Search

CN-121982262-A - Four-dimensional space-time heart map analysis method and application thereof

CN121982262ACN 121982262 ACN121982262 ACN 121982262ACN-121982262-A

Abstract

The invention discloses a four-dimensional space-time heart map analysis method and application thereof. Specifically, the invention provides a construction method of a four-dimensional space-time (4D) heart map, and a cardiovascular disease diagnosis and risk prediction model constructed based on the 4D heart map constructed by the method. The model of the invention can provide more accurate and efficient diagnosis and prediction results of cardiovascular diseases.

Inventors

  • WANG CHENGYAN
  • WANG GUANGMING

Assignees

  • 上海国际人类表型组研究院

Dates

Publication Date
20260505
Application Date
20260409

Claims (10)

  1. 1. A method for constructing a four-dimensional spatiotemporal (4D) heart map, comprising the steps of: (s 1) providing cardiac magnetic resonance imaging (CMR) data, the data comprising a cardiac short axis magnetic resonance image; (s 2) image automatic segmentation and anatomy extraction, comprising the steps of time sequencing the short axis magnetic resonance image of the heart to enable the short axis magnetic resonance image to accord with a complete cardiac cycle, converting the sequenced image file into a standard format file which can be processed by a computer, inputting the converted standard format file into a pre-trained depth convolution neural network based on nnU-Net architecture, thereby realizing full-automatic segmentation of left ventricle, right ventricle and myocardial structure, and converting two-dimensional image pixel information into a space mask with anatomical significance; (s 3) generating and controlling the quality of the heart three-dimensional grid model, which comprises the steps of converting the output of the segmented mask into three-dimensional heart grid representation, smoothing the grid, and removing low-quality samples with incomplete segmentation or topological errors through an automatic quality control flow; (s 4) spatial rigidity alignment and standardization of population morphology, which comprises the steps of adopting an iterative nearest point algorithm to carry out spatial standardization processing on the heart grids of all individuals generated in (s 3) so that all heart models are under a unified standard coordinate system; (s 5) 3D heart map construction, which comprises the steps of carrying out accurate non-rigid registration on a heart grid at the end diastole by utilizing a Deformetrica method under a large-scale variant differential stratospheric mapping frame, constructing a group average heart template at the End Diastole (ED) by jointly optimizing the group average shape and differential stratospheric deformation mapping from the average shape to each individual heart shape, and simultaneously obtaining deformation field parameters describing individual morphological differences, thereby constructing a 3D heart statistical map at the end diastole; (s 6) constructing a 4D heart dynamic map, which comprises the steps of taking the end diastole 3D heart map constructed in the step (s 5) as a unified reference template, and carrying out non-rigid registration on heart morphology of different time phases in the same heart cycle under the condition of keeping the spatial distribution of control points fixed, so as to establish a multi-temporal three-dimensional heart model sequence with consistent topological correspondence in the whole heart cycle, thereby constructing and obtaining the four-dimensional space-time heart map.
  2. 2. The method of claim 1, comprising the substep of, in step (s 5), converting the vertex coordinates X into a vector field X (X) by optimizing a set of control points q and their corresponding momentums μ: (1) Wherein, the (2) Wherein X (X) represents a vector field generated by transformation and is used for mapping an original vertex coordinate X to a target shape, X is the vertex coordinate of a heart grid model, p is the total number of control points, K represents a kernel and is used for calculating the interaction weight between the vertex X and the control point q; Wherein, the formula (2) is a Gaussian kernel function for defining the smoothness of the deformation, y is a position variable in the kernel function, y corresponds to the position of the control point q in the formula (2), and sigma is a scale parameter of the kernel function, which determines the influence range of the deformation, namely the smoothness of the deformation in space.
  3. 3. The method of claim 1, comprising, in (s 6), the sub-steps of: (s 6 a) estimating a set of differential coherent transformations by phase non-rigid registration of the time points t of each cardiac cycle So that each phase surface S t is aligned with the reference phase S 0 : (3); (s 6 b) population-averaging the aligned shapes of all subjects at the same time phase, the population-averaged shape at time t being defined as: (4) wherein N refers to the number of subjects.
