A blood flow reserve fraction prediction method and system based on deep learning

Abstract

The present invention proposes a blood flow reserve fraction prediction method and system based on deep learning, which relates to the field of image processing technology. The specific scheme includes: decomposing the original CTA image into a single-vessel probe image sequence based on the extracted vascular centerline and segment center point; extracting the vascular segment image features containing complete vascular time series information from the single-vessel probe image sequence, and predicting the blood flow reserve fraction of the vascular segment based on the vascular segment image features; the blood flow reserve fraction of the vascular segment is matched one-to-one with the segment center point on the vascular centerline to obtain the complete blood flow reserve fraction prediction result corresponding to the original CTA image; based on 3DCNN and Transformer, the present invention extracts segment features containing time series from the CTA image, and quickly and accurately calculates the blood flow reserve fraction at each position in the blood vessel, with the advantages of simple processing, high timeliness and stable effect.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and in particular relates to a blood flow reserve fraction prediction method and system based on deep learning. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Fractional Flow Reserve (FFR) is a technology used during coronary catheterization to measure the pressure difference at the site of coronary artery stenosis. The currently commonly used method for measuring FFR is invasive measurement using a pressure guidewire, where the ratio of the proximal pressure to the distal pressure measured using the pressure guidewire is used as the FFR. However, using a pressure guidewire to measure FFR is an invasive examination and carries a certain risk of surgical complications. Therefore, the ability to accurately calculate FFR using a non-invasive method has become an urgent issue to be addressed. Due to the large number of patients, the market prospects are very broad.

[0004] With the widespread application of deep learning technology in medical imaging, there has been a lot of research on the detection of lesions such as lung nodules, chest fractures, coronary plaques and stenosis, and it plays a relatively high role in practical applications; however, the existing deep learning-based blood flow reserve fraction prediction method predicts the blood flow reserve fraction of the entire blood vessel and cannot be refined to each location of the blood vessel. The prediction model does not even consider the temporal relationship between blood vessel segments, so the prediction accuracy and precision are insufficient.

[0005] Therefore, how to deeply explore the temporal relationship between vascular segments to improve the accuracy and precision of prediction has become a topic worthy of research. Summary of the Invention

[0006] To overcome the shortcomings of the above-mentioned existing technologies, the present invention provides a deep learning-based blood flow reserve fraction prediction method and system. Based on 3DCNN and Transformer, it extracts segment features containing time series from CTA images, and quickly and accurately calculates the blood flow reserve fraction at various locations in the blood vessels. It has the advantages of simple processing, high timeliness and stable results.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] The first aspect of the present invention provides a method for predicting blood flow reserve fraction based on deep learning;

[0009] A blood flow reserve fraction prediction method based on deep learning, comprising:

[0010] Based on the extracted vascular centerlines and segment center points, the original CTA image is decomposed into a single-vessel probe image sequence;

[0011] Extracting vessel segment image features containing complete vessel temporal information from a single vessel probe image sequence, and predicting the blood flow reserve fraction of the vessel segment based on the vessel segment image features;

[0012] The blood flow reserve fraction of the vascular segment is matched one-to-one with the segment center point on the vascular centerline to obtain the complete blood flow reserve fraction prediction result corresponding to the original CTA image.

[0013] Furthermore, the decomposition steps of the single blood vessel probe image sequence are:

[0014] Use vtk to extract the centerline of each blood vessel from the original CTA image;

[0015] Based on the extracted vascular centerline, the single vessel probe image corresponding to each vessel is obtained from the original CTA image, and a single vessel probe image library consisting of the single vessel probe images of all vessels is obtained;

[0016] By adopting the equal-interval sampling method, blood vessel segments of equal length with the point on the center line as the center point are intercepted from the single blood vessel probe image to obtain the single blood vessel probe image sequence.

[0017] Furthermore, the extraction of the centerline of each blood vessel also requires determining the starting position of each blood vessel based on the original CTA image.

