# US 2026/0073512 A1 · Ophthalmic medical image segmentation method and system and storage medium

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

- Number: US 2026/0073512 A1
- Document ID: US20260073512A1
- Status: Published application
- Kind: A1
- Application number: 19/214,033
- Filed: 2025-05-21
- Published: 2026-03-12
- Earliest claimed: 2024-09-09
- Assignees: Suzhou City University
- Applicants: Suzhou City University
- Inventors: Xuemei LU; Zhihao XU; Qingquan MENG; Lei GAO
- Web page: https://patentgrep.com/patent/US20260073512A1

## CPC

- G06T 7/0012: {Biomedical image inspection}
- A61B 3/14: Arrangements specially adapted for eye photography
- G06T 7/11: Region-based segmentation
- G06T 2207/10024: Color image
- G06T 2207/10101: Optical tomography; Optical coherence tomography [OCT]
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30041: Eye; Retina; Ophthalmic
- G06T 2207/30096: Tumor; Lesion
- Y02T 10/40: Engine management systems
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- A61B: DIAGNOSIS; SURGERY; IDENTIFICATION (usefulness limited to only animals A61D)
- Y02T: CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION

## Abstract

Ophthalmic medical image segmentation method includes dividing the medical image data into a training set and a test set according to an autonomously set proportion; constructing a convolutional neural network model adopting a U-shaped encoding and decoding structure based on an attention mechanism and a weighted loss function, and performing training; transmitting a to-be-segmented medical image to obtain a segmentation result, wherein the attention mechanism is introduced into the U-shaped encoding and decoding structure: a superficial layer feature map ILE of an encoder is subjected to convolution to obtain ILE-1, and a deep layer feature map IHD of a decoder is subjected to up-sampling and convolution to obtain IHD-1; the ILE-1 and the IHD-1 are multiplied to obtain IMul; the IMul and the IHD-1 are summed, and ISum is then output through an activation function; and the IMul and the ISum are spliced, and then output to a target layer.

## Claims

### Claim at position 1

1-17. (canceled)

### Claim 18

An ophthalmic medical image segmentation method, comprising the following steps:
acquiring medical image data of an ophthalmic lesion region, and dividing the medical image data into a training set and a test set according to an autonomously set proportion;
constructing a convolutional neural network model adopting a U-shaped encoding and decoding structure based on an attention mechanism and a weighted loss function;
performing training on the convolutional neural network model based on the training set and the test set; and
transmitting a to-be-segmented medical image to the trained convolutional neural network model to obtain a segmentation result,
wherein the U-shaped encoding and decoding structure includes an encoder, a jump connection part, a bottleneck layer and a decoder; the bottleneck layer is located between the encoder and the decoder; the encoder is connected to the decoder through the jump connection part; and wherein the attention mechanism introduced into the U-shaped encoding and decoding structure comprises the steps of:
subjecting a superficial layer feature map ILE of the encoder to a convolution operation to obtain a first processed feature map ILE-1, and subjecting a deep layer feature map IHD of the decoder to an up-sampling operation and the convolution operation to obtain a second processed feature map IHD-1;
multiplying the first processed feature map ILE-1 and the second processed feature map IHD-1 to obtain a multiplied feature map IMul;
summing the multiplied feature map IMul and the second processed feature map IHD-1, and outputting a result through an activation function to obtain a summed feature map ISum; and
splicing the multiplied feature map IMul and the summed feature map ISum, and then outputting to a target layer.

### Claim 19 (depends on 18)

The ophthalmic medical image segmentation method according to claim 18, wherein the weighted loss function is constructed using a multi-loss fusion method, which is as follows:
a multi-classified logistic loss function, a Dice loss function and a binary cross entropy loss function are fused to obtain the weighted loss function LossAll:
Loss
All
=
λ
1
⁢
L
logistic
(
Y
,
Y
ˆ
)
+
λ
2
⁢
L
Dice
(
Y
,
Y
ˆ
)
+
λ
3
⁢
L
BCE
(
Y
,
Y
ˆ
)
;
wherein λ1 is a weight of the multi-classified logistic loss function; Llogistic(Y,Ŷ) represents the multi-classified logistic loss function; λ2 represents a weight of the Dice loss function; LDice(Y,Ŷ) represents the Dice loss function; λ3 represents a weight of the binary cross entropy loss function; LBCE(Y,Ŷ) represents the binary cross entropy loss function; λ1, λ2, λ3 are all real numbers, and λ1+λ2+λ3=1; and initial values of λ1, λ2, λ3 are 0.4, 0.4, and 0.2 respectively.

