CN-122023375-A - Cardiac MRI image dynamic segmentation and function evaluation system

CN122023375ACN 122023375 ACN122023375 ACN 122023375ACN-122023375-A

Abstract

The invention discloses a dynamic segmentation and function evaluation system for a cardiac MRI image, and relates to the technical field of image semantic segmentation. Firstly, dividing a dynamic sequence based on a U-Net model integrating a Monte Carlo Dropout strategy to generate a division outline, an uncertainty heat map and a global quality evaluation score, secondly, starting a correction flow when an interaction condition is met, receiving a user instruction through an uncertainty perception brush, updating the outline by using a deformation model and recalculating the quality score, thirdly, extracting morphological and dynamic characteristics to calculate cardiac function parameters, and finally, carrying out reliability correction on the function parameters through Monte Carlo sampling and weighting functions based on the quality score to generate a reliability interval. The invention realizes quantification of segmentation uncertainty and credibility transfer of functional evaluation results, and improves reliability and clinical applicability of cardiac MRI analysis.

Inventors

YANG LINA
WANG XI
ZHANG MING
LI DA
JIANG LILI
YANG JING
CHU YUE
TANG SUMEI

Assignees

中国人民解放军总医院第一医学中心

Dates

Publication Date: 20260512
Application Date: 20260212

Claims (7)

1. A system for dynamic segmentation and functional assessment of cardiac MRI images, the system comprising: The initial contour segmentation module is used for processing the MRI dynamic sequence image based on a preset deep learning segmentation model to obtain a segmentation contour of a heart structure and a segmentation certainty degree of each contour point, generating an overall uncertainty heat map and calculating to obtain an initial global quality evaluation score; The interactive correction module is used for starting an interactive correction flow when the uncertainty heat map and the initial global quality assessment score meet preset interaction conditions, receiving a correction instruction of a user, updating the segmentation contour and the uncertainty heat map in real time, and meanwhile, obtaining the contour fitness of the segmentation contour and updating the initial global quality assessment score; The functional parameter acquisition module is used for extracting morphology and dynamics characteristics of the updated segmentation contour, inputting the morphology and dynamics characteristics into the preset functional parameter calculation module and acquiring N functional parameters of the heart; and the parameter interval limiting module is used for carrying out reliability correction on the N calculated functional parameters based on the updated global quality evaluation score to obtain reliability intervals of the N functional parameters.
2. The system for dynamic segmentation and functional assessment of cardiac MRI images according to claim 1, wherein processing the MRI dynamic sequence images based on a predetermined deep learning segmentation model to obtain segmentation contours of cardiac structures and segmentation certainty of each contour point, generating an overall uncertainty heat map, and calculating to obtain an initial global quality assessment score comprises: establishing a deep learning segmentation model based on a U-Net architecture, and starting a Monte Carlo Dropout strategy in a model reasoning stage; Carrying out forward propagation reasoning for T times on each frame of image in the input MRI dynamic sequence to obtain T segmentation probability diagrams; calculating the average value of T times of prediction results of each pixel point in the segmentation probability map as the final class probability of the corresponding pixel point, and generating a final segmentation contour through threshold processing; for each pixel point, calculating the standard deviation of the T times of prediction results, as the segmentation certainty of the corresponding pixel point, and forming the uncertainty heat map based on the segmentation certainty of all the pixel points; And (3) counting the average segmentation certainty degree of all pixel points in the segmentation contour area, and normalizing the average segmentation certainty degree to be used as an initial global quality evaluation score.
3. The cardiac MRI image dynamic segmentation and functional assessment system according to claim 2, wherein the U-Net architecture based deep learning segmentation model comprises: Constructing a U-Net deep learning segmentation model of an encoder-decoder structure; the input of the deep learning segmentation model is a single-frame image of a time point in the preprocessed MRI dynamic sequence; the output of the deep learning segmentation model is a segmentation probability map which has the same size as the input image and comprises a plurality of channels, wherein each channel corresponds to a heart structure type.
4. The cardiac MRI image dynamic segmentation and functional assessment system according to claim 1, wherein when the uncertainty heat map and the initial global quality assessment score satisfy a preset interaction condition, an interactive correction procedure is started, a correction instruction of a user is received, and a segmentation contour and the uncertainty heat map are updated in real time, and a contour fitness of the segmentation contour is obtained, and the initial global quality assessment score is updated, including: presetting an uncertainty threshold, a quality assessment score threshold and the maximum number X of highlight areas; When the initial global quality evaluation score is lower than a quality evaluation score threshold or the number of continuous areas exceeding an uncertainty threshold in the uncertainty heat map is larger than X, judging that a preset interaction condition is met; In the uncertainty heat map, identifying the first X continuous areas with uncertainty values exceeding the uncertainty threshold as highlighting prompt areas; Receiving contour correction operation of a user in a highlight prompt area or a non-highlight prompt area based on an uncertainty perception brush tool, wherein the response weight of a model is higher than that of the non-highlight prompt area when the correction operation is carried out in the highlight prompt area, and the correction operation in the non-highlight prompt area triggers a confirmation prompt; Updating the segmentation contours of the current frame and the adjacent time frames in real time through a deformation model according to contour correction operation of a user, and recalculating an uncertainty heat map of a contour correction area; Calculating the contour matching degree of the segmentation contour based on the difference degree of the corrected segmentation contour and the segmentation contour predicted by the model, wherein the smaller the difference degree is, the higher the contour matching degree is; And re-calculating the average segmentation certainty degree by combining the contour fitness degree and the updated uncertainty heat map, and updating the initial global quality evaluation score, wherein the contour fitness degree is in direct proportion to the updated global quality evaluation score.
5. The cardiac MRI image dynamic segmentation and functional assessment system according to claim 1, wherein extracting morphology and dynamics features of the updated segmentation contour comprises: Calculating the ventricular cavity volume of each time frame based on the ventricular cavity segmentation contour of each time frame, and connecting the volume values of all the time frames to generate a ventricular cavity volume curve; Calculating the volume of the cardiac muscle based on the epicardial contour and the endocardial contour in the end diastole frame, and multiplying the volume by the cardiac muscle density to obtain the cardiac muscle quality; dividing a plurality of segments along the ventricular wall in end diastole and end systole frames, and calculating the wall thickness and the change rate of each segment; The ventricular chamber volume curve, myocardial mass, chamber wall thickness and their rate of change are taken as morphology and kinetic characteristics of the updated segmented contours.
6. The cardiac MRI image dynamic segmentation and functional assessment system as set forth in claim 5, wherein the morphology and dynamics feature is input to a preset functional parameter calculation module to obtain N functional parameters of the heart, comprising: the N functional parameters at least comprise ejection fraction and stroke volume; extracting a maximum value from the ventricular cavity volume curve as an end diastole volume and extracting a minimum value as an end systole volume; Based on the end diastole volume and the end systole volume, the ejection fraction and stroke volume are calculated.
7. The cardiac MRI image dynamic segmentation and functional assessment system according to claim 1, wherein performing reliability correction on the calculated N functional parameters based on the updated global quality assessment score to obtain reliability intervals for the N functional parameters comprises: Generating M sets of possible contour variants conforming to an uncertainty distribution by monte carlo sampling based on the uncertainty heat map; Recalculating all N functional parameters of each group of profile variants; For each functional parameter, forming a statistical distribution by corresponding M calculation results, taking the P% quantile to (100-P)% quantile of the statistical distribution as an initial reliability interval of the functional parameter, and obtaining N initial reliability intervals of N functional parameters, wherein P is more than 0 and less than 100; And carrying out weighted correction on the widths of the N initial reliability intervals by using the updated global quality evaluation score to obtain N reliability intervals of N functional parameters, wherein the higher the updated global quality evaluation score is, the narrower the corrected reliability interval width is.

