# US 12,217,430 B2 · Systems and methods for processing x-ray images

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

- Number: US 12,217,430 B2
- Document ID: US12217430B2
- Status: Granted
- Kind: B2
- Application number: 17/456,393
- Filed: 2021-11-24
- Published: 2022-03-24
- Granted: 2025-02-04
- Earliest claimed: 2019-05-24
- Assignees: Shanghai United Imaging Healthcare Co., Ltd.
- Applicants: Shanghai United Imaging Healthcare Co., Ltd.
- Inventors: Kai Cui; Chunhua Jiang; Liang Xu; Shurui Zhao; Jian Zhong; Jie Niu; Yan'ge Ma; Wanli Teng; Na Zhang; Juan Feng
- Web page: https://patentgrep.com/patent/US12217430B2

## CPC

- G06T 7/11: Region-based segmentation
- G06T 7/12: Edge-based segmentation
- A61B 6/12: Arrangements for detecting or locating foreign bodies
- A61B 6/505: {for diagnosis of bone}
- A61B 6/5205: {involving processing of raw data to produce diagnostic data}
- A61B 6/5217: {extracting a diagnostic or physiological parameter from medical diagnostic data}
- A61B 6/545: {involving automatic set-up of acquisition parameters}
- G06T 7/0012: {Biomedical image inspection}
- G06T 7/155: involving morphological operators
- G06T 7/181: involving edge growing; involving edge linking
- G06T 7/187: involving region growing; involving region merging; involving connected component labelling
- G06T 2207/10081: Computed x-ray tomography [CT]
- G06T 2207/10116: X-ray image
- G06T 2207/20016: Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/20156: Automatic seed setting
- G06T 2207/30004: Biomedical image processing
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)

## Abstract

The present disclosure is related to systems and methods for processing X-ray images. A method for processing an X-ray image may include obtaining an X-ray image; and determining a metal image based on the X-ray image by using a trained metal detection model. The metal image includes information of a metal object in the X-ray image.

## Claims

### Claim 1

A system for processing an X-ray image, comprising:
at least one storage device including a set of instructions for processing an X-ray image; and
at least one processor in communication with the at least one storage device, wherein when executing the set of instructions, the at least one processor is directed to cause the system to perform operations including:
obtaining an X-ray image; and
determining a metal image based on the X-ray image by using a trained metal detection model, wherein the metal image includes information of a metal object in the X-ray image, and the information of the metal object includes at least one of a position of the metal object, a shape of the metal object, a size of the metal object, an orientation of the metal object, or a type of the metal object.

### Claim 2 (depends on 1)

The system of claim 1, wherein the trained metal detection model is determined based on a training process, the training process including:
obtaining a sample set including a plurality of sample pairs, wherein each sample pair of the plurality of sample pairs includes a sample X-ray image and a labeled metal image with respect to the X-ray image; and
obtaining the trained metal detection model by training a preliminary metal detection model based on the sample set.

### Claim 3 (depends on 2)

The system of claim 2, wherein the trained metal detection model includes a deep learning model.

### Claim 4 (depends on 3)

The system of claim 3, wherein the deep learning model is a V-Net neural network model including a skip connection, wherein the skip connection includes:
combing at least one of at least one down-sampling feature corresponding to at least one down-sampling layer during up-sampling.

### Claim 5 (depends on 4)

The system of claim 4, wherein the skip connection further includes:
in each up-sampling layer of a plurality of up-sampling layers, combing a down-sampling feature of a down-sampling layer corresponding to the up-sampling layer and a refined feature of the down-sampling feature.

### Claim 6 (depends on 4)

The system of claim 4, wherein a loss function of the V-Net neural network model is an FL loss function determined based on a Focal Loss function and a Lovasz Loss function.

### Claim 7 (depends on 1)

The system of claim 1, wherein the determining the metal image based on the X-ray image by using the trained metal detection model includes:
determining, by using the trained metal detection model, a probability graph representing a probability that a pixel in the X-ray image belongs to a metal region; and
determining the metal image based on the probability graph.

### Claim 8 (depends on 1)

The system of claim 1, wherein the determining the metal image based on the X-ray image by using the trained metal detection model includes:
determining an initial binary image indicating an initial metal region by using the trained metal detection model, wherein the trained metal detection model is trained from a neural network model for category semantic perception; and
determining the metal image indicating a target metal region based on the initial binary image, wherein the metal image is a target binary image.

### Claim 9 (depends on 8)

The system of claim 8, wherein the determining the metal image indicating the target metal region based on the initial binary image includes:
determining whether the initial metal region is a connected region;
in response to a determination that the initial metal region is the connected region, obtaining the target metal region by filling the initial metal region according to a morphological reconstruction algorithm; and
in response to a determination that the initial metal region is a dis-connected region,
determining a target region edge by performing an edge connection processing on an initial edge of the initial metal region according to a predetermined edge processing algorithm; and
obtaining the target metal region by filling a region within the target region edge according to the morphological reconstruction algorithm.

