KR-20260066509-A - Method and system for segmentation of blood vessel image

Abstract

A method for segmenting a vascular image is disclosed. The method for segmenting a vascular image according to the present invention comprises: a step of determining first segmentation result information that distinguishes between a vascular region and a non-vascular region from a vascular image including a coronary artery using a trained neural network model; and a step of determining second segmentation result information from which noise has been removed by post-processing the first segmentation result information.

Inventors

• 박지용
• 김영인
• 김판기

Assignees

• 주식회사 팬토믹스

Dates

Publication Date

20260512

Application Date

20241104

Claims (6)

1. In a method for segmenting a vascular image using at least one processor, A step of determining first segmentation result information that distinguishes between vascular regions and non-vascular regions from a vascular image including coronary arteries using a trained neural network model; and The method includes the step of determining a second division result information from which noise has been removed by post-processing the first division result information; The above neural network model includes at least one neural network model trained to distinguish between vascular regions and non-vascular regions from vascular images including coronary arteries, and A method for segmenting a vascular image characterized in that at least one of the parameters of the learning and inference algorithms of the neural network model is determined so as to increase sensitivity even if the first segmentation result information contains noise.
2. In paragraph 1, A method for segmenting a vascular image, characterized in that at least one of the above neural network models is trained using the sampling of the vascular image as training data, and is trained using only the positive sampling that includes the vascular region.
3. In paragraph 1, The above neural network models are three in number, each trained using a training dataset in which the ratio of positive sampling and negative sampling including vascular regions among the vascular images is determined. A step of training a first model with a first training dataset of a first sampling rate; A step of training a second model with a second training dataset of a second sampling rate; and Step of training a third model with a third training dataset of a third sampling rate; The method includes the step of determining the first division result information by synthesizing the inference results of each of the first to third models; A method for segmenting a vascular image, characterized in that the first sampling and second sampling ratios are 50:50, and the third sampling includes only positive sampling.
4. In paragraph 3, A step of generating a first group of images by cropping a vascular image into multiple images to have a first overlap ratio; A step of generating a first crop division information group from each of the images included in the first image group; A step of determining first division information from the first crop division information group; A step of generating a second group of images by cropping the vascular images into multiple images to have a second overlap ratio; A step of generating a second crop division information group from each of the images included in the second image group; A step of determining first division information from the second crop division information group; A step of generating a third group of images by cropping the vascular images into multiple images to have a third overlap ratio; A step of generating a third crop division information group from each of the images included in the third image group; and The method includes the step of determining first division information from the third crop division information group; A method for segmenting a vascular image, characterized in that the first to third overlap ratios are 25%, 40%, and 30%, respectively.
5. In paragraph 1, The above neural network model is multiple, and The method includes the step of determining the first division result information by synthesizing the inference results of each of the plurality of neural networks; A method for segmenting a blood vessel image characterized by the synthesis of the above inference results such that any point inferred as a blood vessel by any one of the above multiple neural network models is determined to be a blood vessel.
6. Memory for storing vascular images including coronary arteries; and A processor that distinguishes between vascular regions and non-vascular regions from the above vascular image; is included, The above processor is, A first segmentation result information is determined by using a trained neural network model to distinguish between vascular regions and non-vascular regions from a vascular image including coronary arteries, and a second segmentation result information is determined by post-processing the first segmentation result information to remove noise. The above neural network model includes at least one neural network model trained to distinguish between vascular regions and non-vascular regions from vascular images including coronary arteries, and A segmentation system for vascular images characterized in that at least one of the parameters of the learning and inference algorithms of the neural network model is determined so as to increase sensitivity even if the first segmentation result information contains noise.

Description

Method and system for segmentation of blood vessel image The present invention relates to a method and system for segmenting vascular images. Curved MPR (Multi-Planar Reconstruction) is a two-dimensional planar image extracted from multiple angles of a three-dimensional computed tomography (CT) image (hereinafter also referred to as a ‘CT image’). Curved MPR is useful for diagnosing conditions such as stenosis of the heart blood vessels. In the observation of coronary arteries, the generation of high-quality curved MPR images is an important basis for determining the three-dimensional position of the coronary artery centerline. In generating curved MPR, segmenting blood vessels from CT images is a fundamental step in extracting the centerlines of the vessels. Here, vessel segmentation refers to the process of distinguishing vascular and non-vascular regions within the image using binary information. When segmenting blood vessels, the important task is to ensure that all vessels appear without interruption. However, when separating blood vessels from CT images, there are parts that are difficult to recognize as blood vessels due to various reasons such as stenosis, artifacts around the vessels, and failure of the contrast agent to reach them, so a solution for this is required. Figure 1 shows a vascular image segmentation system according to an embodiment of the present invention. Figure 2 shows in detail a part of the configuration of a vascular image segmentation system according to an embodiment of the present invention. Figure 3 shows a method for segmenting a vascular image according to an embodiment of the present invention. Figure 4 shows in detail a part of the configuration of a method for segmenting a vascular image according to an embodiment of the present invention. Figure 5 shows in detail a part of the configuration of a method for segmenting a vascular image according to an embodiment of the present invention. Figure 6 shows the ground truth of the coronary artery segmented from the vascular image. Figure 7 shows the first segmentation result information inferred by the neural network model. Figure 8 shows the second division result information obtained by post-processing the first division result information. The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the present invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. In describing the present invention, if it is determined that a detailed description of related known technology may obscure the essence of the present invention, such detailed description is omitted. Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. FIG. 1 shows a vascular image segmentation system (100) (hereinafter also briefly referred to as ‘system (100)’) according to an embodiment of the present invention. Referring to FIG. 1, the vascular image segmentation system (100) includes a processor (110), a memory (120), and a communication unit (130). The processor (110) is for performing information processing related to the segmentation of a vascular image and may include at least one processor (110). Here, vascular segmentation can refer to the process of distinguishing vascular regions and non-vascular regions in an image using binary information. The processor (110) may include at least one of a Central Processing Unit, a Graphic Processing Unit, a Micro Processor, and a processor dedicated to artificial intelligence, and the type and number of processors are not limited thereto as long as they perform the functions of the present invention. Memory (120) can store a program, which is a set of data and executable instructions that can be read or written by a processor. The memory (120) includes non-volatile memory (120) which can store data (information) regardless of whether power is provided, and volatile memory which cannot store data when power is not provided, in which data is loaded to be processed by a processor. Non-volatile memory includes flash memory, HDD (hard-disc drive), SSD (solid-state drive), ROM (Read Only Memory), etc., and volatile memory includes buffer, RAM (Random Access Memory), etc. Figure 2 shows the configuration of the memory (120) in detail. Referring to FIG. 2, the memory (120) includes a first model (121), a second model (122), and a third model (123). Each model may include the structure of an artificial intelligence neural network, and the number of artificial intelligence models is exemplary and is not necessarily limited to the embodiments of the present invention. Each model may include information regarding the model architecture and information regarding parameters. Each