technical Field The application relates to the technical field of medical image processing, in particular to an intelligent phase identification method and system for CT scanning based on organ segmentation. Background CT enhanced scanning is an important means of clinical diagnosis, scanning is performed at different time points by intravenous injection of Contrast agent, and images of different phases are obtained, including panning (Non-Contrast, NC), arterial phase (ARTERIALPHASE, AP), portal venous phase (PortalVenousPhase, PVP), and delay phase (DELAYEDPHASE, DP). Panning (NC) refers to a baseline scan without Contrast agent injection, arterial phase (ARTERIALPHASE, AP) refers to 25-35 seconds after Contrast agent injection, arterial vessel and rich blood supply organ enhancement is evident, portal phase (PortalVenousPhase, PVP) refers to 60-70 seconds after injection, portal system and parenchymal organ enhancement peak, delay phase (DELAYEDPHASE, DP) refers to 3-5 minutes after injection, contrast agent is gradually excreted, and parenchymal/equilibration phase refers to extra phase of a specific organ (e.g., kidney). Different phases correspond to different clinical diagnostic values, such as rapid enhancement of liver tumor in arterial phase, enhancement decline in portal vein phase, and different layered structures of kidney in cortical phase and parenchymal phase. In clinical application and scientific research projects, accurate identification of CT scanning phases faces the following challenges that DICOM labels are not standard or missing and multi-center data integration is difficult, names of phases are not uniform (such as 'arterial' vs 'arterial phase' vs 'A phase' vs 'enhancement early stage') for different hospitals and different equipment manufacturers, partial equipment only records phase information in private labels, general software cannot analyze, phase label errors or missing are caused by technician misoperation, follow-up automatic flow is influenced, seriesDescription fields usually contain free text formats, and phase information is difficult to accurately extract by rules. Clinical research projects often relate to a plurality of hospitals, the phase naming standards are different, the time and effort are consumed for manually checking the phases one by one, subjective judgment errors exist, a unified phase normalization method is lacking, and the quality of image data and research conclusion are uneven. In addition, the traditional method relies on keyword matching of SeriesDescription fields (if 'arterial' is included, arterial period is judged), the rule method needs to maintain a huge keyword library and complex judgment logic, and is poor in robustness and easy to misjudge when facing a new naming mode or error labeling. The method can infer the phases according to AcquisitionTime sequences in theory, but actual scanning can cause time sequence confusion due to factors such as patient coordination degree, equipment faults and the like, partial inspection is performed to supplement and sweep a certain phase on different dates, and the time stamp cannot reflect the real phase sequence. The quality control rule cannot be correctly applied due to the phase identification error (such as three-stage integrity check of liver), noise data can be introduced by the phase label error, the model performance is reduced, and misdiagnosis or missed diagnosis can be caused by the phase confusion. Therefore, there is a strong need for an intelligent method that can automatically identify the phase based on the content of the image itself, independent of DICOM tags. Disclosure of Invention In order to solve the problems that a phase label is not standard and is difficult to automatically identify and the like in the prior art, the application provides an intelligent phase identification method and system for CT scanning based on organ segmentation. In a first aspect, the present application provides a method for intelligently identifying phases of CT scan based on organ segmentation, which adopts the following technical scheme: A CT scanning phase intelligent identification method based on organ segmentation comprises the following steps: Receiving a CT enhanced scanning sequence, carrying out standardization processing on the CT enhanced scanning sequence to obtain standardized image data, inputting the standardized image data into a preset segmentation model, determining an organ mask dictionary, and obtaining a corresponding sampling strategy according to the organ mask dictionary; Cutting the key organ structures according to the sampling strategy to obtain corresponding regions of interest, extracting numerical characteristics according to the regions of interest of all the key organ structures, and carrying out characteristic interaction on the updated numerical characteristics to construct p