Description

Cardiac MRI image dynamic segmentation and function evaluation system Technical Field The invention relates to the technical field of image semantic segmentation, in particular to a dynamic segmentation and function evaluation system for cardiac MRI images. Background In the field of medical image analysis, cardiac Magnetic Resonance Imaging (MRI) is an important examination means for acquiring structural and functional information of the heart by virtue of its non-invasive nature and excellent soft tissue resolution. The heart structure in the image is accurately segmented, and functional evaluation is completed based on segmentation results, so that the method is a core technical link for assisting in diagnosing various cardiovascular diseases. However, existing solutions present significant challenges in this procedure. At present, analysis of heart structures mainly depends on two modes, namely, a doctor performs full manual segmentation, the method is time-consuming in process and low in efficiency, segmentation accuracy is greatly limited by subjective experience of operators, and repeatability and consistency of results are difficult to guarantee. The second is to divide by using the full-automatic deep learning model, although the speed is improved, the inherent black box characteristic of the model causes the decision process to lack transparency, and particularly, the model does not have the quantization capability for the reliability of the self-dividing result. This makes it still necessary for the clinician to perform a tedious full-disk manual review of the automatic segmentation results in order to secure diagnosis, without substantially freeing up manpower. Further, in the subsequent functional evaluation stage, the confidence of the calculated critical functional parameters such as ejection fraction, ventricular volume, etc. is doubtful due to the fact that the inherent uncertainty existing in the early segmentation result cannot be perceived and quantified, and finally the accuracy of clinical decision may be affected. Disclosure of Invention Aiming at the technical problems of low efficiency and insufficient reliability caused by the fact that the existing cardiac magnetic resonance imaging analysis depends on manual segmentation or an unreliable full-automatic segmentation model, the invention provides a dynamic segmentation and function evaluation system for cardiac MRI images. The technical scheme for solving the technical problems is as follows: the invention provides a heart MRI image dynamic segmentation and function evaluation system, which comprises: The initial contour segmentation module is used for processing the MRI dynamic sequence image based on a preset deep learning segmentation model to obtain a segmentation contour of a heart structure and a segmentation certainty degree of each contour point, generating an overall uncertainty heat map and calculating to obtain an initial global quality evaluation score; The interactive correction module is used for starting an interactive correction flow when the uncertainty heat map and the initial global quality assessment score meet preset interaction conditions, receiving a correction instruction of a user, updating the segmentation contour and the uncertainty heat map in real time, and meanwhile, obtaining the contour fitness of the segmentation contour and updating the initial global quality assessment score; The functional parameter acquisition module is used for extracting morphology and dynamics characteristics of the updated segmentation contour, inputting the morphology and dynamics characteristics into the preset functional parameter calculation module and acquiring N functional parameters of the heart; and the parameter interval limiting module is used for carrying out reliability correction on the N calculated functional parameters based on the updated global quality evaluation score to obtain reliability intervals of the N functional parameters. The beneficial effects of the invention are as follows: Compared with the prior art, the method provided by the invention has the advantages that firstly, the uncertainty in the segmentation process is quantified, the visual thermodynamic diagram and the global quality score are generated, and an explicit basis is provided for evaluating the reliability of the initial segmentation result. And an interactive correction mechanism is introduced, so that a user can be intelligently guided to carry out priority correction on the high uncertainty region, the segmentation result and the quality evaluation are updated in real time, and the efficiency and pertinence of man-machine interaction are improved. And the segmentation quality is associated with the subsequent function parameter calculation again, so that the data base from morphological characteristics to the dynamic parameter extraction process is ensured to be more reliable. And finally, carrying out reliability correction on th