CN116152253B - Cardiac magnetic resonance mapping quantification method, system and storage medium - Google Patents

Abstract

The invention belongs to the technical field of medical image processing, and particularly relates to a method, a system and a storage medium for quantifying cardiac magnetic resonance mapping. The method comprises the following steps: step 1, inputting original data of a heart magnetic resonance mapping image, respectively acquiring original images of a left ventricular basal layer, a middle layer and a cardiac apex layer, and executing subsequent steps on at least one layer of the original images of the basal layer, the middle layer and the cardiac apex layer; step 2, preprocessing the original image, and determining a general target area of left ventricle cardiac muscle based on contour recognition; step 3, obtaining an accurate target area of the segmentation target through connected domain analysis; step 4, segmenting left ventricular myocardium based on the accurate target region; and 5, calculating the average value of the original image pixels of each segmented region. The invention further provides a system for realizing the method. The method has high accuracy and repeatability of the quantized result and good application prospect.

Description

Cardiac magnetic resonance mapping quantification method, system and storage medium

technical field

The invention belongs to the technical field of medical image processing, and in particular relates to a cardiac magnetic resonance mapping quantification method, system and storage medium.

Background technique

In recent years there has been a steady increase in the use of cardiac MR mapping techniques in the clinical setting, which quantify one or more drivers of MR contrast (i.e., relaxation times) in each pixel of the image. These relaxation times or relaxation parameters are quantifiable properties of tissue in a magnetic field and are strongly dependent on the physiological properties of the tissue. Since the mapped image is no longer dependent on some specific scan parameters (such as RF coil proximity, receiver chain efficiency, or magnetic field inhomogeneity), this gives mapped imaging a huge advantage over qualitative imaging. Quantified relaxation times also reduce inter-observer variability, allow dynamic tracking of tissue parameters across patients, and allow comparison of tissue parameters between individual patients. These mapping images only rely on the interaction of physics and biology, so they are theoretically highly reproducible.

According to the standard post-processing guidelines for cardiac magnetic resonance images released by the Society for Cardiovascular Magnetic Resonance (SCMR) in 2020, the existing post-processing analysis methods for mapping images include visual analysis and manual quantitative analysis. There is no specific recommendation or preferred post-processing package, and analysts should be trained in local standards or the post-processing package of choice. Visual analysis is to observe the general image with the naked eye, aiming to detect image artifacts and verify the quality of diagnostic images, but it cannot obtain quantitative data from mapping images. When measuring global or diffuse cardiomyopathy during manual quantitative analysis, guidelines recommend conservatively delineating a region of interest in the interventricular septum at the short-axis level of the myocardium to reduce the influence of susceptibility artifacts from adjacent tissues. For accurate assessment of diffuse disease, focal lesions should be excluded from the region of interest in the mapping image. For focal lesions, it is recommended to delineate additional ROIs outside the abnormal areas found by visual analysis, and the range of ROIs < 20 pixels should be avoided. In actual clinical application and research, doctors (researchers) often divide the left ventricular myocardium into 16 segments according to the American Heart Association (AHA)-16 segment division method, and measure the whole or segment in the mapping image Myocardial average.

The above existing post-processing methods of cardiac magnetic resonance images mainly have the following problems: the mapping image needs to be imported into a commercial software package dedicated to cardiac imaging, and then the left ventricular endocardium/epicardium and (or) region of interest are manually drawn, and the measurement method consumes There is a certain difference between observers, and the measured data also needs to be output manually, which greatly limits the clinical application and promotion of this technology. If computer software can be used to automatically realize the above segmentation and quantification process, these problems will be effectively solved. However, due to the special anatomical structure of the heart and the interference of image background noise (especially the muscle trabeculae and pericardial tissue adjacent to the endocardium/epicardium), the outline of the target area (such as left ventricular endocardium/epicardium, right ventricle) in cardiac magnetic resonance mapping images The automatic recognition of the insertion part) is still not accurate enough, and there is still a lack of relevant research on how to achieve accurate and fully automatic processing of cardiac magnetic resonance mapping images. There is still an urgent need in this field to develop a method that can automatically segment and quantify cardiac magnetic resonance mapping images by computer methods and systems.

Contents of the invention

In view of the problems in the prior art, the present invention provides a method, system and storage medium for quantifying cardiac magnetic resonance mapping images, so as to achieve the purpose of automatically segmenting and quantifying cardiac magnetic resonance mapping images by computer.

