# US 2021/0174498 A1 · Image processing device and calcification analysis system including the same

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0174498 A1
- Document ID: US20210174498A1
- Status: Published application
- Kind: A1
- Application number: 16/950,573
- Filed: 2020-11-17
- Published: 2021-06-10
- Earliest claimed: 2019-12-10
- Assignees: Electronics and Telecommunications Research Institute
- Applicants: Electronics and Telecommunications Research Institute
- Inventors: Ho-Youl JUNG; Jae Hun CHOI
- Web page: https://patentgrep.com/patent/US20210174498A1

## CPC

- G06T 7/0012: {Biomedical image inspection}
- A61B 6/032: {Transmission computed tomography [CT]}
- A61B 6/466: {adapted to display 3D data}
- A61B 6/503: {for diagnosis of the heart}
- A61B 6/504: {for diagnosis of blood vessels, e.g. by angiography}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06V 10/82: using neural networks
- G06T 2200/04: involving 3D image data
- G06T 2207/10028: Range image; Depth image; 3D point clouds
- G06T 2207/10081: Computed x-ray tomography [CT]
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30048: Heart; Cardiac
- G06T 2207/30101: Blood vessel; Artery; Vein; Vascular
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

The image processing device includes a voxel extractor, a learner, and a predictor. The voxel extractor extracts a target voxel and neighboring voxels adjacent to the target voxel from a 3D image. The learner generates vectors corresponding to the target voxel and the neighboring voxels, respectively, generates vector weights corresponding to each of the vectors, based on the vectors and a parameter group, and adjusts the parameter group, based on an analysis result of the target voxel generated by applying the vector weights to the vectors. The predictor generates vectors corresponding to the target voxel and the neighboring voxels, respectively, generates correlation weights among the vectors by applying a parameter group to the vectors, generates vector weights corresponding to each of the vectors by applying the correlation weights to the vectors, and generates an analysis result of the target voxel by applying the vector weights to the vectors.

## Claims

### Claim 1

An image processing device comprising:
a voxel extractor configured to extract a target voxel and neighboring voxels adjacent to the target voxel from a 3D image; and
a learner configured to generate vectors corresponding to the target voxel and the neighboring voxels, respectively, to generate vector weights corresponding to each of the vectors, based on the vectors and a parameter group, and to adjust the parameter group, based on an analysis result of the target voxel generated by applying the vector weights to the vectors.

### Claim 2 (depends on 1)

The image processing device of claim 1, wherein the number of the neighboring voxels is (2n+1)3−1 with respect to a natural number ‘n’.

### Claim 3 (depends on 1)

The image processing device of claim 1, wherein the learner generates correlation weights among the vectors by applying the parameter group to the vectors, and generates the vector weights by applying the correlation weights to the vectors.

### Claim 4 (depends on 3)

The image processing device of claim 3, wherein the learner generates a vector weight corresponding to a target vector by applying correlation weights between the target vector and the vectors among the correlation weights to the vectors.

### Claim 5 (depends on 1)

The image processing device of claim 1, wherein the learner generates the vectors through a convolutional neural network operation on each of the target voxel and the neighboring voxels.

### Claim 6 (depends on 1)

The image processing device of claim 1, wherein the learner generates the vectors by merging a result of a convolutional neural network operation on each of the target voxel and the neighboring voxels and a result of a dilated convolutional neural network operation on each of the target voxel and the neighboring voxels.

### Claim 7 (depends on 1)

The image processing device of claim 1, wherein the learner adjusts the parameter group until the analysis result is within a reference error from a preset result.

### Claim 8 (depends on 1)

The image processing device of claim 1, wherein the 3D image is a computed tomography image, and wherein the analysis result is a calcification index.

### Claim 9

An image processing device comprising:
a voxel extractor configured to extract a target voxel and neighboring voxels adjacent to the target voxel from a 3D image; and
a predictor configured to generate vectors corresponding to the target voxel and the neighboring voxels, respectively, to generate correlation weights among the vectors by applying a parameter group to the vectors, to generate vector weights corresponding to each of the vectors by applying the correlation weights to the vectors, and to generate an analysis result of the target voxel by applying the vector weights to the vectors.

### Claim 10 (depends on 9)

The image processing device of claim 9, wherein the number of the neighboring voxels is (2n+1)3−1 with respect to a natural number ‘n’.

### Claim 11 (depends on 9)

The image processing device of claim 9, wherein the predictor generates target correlation weights among a target vector and the vectors, based on the target vector corresponding to the target voxel among the vectors, the vectors, and the parameter group.

### Claim 12 (depends on 11)

The image processing device of claim 11, wherein the predictor generates a vector weight corresponding to the target vector by adding values obtained by multiplying each of the vectors to each of the target correlation weights.

### Claim 13 (depends on 9)

The image processing device of claim 9, wherein the parameter group includes parameter values depending on a correlation of the vectors for each of the vectors.

### Claim 14 (depends on 9)

The image processing device of claim 9, wherein the voxel extractor sets each of a plurality of voxels included in the 3D image as the target voxel, and extracts the neighboring voxels, based on the set target voxel.

### Claim 15 (depends on 14)

The image processing device of claim 14, wherein the predictor calculates a calcification index of the 3D image, based on analysis results of each of the voxels.

### Claim 16

A calcification analysis system comprising:
a computed tomography device configured to generate a 3D computed tomography image; and
an image processing device configured to extract a target voxel and neighboring voxels adjacent to the target voxel from the 3D computed tomography image, to generate vectors corresponding to the target voxel and the neighboring voxels, respectively, to generate vector weights corresponding to each of the vectors, based on a correlation among the vectors, and to calculate a calcification index of the target voxel by applying the vector weights to the vectors.

### Claim 17 (depends on 16)

The calcification analysis system of claim 16, wherein the image processing device includes a learner configured to generate correlation weights among the vectors by applying a parameter group to the vectors, to generate the vector weights by applying the correlation weights to the vectors, and to adjust the parameter group, based on the calcification index.

### Claim 18 (depends on 17)

The calcification analysis system of claim 17, wherein the learner adjusts the parameter group until the calcification index is within a reference error from a preset result, and
wherein the adjusted parameter group is stored in a learning model database.

### Claim 19 (depends on 16)

The calcification analysis system of claim 16, wherein the image processing device includes a predictor configured to generate correlation weights among the vectors by applying a parameter group provided from a learning model database to the vectors, to generate the vector weights by applying the correlation weights to the vectors, and to calculate the calcification index of the target voxel by applying the vector weights to the vectors.

### Claim 20 (depends on 19)

The calcification analysis system of claim 19, wherein the image processing device further includes a voxel extractor configured to set each of a plurality of voxels included in the 3D image as the target voxel, and to extract the neighboring voxels, based on the set target voxel, and
wherein the predictor calculates a calcification index of the 3D image, based on analysis results of each of the voxels.
