CN114693928A - Blood vessel segmentation method and imaging method of OCTA image - Google Patents

Blood vessel segmentation method and imaging method of OCTA image Download PDF

Info

Publication number
CN114693928A
CN114693928A CN202210303004.1A CN202210303004A CN114693928A CN 114693928 A CN114693928 A CN 114693928A CN 202210303004 A CN202210303004 A CN 202210303004A CN 114693928 A CN114693928 A CN 114693928A
Authority
CN
China
Prior art keywords
decoder
image
encoder
vessel segmentation
transformation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210303004.1A
Other languages
Chinese (zh)
Inventor
沈海澜
熊雨晨
陈再良
戴培山
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN202210303004.1A priority Critical patent/CN114693928A/en
Publication of CN114693928A publication Critical patent/CN114693928A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Eye Examination Apparatus (AREA)

Abstract

The invention discloses a blood vessel segmentation method of an OCTA image, which comprises the steps of obtaining a retina OCTA image data set and dividing the retina OCTA image data set into a labeled image and a non-labeled image; constructing an encoder and a dual decoder; selecting a plurality of images with labels and images without labels; inputting the image with the label into an encoder and a main decoder to obtain a blood vessel segmentation result, calculating the loss of a supervision part, and reversely propagating and updating parameters of the encoder and the main decoder; inputting the label-free image into an encoder, a main decoder or an auxiliary decoder to obtain a blood vessel segmentation result, calculating consistency loss and reversely propagating and updating parameters of the encoder and the auxiliary decoder; repeating the steps to obtain a final blood vessel segmentation model of the OCTA image; and (3) carrying out vessel segmentation on the actual OCTA image by adopting a vessel segmentation model of the OCTA image. The invention also discloses an imaging method comprising the OCTA image blood vessel segmentation method. The invention has high reliability, good practicability and better segmentation effect.