A method for cardiac magnetic resonance mapping quantification, comprising the steps of:

Step 1, input the original data of cardiac magnetic resonance mapping image, obtain the original image of basal layer, middle layer and apical layer respectively, carry out follow-up step for at least one in the original image of described basal layer, middle layer and apical layer;

Step 2, preprocessing the original image, and determining the approximate target area of the left ventricular myocardium based on contour recognition;

Step 3, obtain the precise target area of the segmented target through connected domain analysis;

Step 4, segmenting based on the precise target area;

Step 5, calculate the average value of the original image pixels of each segmented region after segmentation.

Preferably, the format of the original data is a DICOM file.

Preferably, step 2 specifically includes:

Step 2.1, converting the original image into a grayscale image;

Step 2.2, filter background noise and foreground noise;

Step 2.3, draw a reference circle with a radius of 1/4 width with the center point of the image;

Step 2.4, performing contour detection on the entire image, traversing the detected contours, and those whose center of gravity of the contour is within the reference circle are candidate contours;

Step 2.5, obtain the candidate contour with the largest area as the target contour;

Step 2.6, drawing the extreme points of the target contour;

Step 2.7, obtaining the approximate target area according to the extreme points.

Preferably, step 2.2 specifically includes:

Step 2.2.1, filter the background noise by threshold analysis and median filtering;

Step 2.2.2, filter background noise by connected domain analysis;

Step 2.2.3, the image is inverted, and the foreground noise is filtered through connected domain analysis;

Step 2.2.4, perform image inversion again;

The shape of the approximate target area is a rectangle; step 2.7 specifically includes:

Step 2.7.1, taking the extreme left point as the center point c(x ₀ , y ₀ ) of the approximate target area;

Step 2.7.2, using half of the vertical distance between the upper extreme point and the lower extreme point as the reference value d to calculate the upper left point and lower right point of the approximate target area, the calculation formula is:

Coordinates of upper left point: lt(x ₁ , y ₁ ) = (x ₀ - 0.5*d, y ₀ - 0.8*d);

Coordinates of lower right point: rb(x ₂ , y ₂ ) = (x ₀ + r, y ₀ + 0.8*d);

Step 2.7.3, drawing the approximate target area.

Preferably, step 3 specifically includes:

Step 3.1, preprocessing by threshold analysis;

Step 3.2, through connected domain analysis, remove connected domains with small area, leaving at least 3 connected domains as candidate regions;

Step 3.3, locating the left ventricle by contour detection;

Step 3.4, using the convex hull algorithm to locate the anterior insertion of the right ventricle;

Step 3.5, perform image inversion;

Step 3.6, filtering the background noise by at least one method of threshold analysis, median filtering, corrosion, and expansion;

In step 3.7, the connected domain with the largest area is obtained as the precise target area through connected domain analysis.

Preferably, step 4 specifically includes:

Step 4.1, perform contour repair on the left ventricular endocardium and epicardium, the method of contour repair is: fill the fault through morphological operation, and delete the bloat;

Step 4.2, drawing the contour ring of the epicardium;

Step 4.3, based on the center of gravity of the left ventricular chamber and the anterior insertion of the right ventricle, draw star-shaped scattering lines to evenly segment the contour ring;

Step 4.4, performing corrosion operation on each segmented area;

In step 4.5, number the center of gravity of each segmented region in a clockwise direction, starting from the anterior insertion of the right ventricle.

Preferably, in step 4, the original image of the basal layer or the original image of the middle layer is divided into 6 segments, and the original image of the apical layer is divided into 4 segments.

Preferably, step 5 specifically includes:

Step 5.1, obtain all the pixel points of each segment area after segmentation, and color each segment area with different colors;

Step 5.2, mapping the coordinate position of each pixel point to the original image to obtain the pixel value;

Step 5.3, calculating the pixel average value of each region according to the pixel value of the original image of each region.

The present invention also provides a cardiac magnetic resonance mapping image quantification system, comprising:

The input module is used to input the raw data of the cardiac magnetic resonance mapping image;

A calculation module, configured to process the raw data according to the above-mentioned cardiac magnetic resonance mapping image quantification method;

The output module is used to output the processing result of the calculation module.

