CN-122023919-A - Processing method based on heart three-dimensional image

CN122023919ACN 122023919 ACN122023919 ACN 122023919ACN-122023919-A

Abstract

The invention discloses a processing method, device, equipment and medium based on a heart three-dimensional image. The method comprises the steps of obtaining a heart three-dimensional image and a chamber segmentation result to obtain a left ventricle heart chamber and heart muscle, constructing a three-dimensional inner diameter curved surface according to the inner diameter and the outer diameter of the heart muscle and generating heart muscle wall thickness mapping, determining a pre-trained classification model and constructing a two-channel input of the model, inputting the curved surface and the mapping to a first channel of the model, combining full heart segmentation or original image data of a second channel, processing the two-channel input through the model, and outputting a heart muscle disease classification result. The invention reserves the complete space shape through the mapping of the three-dimensional inner diameter curved surface and the wall thickness, combines the double-channel feature fusion and the interactive calibration of the space guidance mask, realizes the accurate and robust classification of various cardiomyopathy such as hypertrophic cardiomyopathy, dilated cardiomyopathy and the like, and solves the problems of efficiently and accurately extracting the features from the complex three-dimensional heart image and automatically identifying the diseases.

Inventors

LI CHENGUANG
HUANG XINGSHENG
MA JUN
ZHOU JINYING
GE JUNBO

Assignees

复旦大学附属中山医院

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (13)