Description

Blood vessel segmentation method and imaging method of OCTA image
Technical Field
The invention belongs to the field of image processing, and particularly relates to a blood vessel segmentation method and an imaging method of an OCTA image.
Background
With the development of economic technology and the improvement of living standard of people, people pay more and more attention to health. With the development of image processing technology and deep learning technology, more and more deep learning schemes and image processing schemes are applied to the medical field, and bring endless convenience to people.
The state of retinal blood vessels is of great significance in both clinical and experimental studies. Currently, with the development of technology, a non-invasive imaging technology, Optical Coherence Tomography (oct), is emerging. The technology is based on the existence of flowing blood cells in the fundus blood vessel, the same cross section is imaged for a plurality of times, blood flow signals are calculated, and then the blood flow is reconstructed in a three-dimensional mode to recover the image of the fundus blood vessel; this technique thus enables observation of the morphological structure of the fine capillaries of the fundus. Therefore, it is very important to automatically detect and segment blood vessels in the OCTA image.
There are some research methods for retinal OCTA image vessel segmentation, which can be roughly divided into a conventional segmentation method and a segmentation method based on deep learning.
(1) Conventional segmentation method
In 2016, Gopinath et al proposed a method for classifying glaucoma using the capillary characteristics of the OCTA images; the method practices the segmentation of capillary vessels in OCTA, but the segmentation effect is not ideal enough, and a large error can be caused. Eladawi et al in 2017 propose a method for segmenting blood vessels in an OCTA image by using a markov-gibbs random field, wherein the method integrates an apparent model and a spatial model in addition to a prior probability model of the OCTA image. A high-order MGRF (HO-MGRF) model is used to consider spatial information in addition to the first-order intensity model to overcome low contrast between blood vessels and other tissues; the method achieves higher accuracy, but the segmentation precision of some micro-vessel regions is not ideal enough. In 2020, Le et al propose a filter based on Hessian matrix and an Ostu threshold method to detect and enhance small blood vessels; the method is simple but the segmentation effect is not ideal. These filter-based approaches typically rely heavily on manual parameter adjustments in the implementation.
(2) Segmentation method based on deep learning
Deng et al, university of iowa in 2019, proposed a method for measuring vascular loss in OCTA; the method skillfully avoids the position where the micro-vessel structure is difficult to be segmented, can accurately measure the blood vessel missing area in the OCTA, but cannot obtain the parameter information such as the density of the capillary vessels because the blood vessel network cannot be segmented accurately. Mou et al propose a unified curve structure segmentation network based on U-net's channel and spatial attention; however, the method is a fully supervised method, and the OCTA is particularly difficult to label for the vascular network, and requires an experienced doctor to spend a great deal of time and energy on manual labeling, thereby increasing the burden of the doctor. In 2020, Li et al propose an end-to-end Image Projection Network (IPN) structure, which can realize three-dimensional to two-dimensional image segmentation in an OCTA image; however, this method only segments the large vessels on the OCTA, and does not segment the microvascular network. Mou et al propose a curve-structured segmentation network, which contains an attention mechanism in the encoder and decoder to learn the rich hierarchical representation of the curve structure, but the method is based on a fully supervised segmentation method, and relies on large-scale labeled training data, but the distribution of the large and small blood vessels of the OCTA images is densely staggered, the image contrast is not high, the labeling is more time-consuming and labor-consuming, and a large amount of labeled data is difficult to collect. In 2021, Xu et al proposed a segmentation method of local supervised learning, which performs local labeling on a part of images, and which adopts an active learning strategy to select a patch with the largest information content for further labeling, but the method is more complicated and the blood vessel segmentation accuracy needs to be further improved.
Disclosure of Invention
One of the purposes of the invention is to provide a blood vessel segmentation method of an OCTA image, which is based on image transformation and characteristic disturbance, and has high reliability, good practicability and good segmentation effect.
Another object of the present invention is to provide an imaging method including the blood vessel segmentation method for the OCTA image.
The invention provides a blood vessel segmentation method of an OCTA image, which comprises the following steps:
s1, obtaining a retina OCTA image data set, and dividing the retina OCTA image data set into a labeled image and a non-labeled image;
s2, constructing an encoder and a double decoder so as to form a segmentation model; the dual decoder includes a primary decoder and a secondary decoder;
s3, selecting a plurality of images with labels and images without labels from the images obtained in the step S1;
s4, sequentially inputting the images with the labels into an encoder and a main decoder to obtain a blood vessel segmentation result, calculating loss of a supervision part, and performing reverse propagation to update parameters of the encoder and the main decoder;
s5, inputting the label-free image into an encoder, inputting the output result of the encoder into a main decoder or an auxiliary decoder to obtain a blood vessel segmentation result, calculating consistency loss, and performing reverse propagation to update parameters of the encoder and the auxiliary decoder;