The present invention also provides a computer-readable storage medium, on which is stored: a computer program for realizing the above-mentioned cardiac magnetic resonance mapping quantification method.

The present invention designs the flow of the method for automatic quantification of cardiac magnetic resonance mapping images for the first time, and can realize the purpose of using a computer to automatically process cardiac magnetic resonance mapping images. It solves the time-consuming problem of manually delineating the left ventricular endocardium/epicardium and manually positioning the interventricular septum insertion part in the current quantitative measurement of myocardial mapping images, and realizes fully automatic measurement; the operation is simple and fast, and the obtained results are reliable and repeatable High reliability, which solves the problem of large differences in manual measurement data caused by different proficiency levels of analysts. Therefore, the present invention has good application prospects.

Apparently, according to the above content of the present invention, according to common technical knowledge and conventional means in this field, without departing from the above basic technical idea of the present invention, other various forms of modification, replacement or change can also be made.

The above-mentioned content of the present invention will be further described in detail below through specific implementation in the form of examples. However, this should not be construed as limiting the scope of the above-mentioned subject matter of the present invention to the following examples. All technologies realized based on the above contents of the present invention belong to the scope of the present invention.

Description of drawings

FIG. 1 is a schematic flow chart of obtaining an approximate target area in Embodiment 1 of the present invention.

FIG. 2 is a schematic flow diagram of obtaining a precise target area in Embodiment 1 of the present invention.

Fig. 3 is a schematic flow chart of contour segmentation and data extraction in Embodiment 1 of the present invention.

Figure 4 shows the original image preprocessing steps. A. DICOM image grayscale image; B. Filter background noise by threshold analysis and median filter; C. Filter background noise by connected domain analysis; D. After image inversion, filter foreground noise by connected domain analysis. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 5 shows the approximate target area obtained according to the extreme points. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 6 shows the candidate regions obtained according to connected domain analysis. A. Invert the image; B. Remove the small-area connected domain; C. Get the candidate area. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 7 shows the location of the anterior insertion of the right ventricle using the convex hull algorithm. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 8 is for filtering background noise. A. Filtering background noise by threshold analysis; B. Filtering background noise by median filtering; C. Filtering background noise by erosion; D. Filtering background noise by dilation. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 9 shows the acquisition of the largest area connected domain as the precise target area. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 10 shows the contour patching of the target area. A. Fill the fault; B. Delete the bloat; C. Draw the contour ring of the left ventricular myocardium. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 11 is the segmented left ventricular myocardium outline ring. A. Draw a star-shaped scattering line; B. Split the contour ring; C. Get the center of gravity of each segment. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Figure 12 is to obtain and color the pixel points of each area. In the figure, the unit of the axis of abscissa and the axis of ordinate is pixel.

Detailed ways

It should be noted that the algorithms for the steps of data collection, transmission, storage and processing not specifically described in the embodiments, as well as the hardware structures and circuit connections not specifically described can be realized by the disclosed content of the prior art.

Example 1 Cardiac magnetic resonance mapping image quantification method and system

This embodiment provides a system for automatically segmenting and quantifying cardiac magnetic resonance mapping images, the system comprising:

The input module is used to input the raw data of the cardiac magnetic resonance mapping image;

Calculation module, used to process the raw data by the cardiac magnetic resonance mapping image quantification method;

The output module is used to output the processing result of the calculation module.

Among them, the workflow of the algorithm in the calculation module is shown in Figure 1-Figure 3, and Figure 4-Figure 12 uses a set of DICOM files of myocardial T1mapping as an example operation, specifically including the following steps:

Step 1, input the original data of the cardiac magnetic resonance mapping image, obtain the original images of the basal layer, the middle layer and the apical layer respectively, and perform subsequent steps for the original images of the basal layer, the middle layer and the apical layer respectively; it should be noted that, in In the subsequent steps, the original image of the basal layer or the original image of the middle layer is divided into 6 segments, and the original image of the apical layer is divided into 4 segments;

specific:

Step 1.1, export the raw data of a set of cardiac magnetic resonance mapping images that need to be measured (the example in this embodiment is a set of myocardial T1 mapping images) in DICOM form and classify them into a folder, including basal layer, middle layer, and apex Layer three DICOM files.