  4. 4. A method of constructing a cardiovascular disease diagnostic model based on a four-dimensional spatiotemporal cardiac atlas constructed by the method of claim 1, comprising the steps of: (Z1) multi-dimensional depth phenotype extraction, comprising decoupling two types of key cardiac phenotypes from the four-dimensional space-time cardiac atlas, namely a morphological phenotype and a motor phenotype, wherein the morphological phenotype is a group heart morphological variation mode, and the motor phenotype is a dynamic track feature set reflecting heart contraction and relaxation functions; (Z2) space-time feature extraction, which comprises the steps of constructing a PointNet-based deep learning frame based on the step (Z1), directly encoding unordered 3D point clouds of each phase of a heart, extracting high-level space features, introducing a time aggregation module to integrate a cross-phase motion rule, and thus obtaining unified 4D space-time feature representation simultaneously containing heart morphology information and motion information; (Z3) constructing a model, namely respectively inputting the 4D space-time characteristics into a LightGBM classifier and a logistic regression model for training and testing, and selecting or parallelly outputting a model with optimal performance as a cardiovascular disease diagnosis model; The cardiovascular disease diagnosis model is used for diagnosing whether a subject currently suffers from a specific cardiovascular disease.
  5. 5. The method of claim 4, wherein the cardiovascular disease is selected from the group consisting of chronic ischemic heart disease (CHID), atrial Fibrillation (AF), angina Pectoris (AP), acute Myocardial Infarction (AMI), cardiac complications and undefined heart disease (I51-HD), other Cardiac Arrhythmias (OCA), or combinations thereof.
  6. 6. A method of constructing a cardiovascular disease risk prediction model based on a four-dimensional spatiotemporal (4D) heart map constructed by the method of claim 1, comprising the steps of: (Z1) multidimensional depth phenotype extraction, comprising decoupling two key cardiac phenotypes from the 4D cardiac atlas, namely a morphological phenotype and a motor phenotype, wherein the morphological phenotype is a normal variation mode of cardiac morphology in a population, and the motor phenotype is a dynamic track feature set reflecting systolic and diastolic functions; (Z2) space-time feature extraction, which comprises the steps of constructing a PointNet-based deep learning frame based on the step (Z1), directly encoding unordered 3D point clouds of each phase of a heart, extracting high-level space features, introducing a time aggregation module to integrate a cross-phase motion rule, and thus obtaining unified 4D space-time feature representation simultaneously containing heart morphology information and motion information; The method comprises the steps of (Z3) constructing a model, namely inputting a result of whether the corresponding cardiovascular disease occurs in a future preset time window or not of an individual corresponding to the 4D space-time characteristics and the 4D space-time characteristics into a machine learning model for training and testing, so as to construct and obtain a cardiovascular disease risk prediction model; the cardiovascular disease risk prediction model is used for predicting whether the risk of the specific cardiovascular disease occurs in a future preset time window of the object to be detected.
  7. 7. A device for cardiovascular disease diagnosis or risk prediction, comprising: (a) An input module configured to input 4D spatiotemporal features of an object to be tested; (b) An evaluation module configured to receive the 4D spatiotemporal features and input the 4D spatiotemporal features into a cardiovascular disease diagnosis model constructed by the method of claim 4 or a cardiovascular disease risk prediction model constructed by the method of claim 6 for diagnosis or risk prediction, and obtain an evaluation result of whether the subject to be tested has a specific cardiovascular disease or whether a risk of the specific cardiovascular disease occurs within a future preset time window; (c) And outputting the evaluation result.
  8. 8. An electronic device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the method of claim 1,4 or 6 when executing the computer program.
  9. 9. A computer readable storage medium storing a computer program, characterized in that the computer program when executed by a processor implements the method of claim 1, 4 or 6.
  10. 10. A computer program product comprising computer-executable instructions or a computer program, which, when executed by a processor, implements the method of claim 1,4 or 6.