s6, repeating the steps S3-S5 until the condition that the model training is finished is reached, and obtaining a final blood vessel segmentation model of the OCTA image;
and S7, performing blood vessel segmentation on the actual OCTA image by adopting the blood vessel segmentation model of the final OCTA image obtained in the step S6.
Step S1, obtaining a retinal OCTA image dataset and dividing the retinal OCTA image dataset into a labeled image and a non-labeled image, specifically obtaining a retinal OCTA image dataset and dividing the retinal OCTA image dataset into a labeled image
Figure BDA0003563567560000041
And unlabelled images
Figure BDA0003563567560000042
Wherein the content of the first and second substances,
Figure BDA0003563567560000043
for the i-th image with a label,
Figure BDA0003563567560000044
for the tag carried by the ith tagged image,
Figure BDA0003563567560000045
for the ith unlabeled image, NLFor the total number of tagged images, NUThe total number of unlabeled images.
Constructing an encoder and a dual decoder as described in step S2, thereby constructing a segmentation model; the dual decoder includes a primary decoder and a secondary decoder; specifically, an encoder and a double decoder are constructed, so that a segmentation model is formed; the dual decoder includes a primary decoder and a secondary decoder; the parameter of the encoder is thetaeThe parameter of the main decoder is thetamThe parameter of the auxiliary decoder is thetaa。
The step S4 of sequentially inputting the labeled images into the encoder and the main decoder to obtain the blood vessel segmentation result, calculating the loss of the monitoring part, and performing back propagation to update the parameters of the encoder and the main decoder specifically includes the following steps:
sequentially inputting the images with the labels into an encoder and a main decoder to obtain a blood vessel segmentation result;
supervision of partial loss LsDefined as cross entropy pixel by pixel as a loss function; cross entropy loss function LCE(XL,YL；θe,θm) Is defined as
Figure BDA0003563567560000046
In the formula XLFor images with labels, YLFor labels corresponding to the images with labels, θeAs a parameter of the encoder, θmIs the parameter of the main decoder, h is the height of the picture, w is the width of the picture, c is the number of classes, yLIs a real label, log is a logarithm taking 2 as a base number,
Figure BDA0003563567560000047
as a function of the decoding operation of the primary decoder,
Figure BDA0003563567560000048
as a function of the encoding operation of the encoder, xLLabeling the sample;
the parameters of the encoder and the main decoder are updated by back propagation through the gradient descent algorithm.
Step S5, inputting the unlabeled image into the encoder, and inputting the output result of the encoder into the main decoder or the auxiliary decoder to obtain the blood vessel segmentation result, calculating the consistency loss, and performing back propagation to update the parameters of the encoder and the auxiliary decoder, specifically including the following steps:
A. sequentially inputting the unlabeled images into the encoder and the main decoder to obtain the blood vessel segmentation result of the unlabeled main decoder
Figure BDA0003563567560000051
B. Randomly transforming unlabeled images
Figure BDA0003563567560000052
And converting the image without label
Figure BDA0003563567560000053
Inputting the data into an encoder to obtain the random transformation coding characteristics without labels
Figure BDA0003563567560000054
Figure BDA0003563567560000055
For making a random change
Figure BDA0003563567560000056
The operation function of (1);
C. random transformation coding features without labels
Figure BDA0003563567560000057
Inputting the result into an auxiliary decoder to obtain a blood vessel segmentation result of the label-free random transformation auxiliary decoder
Figure BDA0003563567560000058
D. The obtained label-free random transformation is assisted with the vessel segmentation result of the decoder
Figure BDA0003563567560000059
Performing inverse transformation to obtain the result of inverse transformation vessel segmentation of the unlabeled random transformation auxiliary decoder
Figure BDA00035635675600000510
E. Based on unlabeled random transform, assisting decoder to inversely transform vessel segmentation result
Figure BDA00035635675600000511
And the unlabeled main decoder blood vessel segmentation result
Figure BDA00035635675600000512
Calculating a consistency loss based on the image transformation;
F. random transformation coding features without labels
Figure BDA00035635675600000513
Inverse transformation is carried out to obtain the coding characteristics of label-free random transformation inverse transformation
Figure BDA00035635675600000514
And inputting the result into an auxiliary decoder to obtain a blood vessel segmentation result of the unlabeled random transformation inverse transformation auxiliary decoder
Figure BDA00035635675600000515
G. According to nothingLabel random transformation inverse transformation auxiliary decoder blood vessel segmentation result
Figure BDA00035635675600000516
And the unlabeled main decoder blood vessel segmentation result
Figure BDA00035635675600000517
Calculating consistency loss based on the characteristic disturbance;
H. and performing back propagation through a gradient descent algorithm, and updating parameters of the encoder and the auxiliary decoder.
Random transformation as described in step B
Figure BDA0003563567560000061
Specifically, random grid transformation and random rotation transformation are adopted for random transformation; wherein the random grid transformation adopts 2 × 2 and 3 × 3 grids, and the random rotation range is [ -180 deg., 180 deg. °]。
Step E, calculating the consistency loss based on the image transformation, specifically, calculating the consistency loss L based on the image transformation by adopting the following formulad(XU；θe,θm,θa)：
Figure BDA0003563567560000062
In the formula XUIs a no-label image; thetaeIs a parameter of the encoder; theta.theta.mIs a parameter of the primary decoder; thetaaIs a parameter of the secondary decoder; mse () is the mean square error function; x is the number ofUIs a currently computed label-free image;
Figure BDA0003563567560000063
for making a random change