1. The processing method based on the heart three-dimensional image is characterized by comprising the following steps of: Acquiring three-dimensional image data of a heart and a three-dimensional segmentation result of a heart chamber, wherein the three-dimensional segmentation result of the heart chamber comprises a left ventricular chamber and a left ventricular myocardium; constructing a three-dimensional inner diameter curved surface and generating a myocardial wall thickness map according to the inner diameter and the outer diameter of the left ventricular myocardium, wherein the myocardial wall thickness map represents the shortest distance from an inner diameter pixel point to the outer diameter; Determining a pre-trained classification model, and constructing a dual-channel input of the pre-trained classification model; a first input channel for mapping the three-dimensional inner diameter surface and the myocardial wall thickness to the pre-trained classification model; processing the two-channel input through the pre-trained classification model, and outputting a myocardial disease classification model result; the dual channel input of the pre-trained classification model includes the first input channel with respect to the three-dimensional inside diameter surface to myocardial wall thickness mapping.
2. The method for processing a three-dimensional image of a heart according to claim 1, wherein the acquiring three-dimensional image data of the heart and three-dimensional segmentation results of a heart chamber comprises: acquiring original medical image data of a heart; Processing the original medical image data through a pre-established segmentation algorithm to generate a heart chamber three-dimensional segmentation result; Extracting boundary structure information of a left ventricular chamber and left ventricular myocardium based on the three-dimensional segmentation result of the heart chamber; And determining the inner diameter and the outer diameter of the left ventricular myocardium according to the boundary structure information, wherein the inner diameter is the juncture of the left ventricular myocardium and the left ventricular chamber, and the outer diameter is the juncture of the left ventricular myocardium and the right ventricle or the background.
3. The method of processing a three-dimensional image of the heart of claim 2, wherein determining the inner and outer diameters of the left ventricular myocardium from the boundary structure information comprises: extracting the centroid of the left ventricle base as a first reference point according to the boundary structure information; Determining a point in the heart chamber of the left ventricle, which is farthest from the first reference point, as a second reference point; Constructing a long axis of the left ventricle from the first reference point and the second reference point; Determining the edge of the left ventricular myocardium at the left ventricular chamber side as an inner diameter according to the boundary structure information; The edge of the left ventricular myocardium distinguished from the right ventricle and the background is determined as the outer diameter.
4. The method of processing a three-dimensional image of the heart of claim 1, wherein constructing a three-dimensional inside diameter surface and generating a myocardial wall thickness map from the inside diameter and outside diameter of the left ventricular myocardium comprises: Traversing all pixel points of the inner diameter of the left ventricle cardiac muscle, and calculating the shortest distance between each pixel point of the inner diameter and the central distance between each pixel point of the outer diameter; generating a myocardial wall thickness map based on the shortest distance, the myocardial wall thickness map reflecting the spatial distribution of the inside diameter pixel points and the corresponding wall thickness intensity; Constructing a three-dimensional inner diameter curved surface based on the spatial distribution of the inner diameter pixel points, wherein the three-dimensional inner diameter curved surface reserves the original spatial form of left ventricular myocardium; and processing the myocardial wall thickness map and the three-dimensional inner diameter curved surface by applying a preset smoothing algorithm to obtain a smooth three-dimensional inner diameter curved surface and a myocardial wall thickness map.
5. The method of claim 1, wherein the two-channel input of the pre-trained classification model comprises a second input channel for full heart segmentation results or raw medical image data corresponding to full heart segmentation.
6. The method of processing a three-dimensional image of a heart of claim 5, wherein processing the two-channel input through the pre-trained classification model to output a cardiomyopathy classification model result comprises: Based on the spatial distribution of the inner diameter of the left ventricular myocardium, intercepting a three-dimensional space frame containing the inner diameter of the left ventricular myocardium to obtain cut three-dimensional data; The three-dimensional space frame covers the directions of an x axis, a y axis and a z axis of the inner diameter; Processing the cut three-dimensional data and the myocardial wall thickness map by using a self-adaptive average pooling 3d (AdaptiveAvgPool d) operator to generate output data in a uniform format; inputting the output data in the unified format to the pre-trained classification model through the first input channel, and Acquiring a full heart segmentation result or corresponding original medical image data as a second channel input, and inputting the full heart segmentation result or corresponding original medical image data into the pre-trained classification model; the second channel input includes chamber morphology information or texture information; and processing the two-channel input through the pre-trained classification model, and outputting a cardiomyopathy classification model result.
7. The method of claim 6, wherein acquiring the full heart segmentation result or corresponding raw medical image data as a second channel input to the pre-trained classification model, further comprises: Extracting features of the two-channel input through a 3D convolution network to obtain high-dimensional features, wherein the high-dimensional features comprise spatial morphology and wall thickness distribution information of left ventricular myocardium; classifying based on the high-dimensional features to generate a cardiomyopathy classification result; the cardiomyopathy classification result comprises hypertrophic cardiomyopathy, dilated cardiomyopathy, restricted cardiomyopathy, amyloidosis, myocarditis and normal type; And aiming at the myocardial disease classification result, adjusting classification model parameters by applying a preset optimization algorithm to obtain an optimized myocardial disease classification model result so as to output the myocardial disease classification model result.
8. The method for processing a three-dimensional image of a heart of claim 5, wherein the step of processing the two-channel input through the pre-trained classification model and outputting a cardiomyopathy classification model result comprises: Determining the spatial weight of each sampling point according to the deviation degree of each sampling point in the myocardial wall thickness map relative to a preset myocardial wall thickness reference value, and generating a spatial guidance mask matrix M aligned with the spatial scale of the three-dimensional image data of the heart; Inputting the three-dimensional image data of the heart into a second input channel of the classification model, and extracting deep space feature tensor T containing anatomical texture information through multidimensional convolution; Introducing a dynamic adjustment factor beta, and performing space coordinate-by-space coordinate weight modulation on the deep space feature tensor T by utilizing the space guidance mask matrix M so as to enhance the feature intensity corresponding to the wall thickness abnormal region and generate a calibrated space interaction feature map F; And performing space dimension compression on the space interaction feature map F to extract a core pathological feature vector, mapping the core pathological feature vector to a category space, and outputting a cardiomyopathy classification model result.
9. The method for processing a three-dimensional image of the heart of claim 4, wherein constructing a three-dimensional inside diameter surface based on the spatial distribution of inside diameter pixels comprises: generating a three-dimensional grid structure based on the coordinate information of the inner diameter pixel points, wherein the three-dimensional grid structure represents an inner diameter curved surface of left ventricular myocardium; Filling missing curved surface data by applying a preset interpolation algorithm aiming at the three-dimensional grid structure; according to the spatial distribution of the inner diameter pixel points, the geometric shape of the three-dimensional grid structure is adjusted to obtain a three-dimensional inner diameter curved surface with the original spatial shape reserved; and generating corresponding color codes aiming at the three-dimensional inner diameter curved surface, wherein the color codes reflect the spatial position information of the curved surface.
10. The method of claim 6, wherein capturing a three-dimensional spatial frame containing the inside diameter of the left ventricular myocardium based on the spatial distribution of the inside diameter of the left ventricular myocardium, comprises: determining clipping ranges in the directions of an x axis, a y axis and a z axis based on the spatial distribution of the pixel points of the inner diameter; generating a three-dimensional space frame containing an inner diameter for the clipping range, wherein the three-dimensional space frame covers a main distribution area of the inner diameter of the left ventricular myocardium; And cutting the three-dimensional inner-diameter curved surface and myocardial wall thickness mapping based on the three-dimensional space frame to obtain cut three-dimensional data.
11. A processing device based on a three-dimensional image of a heart, comprising: The acquisition module is used for acquiring three-dimensional image data of the heart and a three-dimensional segmentation result of a heart chamber, wherein the three-dimensional segmentation result of the heart chamber comprises a left ventricular chamber and a left ventricular myocardium; The processing module is used for constructing a three-dimensional inner diameter curved surface and generating a myocardial wall thickness map according to the inner diameter and the outer diameter of the left ventricular myocardium, wherein the myocardial wall thickness map represents the shortest distance from an inner diameter pixel point to the outer diameter; The training module is used for determining a pre-trained classification model and constructing a double-channel input of the pre-trained classification model; the training module is further used for mapping and inputting the three-dimensional inner-diameter curved surface and the myocardial wall thickness into a first input channel of the pre-trained classification model; The training module is also used for processing the two-channel input through the pre-trained classification model and outputting a myocardial disease classification model result; the dual channel input of the pre-trained classification model includes the first input channel with respect to the three-dimensional inside diameter surface to myocardial wall thickness mapping.
12. A non-transitory computer readable storage medium having stored thereon computer executable instructions, which when executed by a processor, perform the steps of: Acquiring three-dimensional image data of a heart and a three-dimensional segmentation result of a heart chamber, wherein the three-dimensional segmentation result of the heart chamber comprises a left ventricular chamber and a left ventricular myocardium; constructing a three-dimensional inner diameter curved surface and generating a myocardial wall thickness map according to the inner diameter and the outer diameter of the left ventricular myocardium, wherein the myocardial wall thickness map represents the shortest distance from an inner diameter pixel point to the outer diameter; Determining a pre-trained classification model, and constructing a dual-channel input of the pre-trained classification model; a first input channel for mapping the three-dimensional inner diameter surface and the myocardial wall thickness to the pre-trained classification model; processing the two-channel input through the pre-trained classification model, and outputting a myocardial disease classification model result; the dual channel input of the pre-trained classification model includes the first input channel with respect to the three-dimensional inside diameter surface to myocardial wall thickness mapping.
13. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 10 when the computer program is executed.