Description

Four-dimensional space-time heart map analysis method and application thereof Technical Field The invention belongs to the technical field of image processing, and particularly relates to a four-dimensional space-time heart map analysis method and application thereof. Background Cardiovascular disease is the leading cause of death in the global population, and early accurate diagnosis and risk assessment have great clinical significance. Cardiac magnetic resonance imaging (CMR) has become the gold standard for assessing cardiac structure and function due to its excellent soft tissue resolution and multiparameter imaging capabilities. The heart is a complex four-dimensional dynamic organ that contains rich pathophysiological information in terms of spatial deformation and temporal evolution over the cardiac cycle. However, how to systematically quantify and extract spatiotemporal phenotypes with high discriminant from massive amounts of CMR image data is a key technical bottleneck faced in the current medical image processing and auxiliary diagnosis fields. In the prior art, although a certain progress is made in the aspects of statistical analysis of heart structures, calculation of local function indexes, prediction of image-level diseases and the like, the method still mainly stays on the three-dimensional static modeling or local and low-dimensional dynamic analysis level. There is a lack of a unified four-dimensional cardiac modeling and phenotype extraction technical framework that can simultaneously compromise anatomical consistency, temporal continuity, and deep learning availability at a large population level. In addition, in the process of constructing a group-level four-dimensional heart map based on heart magnetic resonance images and being used for disease prediction, the following key technical problems and implementation difficulties are faced in the field: (1) In the presence of significant non-rigid periodic motion of the heart, how to achieve spatiotemporal consistent alignment across individual, across cardiac phases in a large scale sample. The heart is subjected to complex three-dimensional deformation in the contraction and relaxation processes, different individuals have obvious differences in terms of anatomical structures and heart rate and time phase division, and if a stable space-time alignment mechanism is lacked, the same anatomical position is difficult to ensure that the same anatomical position has comparability between different individuals and different time phases, so that population statistics results are distorted. (2) On the basis of uniform alignment, the static anatomical difference and dynamic motion characteristics of the heart are simultaneously described, and the mutual interference of two types of information in high-dimensional representation is avoided. The traditional method usually only focuses on a single time phase or adopts an integral index to describe dynamic change, and is difficult to carry out decoupling modeling on heart structure reconstruction and abnormal functions, so that the discrimination capability and clinical interpretation of the extracted phenotype are limited. (3) How to efficiently model and learn the high-dimensional trajectory data formed by large-scale cardiac anatomical landmarks over a complete cardiac cycle. The data has the characteristics of high dimensionality, remarkable noise accumulation, complex time sequence correlation and the like, feature redundancy is easily introduced or training is unstable by directly adopting a conventional deep learning model, and the data is difficult to keep enough sensitivity to tiny but clinically significant motion anomalies. Therefore, there is a need in the art to develop a population-level four-dimensional heart statistical map suitable for ultra-large-scale population, and to realize stable alignment and comparable analysis of heart structures under different individual and different cardiac phases through a unified space-time reference frame. Disclosure of Invention The invention provides a population-level four-dimensional heart statistical map suitable for ultra-large-scale population (more than fifty thousand cases), and stable alignment and comparable analysis of heart structures under different individuals and different concentric phases are realized through a unified space-time reference frame. The invention also aims to realize unified quantitative modeling of heart anatomy morphological characteristics and motion track characteristics in a heart cycle on the basis of the four-dimensional heart map, and respectively represent static structure differences and dynamic function changes in a characteristic decoupling mode. The invention also aims to form a space-time depth phenotype system with good universality and expandability, so that the extracted heart phenotype can be reused in different sample scales, different disease types and different downstream tasks. The invention also