[0018] Furthermore, after obtaining the single-vessel probe image, the tail of the blood vessels that do not meet the length requirement is padded with blank segments.

[0019] Furthermore, the trained prediction model is used to extract features and calculate the blood flow reserve score. The prediction model consists of 3DCNN, Transformer and sigmoid function.

[0020] The prediction model takes the probe image as input and outputs the blood flow reserve fraction.

[0021] Furthermore, the prediction model is specifically used to:

[0022] The probe image is input into the trained 3DCNN to extract image features containing short time series information;

[0023] The extracted image features are flattened and put into the Transformer for temporal feature enhancement to obtain the vascular segment image features containing the complete temporal information of the blood vessels;

[0024] Based on the extracted vascular segment image features, the blood flow reserve fraction is calculated using the sigmoid function.

[0025] Furthermore, when constructing the training data set of the prediction model, blood vessels are screened by the minimum blood flow reserve fraction of the blood vessels.

[0026] A second aspect of the present invention provides a blood flow reserve fraction prediction system based on deep learning.

[0027] A blood flow reserve fraction prediction system based on deep learning, including a probe sequence extraction module, a segmented score prediction module and a final score generation module;

[0028] The probe sequence extraction module is used to decompose the original CTA image into a single-vessel probe image sequence based on the extracted vessel centerline and segment center point;

[0029] A segmented score prediction module is used to extract the image features of a blood vessel segment containing complete temporal information of the blood vessel from a single-vessel probe image sequence, and to predict the blood flow reserve fraction of the blood vessel segment based on the image features of the blood vessel segment;

[0030] The final score generation module is used to correspond the blood flow reserve score of the vascular segment to the segment center point on the vascular centerline one by one, and obtain the complete blood flow reserve score prediction result corresponding to the original CTA image.

[0031] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a blood flow reserve fraction prediction method based on deep learning as described in the first aspect of the present invention.

[0032] A fourth aspect of the present invention provides an electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of a deep learning-based blood flow reserve fraction prediction method as described in the first aspect of the present invention are implemented.

[0033] One or more of the above technical solutions have the following beneficial effects:

[0034] The deep learning-based blood flow reserve fraction prediction method provided by the present invention extracts the centerline of the coronary artery and decomposes the original CTA image into a single-vessel probe image sequence based on the centerline, so that image features containing short-term time series information can be extracted using 3DCNN, and then features containing complete time series information of the blood vessels can be obtained using Transformer. By using deep learning technology, the blood flow reserve fraction at various locations in the patient's coronary arteries can be quickly and accurately predicted under non-invasive conditions. The coronary CTA images can be effectively processed, with the advantages of simple processing, high timeliness and stable results.

[0035] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0037] Figure 1 This is a flow chart of the method of the first embodiment.

[0038] Figure 2 These are three imaging mode diagrams of the first embodiment.

[0039] Figure 3 This is a structural diagram of the prediction model of the first embodiment.

[0040] Figure 4 Schematic diagram of the three-dimensional coronary artery model of the first embodiment.

[0041] Figure 5 This is a system structure diagram of the second embodiment. DETAILED DESCRIPTION

[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0043] Example 1

[0044] This embodiment discloses a blood flow reserve fraction prediction method based on deep learning;

[0045] like Figure 1 As shown, a blood flow reserve fraction prediction method based on deep learning includes:

[0046] Step S1: Decompose the original CTA image into a single vessel probe image sequence based on the extracted vessel centerline and segment center point, specifically:

[0047] To obtain coronary CTA images, CTA images are obtained by intravenous injection of iodinated contrast agent, followed by thin-layer scanning, and finally image post-processing; image post-processing has three imaging modes, such as Figure 2 As shown, they are surface imaging, curved surface reconstruction, and coronary probe mode. The subsequent processing of this embodiment is mainly for the CTA image in the coronary probe mode; the CTA image is semantically segmented and then three-dimensionally reconstructed to obtain the angiogenesis tree.