Step 1.2, use the pydicom library to read the Slice Location of each DICOM file, and sort each DICOM file in reverse order through SliceLocation to determine the location of each DICOM (ie, basal layer, middle layer, and apical layer).

Step 2, performing preprocessing on the original image, and determining an approximate target area of the left ventricular myocardium based on contour recognition; the approximate target area is rectangular in shape;

specific:

Step 2.1, converting the original image into a grayscale image;

Step 2.2, filter background noise and foreground noise;

Step 2.2 specifically includes:

Step 2.2.1, filter background noise by threshold analysis, median filter (Figure 4);

Step 2.2.2, filter background noise by connected domain analysis (Figure 4);

Step 2.2.3, the image is inverted, and the foreground noise is filtered through connected domain analysis (Figure 4);

Step 2.2.4, perform image inversion again;

Step 2.3, draw a reference circle with a radius of 1/4 width with the center point of the image;

Step 2.4, performing contour detection on the entire image, traversing the detected contours, and those whose center of gravity of the contour is within the reference circle are candidate contours;

Step 2.5, obtain the candidate contour with the largest area as the target contour;

Step 2.6, drawing the extreme points of the target contour;

Step 2.7, obtaining the approximate target area according to the extreme points (Figure 5);

Step 2.7 specifically includes:

Step 2.7.1, taking the extreme left point as the center point c(x ₀ , y ₀ ) of the approximate target area;

Step 2.7.2, using half of the vertical distance between the upper extreme point and the lower extreme point as the reference value d to calculate the upper left point and lower right point of the approximate target area, the calculation formula is:

Coordinates of upper left point: lt(x ₁ , y ₁ ) = (x ₀ - 0.5*d, y ₀ - 0.8*d);

Coordinates of lower right point: rb(x ₂ , y ₂ ) = (x ₀ + r, y ₀ + 0.8*d);

Step 2.7.3, drawing the approximate target area.

Step 3, obtain the precise target area of the segmented target through connected domain analysis;

specific:

Step 3.1, preprocessing by threshold analysis;

Step 3.2, through connected domain analysis, remove the connected domains with small area, leaving at least 3 connected domains as candidate areas (Figure 6);

Step 3.3, locate the left ventricle through contour detection; specifically: obtain the contours of the first three connected domains with the largest areas in the candidate area, and then locate the left ventricle through the extreme lower point and extreme left point of the connected domain;

Step 3.4, using the convex hull algorithm to locate the anterior insertion of the right ventricle (Figure 7);

Step 3.5, perform image inversion;

Step 3.6, filter background noise by threshold analysis, median filter, erosion, dilation (Figure 8);

In step 3.7, the connected domain with the largest area is obtained through connected domain analysis as the precise target area (Figure 9).

Step 4, segmenting based on the precise target area;

specific:

Step 4.1, performing contour repair on the left ventricular endocardium and epicardium, the method of contour repair is: filling faults through morphological operations and deleting bloat (Figure 10);

Step 4.2, drawing the contour ring of the left ventricular myocardium (Figure 10);

Step 4.3, based on the center of gravity of the left ventricular chamber and the anterior insertion of the right ventricle, draw star-shaped scattering lines to evenly segment the contour ring (Fig. 11);

Step 4.4, performing corrosion operation on each segmented area;

In step 4.5, number the center of gravity of each segmented region in a clockwise direction, starting from the anterior insertion of the right ventricle.

Step 5, calculate the average value of the original image pixels of each segmented region after segmentation.

specific:

Step 5.1, get all the pixels of each segment area after segmentation, and color each segment area with different colors (Figure 12);

Step 5.2, mapping the coordinate position of each pixel point to the original image to obtain the pixel value;

Step 5.3, calculating the pixel average value of each region according to the pixel value of the original image of each region.

Taking T1 mapping images of 50 healthy people (a total of 150 DICOM original images) as an example for verification, the success rate of accurately identifying the target area reached 91.3%.

It can be seen from the above method that the present invention achieves the purpose of using computer to automatically process cardiac magnetic resonance mapping images, has extremely high accuracy and high repeatability, and has the advantages of reducing the workload of doctors, and has a good application prospect .