Description

Processing method based on heart three-dimensional image Technical Field The present invention relates to the field of information technologies, and in particular, to a method, an apparatus, a device, and a medium for processing a three-dimensional image based on a heart. Background How to accurately and efficiently extract the left ventricular myocardial features from complex three-dimensional heart image data and realize myocardial disease classification is a core technical problem to be solved urgently. The problem is caused by the complexity and diversity of medical images and the remarkable difference of the pathological features of cardiomyopathy types such as hypertrophic type and dilatation type, so that in an actual business scene, the traditional method is difficult to simultaneously consider the accuracy of feature extraction, the accuracy of classification and the high efficiency of processing. Specifically, the original heart image data presents a high-dimensional space structure under a stereoscopic microscope, including noise, individual differences and data heterogeneity brought by imaging equipment, and how to accurately divide the inner diameter and the outer diameter of the left ventricular myocardium to generate high-quality wall thickness distribution mapping and an inner diameter curved surface becomes a primary difficulty. In addition, how to effectively reduce the computational complexity by clipping and feature extraction while maintaining the three-dimensional space morphological information and simultaneously ensure the distinguishing capability of the features on different types of cardiomyopathy is a key contradiction. Further, the classification model is prone to over-fitting or insufficient robustness in the face of small sample data or slight differences in pathological features, affecting the reliability of detection. The sub-problems are commonly directed to a core challenge of how to realize accurate detection of cardiomyopathy in complex and changeable heart image data by efficient segmentation, feature extraction and classification processes, and considering precision, robustness and calculation efficiency. This problem is particularly acute in practical clinical settings, because the accuracy of the detection is directly related to the patient's treatment decisions, while the efficiency affects the distribution and detection speed of medical resources. Disclosure of Invention Based on the existing complex and changeable heart image data, the high-efficiency segmentation, feature extraction and classification processes cannot be realized, the precision, the robustness and the calculation efficiency are taken into account, the accurate detection of the cardiomyopathy is realized, the invention provides a processing method based on the heart three-dimensional image, and the high-efficiency and accurate technical support is provided for the detection of the cardiomyopathy. The invention provides a processing method based on a heart three-dimensional image, which mainly comprises the following steps: Acquiring three-dimensional image data of a heart and a three-dimensional segmentation result of a heart chamber, wherein the three-dimensional segmentation result of the heart chamber comprises a left ventricular chamber and a left ventricular myocardium; constructing a three-dimensional inner diameter curved surface and generating a myocardial wall thickness map according to the inner diameter and the outer diameter of the left ventricular myocardium, wherein the myocardial wall thickness map represents the shortest distance from an inner diameter pixel point to the outer diameter; Determining a pre-trained classification model, and constructing a dual-channel input of the pre-trained classification model; a first input channel for mapping the three-dimensional inner diameter surface and the myocardial wall thickness to the pre-trained classification model; processing the two-channel input through the pre-trained classification model, and outputting a myocardial disease classification model result; the dual channel input of the pre-trained classification model includes the first input channel with respect to the three-dimensional inside diameter surface to myocardial wall thickness mapping. Further, acquiring three-dimensional image data of the heart and three-dimensional segmentation results of the heart chamber includes: acquiring original medical image data of a heart; Processing the original medical image data through a pre-established segmentation algorithm to generate a heart chamber three-dimensional segmentation result; Extracting boundary structure information of a left ventricular chamber and left ventricular myocardium based on the three-dimensional segmentation result of the heart chamber; And determining the inner diameter and the outer diameter of the left ventricular myocardium according to the boundary structure information, wherein the inner diameter is the ju