[0048] The third-party library vtk is used to extract the centerline of each blood vessel and the starting position of the blood vessel is calculated based on the angiogenesis tree;

[0049] Based on the extracted centerline, a single vessel probe image corresponding to each blood vessel is obtained from the original CTA image. That is, a probe image of each blood vessel is obtained. After obtaining the single vessel probe images of all blood vessels, a single vessel probe image library corresponding to the original CTA image is formed.

[0050] Using the equally spaced sampling method, with the point on the center line as the center point, equal-length cubes are intercepted from the single-vessel probe image as vessel segments. The vessel segments arranged in the interception order form a single-vessel probe image sequence. Because the lengths of blood vessels vary, in order to ensure the consistency of the data length input into the prediction model, blank segments are used to pad the tails of blood vessels that do not reach the required length.

[0051] Step S2: extracting vessel segment image features containing complete vessel temporal information from the single vessel probe image sequence, and predicting the blood flow reserve fraction of the vessel segment based on the vessel segment image features;

[0052] In a specific embodiment, a trained prediction model is used to extract features and calculate blood flow reserve fraction. The prediction model is composed of 3DCNN, Transformer and sigmoid function. The specific structure is as follows Figure 3 As shown, the specific steps of the prediction model are:

[0053] The probe image is input into the trained 3DCNN to extract image features containing short-term temporal information;

[0054] The extracted image features are flattened and put into the Transformer for temporal feature enhancement to obtain the vascular segment image features containing the complete temporal information of the blood vessels;

[0055] Based on the extracted vascular segment image features, the blood flow reserve fraction is calculated using the sigmoid function.

[0056] The blood flow reserve fraction results range from 0 to 1, and all FFR values ​​before the starting position in the angiogenesis tree are set to 1.

[0057] When constructing the training data set for the prediction model, blood vessels are screened based on their minimum blood flow reserve score. Specifically, if the minimum blood flow reserve score of a certain blood vessel is greater than 0.95, then for the sake of data balance, this vessel will be discarded and will not appear in the training data set.

[0058] Step S3: The blood flow reserve fraction of the vessel segment is matched one-to-one with the segment center point on the vessel centerline to obtain a complete blood flow reserve fraction prediction result corresponding to the original CTA image.

[0059] Specifically, the blood flow reserve fractions of the vascular segments are filled based on the vascular segments obtained by sampling at equal intervals, so that the number of blood flow reserve fractions matches the vascular length and corresponds one-to-one with the center points of the segments on the vascular centerline. The results corresponding to all the center points on the centerline are saved as a json file and mapped to the three-dimensional coronary model according to the coordinates of the points. The three-dimensional coronary model is as follows: Figure 4 As shown in FIG, the three-dimensional coronary artery model is used as the prediction result of the complete blood flow reserve fraction corresponding to the original CTA image.

[0060] Example 2

[0061] This embodiment discloses a blood flow reserve fraction prediction system based on deep learning;

[0062] like Figure 5 As shown, a blood flow reserve fraction prediction system based on deep learning includes a probe sequence extraction module, a segmented score prediction module and a final score generation module;

[0063] The probe sequence extraction module is used to decompose the original CTA image into a single-vessel probe image sequence based on the extracted vessel centerline and segment center point;

[0064] A segmented score prediction module is used to extract the image features of a blood vessel segment containing complete temporal information of the blood vessel from a single-vessel probe image sequence, and to predict the blood flow reserve fraction of the blood vessel segment based on the image features of the blood vessel segment;

[0065] The final score generation module is used to correspond the blood flow reserve score of the vascular segment to the segment center point on the vascular centerline one by one, and obtain the complete blood flow reserve score prediction result corresponding to the original CTA image.

[0066] Example 3

[0067] The purpose of this embodiment is to provide a computer-readable storage medium.

[0068] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a method for predicting blood flow reserve fraction based on deep learning as described in the first embodiment of the present disclosure.

[0069] Example 4

[0070] The purpose of this embodiment is to provide an electronic device.