Claims (7)

1. A cardiac magnetic resonance mapping quantification method, is characterized in that, comprises the steps: Step 1, input the original data of the cardiac magnetic resonance mapping image, obtain the original images of the left ventricular basal layer, middle layer and apical layer respectively, and perform subsequent steps on at least one layer of the original images of the basal layer, middle layer and apical layer; Step 2, preprocessing the original image, and determining the target area of the left ventricular myocardium based on contour recognition; Step 3, obtain the precise target area of the segmented target through connected domain analysis; Step 4, segmenting based on the precise target area; Step 5, calculating the average value of the original image pixels of each segmented region after segmentation; Step 4 specifically includes: Step 4.1, performing contour repair on the left ventricular endocardium and epicardium, the method of contour repairing is: filling faults through morphological operations and deleting bloat; Step 4.2, drawing the contour ring of the left ventricular myocardium; Step 4.3, based on the center of gravity of the left ventricular chamber and the anterior insertion of the right ventricle, draw star-shaped scattering lines to evenly segment the contour ring; Step 4.4, performing corrosion operation on each segmented area; Step 4.5, starting from the anterior insertion of the right ventricle, number the center of gravity of each segment in a clockwise direction; In step 4, the original image of the basal layer or the original image of the middle layer is divided into 6 segments, and the original image of the apical layer is divided into 4 segments; Step 5 specifically includes: Step 5.1, obtain all the pixel points of each segment area after segmentation, and color each segment area with different colors; Step 5.2, mapping the coordinate position of each pixel point to the original image to obtain the pixel value; Step 5.3, calculating the pixel average value of each region according to the pixel value of the original image of each region. 2. according to the described cardiac magnetic resonance mapping quantitative method of claim 1, it is characterized in that, the format of described raw data is DICOM file. 3. according to the cardiac magnetic resonance mapping quantification method of claim 1, it is characterized in that, step 2 specifically comprises: Step 2.1, converting the original image into a grayscale image; Step 2.2, filter background noise and foreground noise; Step 2.3, draw a reference circle with a radius of 1/4 width with the center point of the image; Step 2.4, performing contour detection on the entire image, traversing the detected contours, and those whose center of gravity of the contour is within the reference circle are candidate contours; Step 2.5, obtain the candidate contour with the largest area as the target contour; Step 2.6, drawing the extreme points of the target contour; Step 2.7, acquiring the target area according to the extreme points. 4. according to the cardiac magnetic resonance mapping quantitative method described in claim 3, it is characterized in that, step 2.2 specifically comprises: Step 2.2.1, filter the background noise by threshold analysis and median filtering; Step 2.2.2, filter background noise by connected domain analysis; Step 2.2.3, the image is inverted, and the foreground noise is filtered through connected domain analysis; Step 2.2.4, perform image inversion again; The shape of the target area is a rectangle; step 2.7 specifically includes: Step 2.7.1, taking the extreme left point as the center point c(x ₀ , y ₀ ) of the target area; Step 2.7.2, using half of the vertical distance between the upper extreme point and the lower extreme point as the reference value d to calculate the upper left point and lower right point of the target area, the calculation formula is: Coordinates of upper left point: lt(x ₁ , y ₁ ) = (x ₀ - 0.5*d, y ₀ - 0.8*d); Coordinates of lower right point: rb(x ₂ , y ₂ ) = (x ₀ + r, y ₀ + 0.8*d); Step 2.7.3, drawing the target area. 5. according to the described cardiac magnetic resonance mapping quantitative method of claim 1, it is characterized in that, step 3 specifically comprises: Step 3.1, preprocessing by threshold analysis; Step 3.2, through connected domain analysis, remove connected domains with small area, leaving at least 3 connected domains as candidate regions; Step 3.3, locating the left ventricle by contour detection; Step 3.4, using the convex hull algorithm to locate the anterior insertion of the right ventricle; Step 3.5, perform image inversion; Step 3.6, filtering the background noise by at least one method of threshold analysis, median filter, and corrosion expansion; In step 3.7, the connected domain with the largest area is obtained as the precise target area through connected domain analysis. 6. A cardiac magnetic resonance mapping quantification system, characterized in that, comprising: The input module is used to input the raw data of the cardiac magnetic resonance mapping image; A calculation module, configured to process the raw data according to the cardiac magnetic resonance mapping quantification method according to any one of claims 1-5; The output module is used to output the processing result of the calculation module. 7. A computer-readable storage medium, wherein a computer program for realizing the cardiac magnetic resonance mapping quantification method according to any one of claims 1-5 is stored thereon.

Images