### Claim 20 (depends on 19)

The ophthalmic medical image segmentation method according to claim 19, wherein the multi-classified logistic loss function Llogistic(Y,Ŷ) is as follows:
L
logistic
(
Y
,
Y
ˆ
)
=
-
∑
x
i
∈
Ω
i
=
1
N
ω
logloss
(
x
i
)
·
Y
⁡
(
x
i
)
·
log
[
Y
ˆ
(
x
i
)
]
;
ω
logloss
(
x
i
)
=
{
10
,
x
i
∈
ROI
1
,
x
i
∈
BACKGROUND
;
wherein N is the number of pixel points, xi represents the ith pixel point, Ω represents a category of classification, Σ is a summing symbol, Y(xi) represents a network segmentation result of the ith pixel point xi, log represents that a logarithm is taken, Ŷ(xi) represents an expected segmentation result of the ith pixel point xi, ROI represents a lesion region of interest, and BACKGROUND represents a background region;
the Dice loss function LDice(Y,Ŷ) is as follows:
L
Dice
(
Y
,
Y
ˆ
)
=
1
-
2
·
[
Y
⁡
(
x
i
)
⊙
Y
^
(
x
i
)
]
∑
i
=
1
N
⁢
❘
"\[LeftBracketingBar]"
Y
⁡
(
x
i
)
❘
"\[RightBracketingBar]"
+
∑
i
=
1
N
⁢
❘
"\[LeftBracketingBar]"
Y
^
(
x
i
)
❘
"\[RightBracketingBar]"
;
wherein ⊙ represents that the corresponding pixels are multiplied, and ∥ represents that an absolute value is taken; and
the binary cross entropy loss function LBCE(Y,Ŷ) is as follows:
L
BCE
(
Y
,
Y
ˆ
)
=
-
1
N
⁢
∑
i
=
1
N
{
Y
⁡
(
x
i
)
⁢
log
[
Y
ˆ
(
x
i
)
]
}
+
[
1
-
Y
⁡
(
x
i
)
]
·
log
[
1
-
Y
ˆ
(
x
i
)
]
.

### Claim 21 (depends on 18)

The ophthalmic medical image segmentation method according to claim 18, wherein the encoder is configured with a plurality of layers, and each layer is subjected to the convolution operation, a batch normalization operation and a maximum pooling operation; and the size of a convolution kernel and the number of operations for each layer to perform the convolution operation can be set autonomously.

### Claim 22 (depends on 21)

The ophthalmic medical image segmentation method according to claim 21, wherein the encoder is configured with four layers, each layer is subjected to the convolution operation twice based on a convolution layer having a convolution kernel size of 3*3 and a step size of 1, and each layer is subjected to the maximum pooling operation based on a pooling layer having a pooling kernel size of 2*2.

### Claim 23 (depends on 18)

The ophthalmic medical image segmentation method according to claim 18, wherein the medical image data comprises OCT image data of ophthalmic choroidal neovascularization and fundus color image data of glaucoma, and the OCT image data and the fundus color image data both contain original image data and corresponding gold standard image data.

### Claim 24 (depends on 23)

The ophthalmic medical image segmentation method according to claim 23, wherein background pixel values in the OCT image data and the fundus color image data are assigned to 0, and the same pixel values are assigned to the lesion regions of the respective categories in the corresponding gold standard image data and incremented sequentially according to their categories.

### Claim 25 (depends on 18)

A non-transitory computer-readable storage medium, having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method according to claim 18.

### Claim 26 (depends on 19)

A non-transitory computer-readable storage medium, having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method according to claim 19.

### Claim 27 (depends on 20)

A non-transitory computer-readable storage medium, having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method according to claim 20.

### Claim 28 (depends on 21)

A non-transitory computer-readable storage medium, having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method according to claim 21.

### Claim 29 (depends on 22)

A non-transitory computer-readable storage medium, having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method according to claim 22.

### Claim 30 (depends on 23)

A non-transitory computer-readable storage medium, having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method according to claim 23.

### Claim 31 (depends on 24)

A non-transitory computer-readable storage medium, having stored a computer program therein, the computer program, when executed by a processor, implementing the steps of the ophthalmic medical image segmentation method according to claim 24.

### Claim 32

An ophthalmic medical image segmentation system, comprising:
a data acquisition module, configured to acquire medical image data of an ophthalmic lesion region, and divide the medical image data into a training set and a test set according to an autonomously set proportion;
a model construction module, configured to construct a weighted loss function by using a multi-loss fusion manner and construct a convolutional neural network model adopting a U-shaped encoding and decoding structure based on an attention mechanism and the weighted loss function;
a model training module, configured to perform training on the convolutional neural network model based on the training set and the test set; and
an image segmentation module, configured to transmit a to-be-segmented medical image to the trained convolutional neural network model to obtain a segmentation result,
wherein the U-shaped encoding and decoding structure includes an encoder, a bottleneck layer, a decoder and a jump connection part; the bottleneck layer is located between the encoder and the decoder; the attention mechanism is introduced into the decoder and the jump connection part, said attention mechanism configured to:
subject a superficial layer feature map ILE of the encoder to a convolution operation to obtain a first processed feature map ILE-1, and subject a deep layer feature map IHD of the decoder to an up-sampling operation and the convolution operation to obtain a second processed feature map IHD-1;
multiply the first processed feature map ILE-1 and the second processed feature map IHD-1 to obtain a multiplied feature map IMul;
sum the multiplied feature map IMul and the second processed feature map IHD-1, and output a result through an activation function to obtain a summed feature map ISum; and
splice the multiplied feature map IMul and the summed feature map ISum, and then output to a target layer.