### Claim 10 (depends on 8)

The system of claim 8, wherein the trained metal detection model is determined based on a training process, the training process including:
obtaining a sample set including a plurality of sample pairs, wherein each sample pair of the plurality of sample pairs includes a sample X-ray image and a reference binary image indicating a reference metal region with respect to the X-ray image;
obtaining the trained metal detection model by training a preliminary metal detection model based on the sample set, wherein
the preliminary metal detection model is the neural network model for category semantic perception,
model parameters of the preliminary metal detection model include initial model parameters of the neural network model for category semantic perception, the initial model parameters including a random number based on a zero mean and a predetermined variance, and
a loss function of the preliminary metal detection model includes a multi-category comparison loss function of the neural network model for category semantic perception.

### Claim 11 (depends on 8)

The system of claim 8, wherein the neural network model for category semantic perception includes:
an initial feature extraction structure including a first convolution layer, at least three transition layers, a plurality of residual blocks, and a second convolution layer, wherein
each of the plurality of residual blocks connects to a respective one of the at least three transition layers,
each of the plurality of residual block is used to generate a first initial feature,
the second convolution layer is used to generate a second initial feature, and
the second convolution layer includes interpolations and convolutions,
a fusion feature extraction structure including at least one up-sampling layer, wherein for each of the at least one up-sampling layer,
the up-sampling layer includes at least one transposed convolution,
the up-sampling layer connects to a residual block of the plurality of residual blocks, and
the up-sampling layer is used to generate a fusion feature by performing a secondary feature extraction on the first initial feature generated by the residual block; and
a feature fusion structure connecting to the second convolution layer and each of the at least one up-sampling layer, wherein the feature fusion structure is used to generate the initial binary image by fusing the second initial feature generated by the second convolution layer and the fusion feature generated by each of the at least one up-sampling layer.

### Claim 12 (depends on 11)

The system of claim 11, wherein
a first count of the at least one up-sampling layer equals to a second count of the plurality of residual blocks, and
for a first residual block deeper than a second residual block, a third count of at least one transposed convolution of an up-sampling layer connected to the first residual block is greater than or equals to a fourth count of at least one transposed convolution of an up-sampling layer connected to the second residual block.

### Claim 13 (depends on 12)

The system of claim 12, wherein
the initial feature extraction structure includes a superficial-layer residual block, an interlayer residual block, and a deep-layer residual block; and
the fusion feature extraction structure includes:
a first up-sampling layer connected to the superficial-layer residual block, wherein the first up-sampling layer includes one transposed convolution;
a second up-sampling layer connected to the interlayer residual block, wherein the second up-sampling layer includes two transposed convolutions; and
a third up-sampling layer connected to the deep-layer residual block, wherein the third up-sampling layer includes four transposed convolutions.

### Claim 14

A system for processing an X-ray image, comprising:
at least one storage device including a set of instructions for processing an X-ray image; and
at least one processor in communication with the at least one storage device, wherein when executing the set of instructions, the at least one processor is directed to cause the system to perform operations including:
obtaining an X-ray image;
determining an initial binary image indicating an initial metal region by using a trained metal detection model based on the X-ray image, wherein the trained metal detection model is trained from a neural network model for category semantic perception; and
determining a metal image indicating a target metal region based on the initial binary image, wherein the metal image is a target binary image, and the metal image includes information of a metal object in the X-ray image, and the information of the metal object includes at least one of a position of the metal object, a shape of the metal object, a size of the metal object, an orientation of the metal object, or a type of the metal object.

### Claim 15

A method for processing an X-ray image, comprising:
obtaining an X-ray image; and
determining a metal image based on the X-ray image by using a trained metal detection model, wherein the metal image includes information of a metal object in the X-ray image, and the information of the metal object includes at least one of a position of the metal object, a shape of the metal object, a size of the metal object, an orientation of the metal object, or a type of the metal object.

### Claim 16 (depends on 15)

The method of claim 15, wherein the trained metal detection model is determined based on a training process, the training process including:
obtaining a sample set including a plurality of sample pairs, wherein each sample pair of the plurality of sample pairs includes a sample X-ray image and a labeled metal image with respect to the X-ray image; and
obtaining the trained metal detection model by training a preliminary metal detection model based on the sample set.

### Claim 17 (depends on 16)

The method of claim 16, wherein the trained metal detection model includes a deep learning model.

### Claim 18 (depends on 17)

The method of claim 17, the deep learning model is a V-Net neural network model including a skip connection, wherein the skip connection includes:
combing at least one of at least one down-sampling feature corresponding to at least one down-sampling layer during up-sampling.

### Claim 19 (depends on 15)

The method of claim 15, wherein the determining the metal image based on the X-ray image by using the trained metal detection model includes:
determining, by using the trained metal detection model, a probability graph representing a probability that a pixel in the X-ray image belongs to a metal region; and
determining the metal image based on the probability graph.

### Claim 20 (depends on 15)

The method of claim 15, wherein the determining the metal image based on the X-ray image by using the trained metal detection model includes:
determining an initial binary image indicating an initial metal region by using the trained metal detection model, wherein the trained metal detection model is trained from a neural network model for category semantic perception; and
determining the metal image indicating a target metal region based on the initial binary image, wherein the metal image is a target binary image.