[0071] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of a deep learning-based blood flow reserve prediction method as described in the first embodiment of the present disclosure are implemented.

[0072] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A blood flow reserve fraction prediction method based on deep learning, characterized in that: include: Based on the extracted vascular centerlines and segment center points, the original CTA image is decomposed into a single vessel probe image sequence. The decomposition steps of the single vessel probe image sequence are as follows: The centerline of each blood vessel is extracted from the original CTA image using vtk. Based on the extracted blood vessel centerlines, the single vessel probe image corresponding to each blood vessel is obtained from the original CTA image, thereby obtaining a single vessel probe image library consisting of the single vessel probe images of all blood vessels. Using the equal-interval sampling method, blood vessel segments of equal length with the point on the center line as the center point are intercepted from the single-vessel probe image to obtain a single-vessel probe image sequence; Extracting vessel segment image features containing complete vessel temporal information from a single vessel probe image sequence, and predicting the blood flow reserve fraction of the vessel segment based on the vessel segment image features; The blood flow reserve fraction of the vascular segment is matched one-to-one with the segment center point on the vascular centerline to obtain the complete blood flow reserve fraction prediction result corresponding to the original CTA image.

2. The method for predicting blood flow reserve fraction based on deep learning according to claim 1, wherein: The extraction of the center line of each blood vessel also requires determining the starting position of each blood vessel based on the original CTA image.

3. The method for predicting blood flow reserve fraction based on deep learning according to claim 1, wherein: After obtaining the single-vessel probe image, blank segments are used to fill in the tail of the blood vessels that do not meet the length requirement.

4. The method for predicting blood flow reserve fraction based on deep learning according to claim 1, wherein: Using the trained prediction model, features are extracted and blood flow reserve fraction is calculated. The prediction model consists of 3DCNN, Transformer and sigmoid function. The prediction model takes the probe image as input and outputs the blood flow reserve fraction.

5. The method for predicting blood flow reserve fraction based on deep learning according to claim 4, characterized in that: The prediction model is specifically used to: The probe image is input into the trained 3DCNN to extract image features containing short time series information; The extracted image features are flattened and put into the Transformer for temporal feature enhancement to obtain the vascular segment image features containing the complete temporal information of the blood vessels; Based on the extracted vascular segment image features, the blood flow reserve fraction is calculated using the sigmoid function.

6. The method for predicting blood flow reserve fraction based on deep learning according to claim 5, characterized in that: When constructing the training data set of the prediction model, blood vessels are screened by the minimum blood flow reserve fraction of the blood vessels.

7. A blood flow reserve fraction prediction system based on deep learning, characterized in that: It includes a probe sequence extraction module, a segmentation score prediction module, and a final score generation module; The probe sequence extraction module is used to decompose the original CTA image into a single vessel probe image sequence based on the extracted vessel centerline and segment center point. The decomposition steps of the single vessel probe image sequence are as follows: Use vtk to extract the centerline of each blood vessel from the original CTA image; Based on the extracted vascular centerline, the single vessel probe image corresponding to each vessel is obtained from the original CTA image, and a single vessel probe image library consisting of the single vessel probe images of all vessels is obtained; Using the equal-interval sampling method, blood vessel segments of equal length with the point on the center line as the center point are intercepted from the single-vessel probe image to obtain a single-vessel probe image sequence; A segmented score prediction module is used to extract the image features of a blood vessel segment containing complete temporal information of the blood vessel from a single-vessel probe image sequence, and to predict the blood flow reserve fraction of the blood vessel segment based on the image features of the blood vessel segment; The final score generation module is used to correspond the blood flow reserve score of the vascular segment to the segment center point on the vascular centerline one by one, and obtain the complete blood flow reserve score prediction result corresponding to the original CTA image.

8. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the blood flow reserve fraction prediction method based on deep learning are implemented.

9. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the blood flow reserve fraction prediction method based on deep learning are implemented as described in any one of claims 1 to 6.

Images