Figure BDA0003563567560000064
The operation function of (1);
Figure BDA0003563567560000065
is changed randomlyChangeable pipe
Figure BDA0003563567560000066
The inverse transform operation function of (1);
Figure BDA0003563567560000067
an encoding operation function for the encoder;
Figure BDA0003563567560000068
is a decoding operation function of the primary decoder;
Figure BDA0003563567560000069
is a function of the decoding operation of the secondary decoder.
And G, calculating the consistency loss based on the characteristic disturbance, specifically adopting the following formula to calculate the consistency loss L based on the characteristic disturbancef(XU；θe,θm,θa)：
Figure BDA00035635675600000610
In the formula XUIs a no-label image; thetaeIs a parameter of the encoder; thetamIs a parameter of the primary decoder; thetaaIs a parameter of the secondary decoder; mse () is the mean square error function; x is the number ofUIs a currently computed label-free image;
Figure BDA00035635675600000611
for making a random change
Figure BDA00035635675600000612
The operation function of (1);
Figure BDA00035635675600000613
for making a random change
Figure BDA00035635675600000614
The inverse transform operation function of (1);
Figure BDA00035635675600000615
an encoding operation function for the encoder;
Figure BDA00035635675600000616
a decoding operation function for the primary decoder;
Figure BDA00035635675600000617
is a function of the decoding operation of the secondary decoder.
In step S6, the final vessel segmentation model of the OCTA image is specifically obtained by averaging the segmentation results of the primary decoder and the secondary decoder, respectively, so as to obtain a final segmentation result.
The invention also discloses an imaging method comprising the blood vessel segmentation method of the OCTA image, which specifically comprises the following steps:
(1) obtaining an OCTA image;
(2) segmenting the blood vessels in the OCTA image acquired in the step (1) by adopting the blood vessel segmentation method of the OCTA image;
(3) according to the segmentation result obtained in the step (2), marking and carrying out secondary imaging on the blood vessel segmentation result obtained in the step (1);
(4) and outputting the OCTA image with the blood vessel marker to finish the final imaging process.
According to the vessel segmentation method and the imaging method of the OCTA image, the consistency based on the characteristic disturbance is introduced under the condition that the network structure is not required to be modified, the potential information of the label-free data is fully utilized through double consistency regularization, and multi-scale random grid transformation is adopted, so that network overfitting is relieved; the method improves the segmentation precision of the model under the condition of only using limited label data and a large amount of label-free data, so that the method is more suitable for scenes with few OCTA vessel image labels in practice, and has high reliability, good practicability and better segmentation effect.
Drawings
FIG. 1 is a schematic flow chart of the segmentation method of the present invention.
FIG. 2 is a diagram illustrating the effect of the random transformation according to the present invention.
Fig. 3 is a schematic view of the visualization of the segmentation result of the present invention.
Fig. 4 is a method flow diagram of the imaging method of the present invention.
Detailed Description
Fig. 1 is a schematic flow chart of the segmentation method of the present invention: the invention provides a blood vessel segmentation method of an OCTA image, which comprises the following steps:
s1, obtaining a retina OCTA image data set, and dividing the retina OCTA image data set into a labeled image and a non-labeled image; in particular, a retina OCTA image data set is acquired and divided into labeled images
Figure BDA0003563567560000081
And unlabelled images
Figure BDA0003563567560000082
Wherein the content of the first and second substances,
Figure BDA0003563567560000083
for the i-th image with a label,
Figure BDA0003563567560000084
for the tag carried by the ith tagged image,
Figure BDA0003563567560000085
for the ith unlabeled image, NLFor the total number of tagged images, NUIs the total number of unlabeled images;
s2, constructing an encoder and a double decoder so as to form a segmentation model; the dual decoder includes a primary decoder and a secondary decoder; specifically, an encoder and a double decoder are constructed, so that a segmentation model is formed; the dual decoder includes a primary decoder and a secondary decoder; the parameter of the encoder is thetaeThe parameter of the main decoder is thetamThe parameter of the auxiliary decoder is thetaa；
S3, selecting a plurality of images with labels and images without labels from the images obtained in the step S1;
s4, sequentially inputting the images with the labels into an encoder and a main decoder to obtain a blood vessel segmentation result, calculating loss of a supervision part, and performing reverse propagation to update parameters of the encoder and the main decoder; sequentially inputting the images with the labels into an encoder and a main decoder to obtain a blood vessel segmentation result;
supervision of partial loss LsDefined as cross entropy pixel by pixel as a loss function; cross entropy loss function LCE(XL,YL；θe,θm) Is defined as
Figure BDA0003563567560000086
In the formula XLFor images with labels, YLFor labels corresponding to the images with labels, θeAs a parameter of the encoder, θmIs the parameter of the main decoder, h is the height of the picture, w is the width of the picture, c is the number of classes, yLIs a real label, log is a logarithm taking 2 as a base number,
Figure BDA0003563567560000087
as a function of the decoding operation of the primary decoder,
Figure BDA0003563567560000088
as a function of the encoding operation of the encoder, xLLabeling the sample;
performing back propagation through a gradient descent algorithm, thereby updating parameters of the encoder and the main decoder;
s5, inputting the label-free image into an encoder, inputting the output result of the encoder into a main decoder or an auxiliary decoder to obtain a blood vessel segmentation result, calculating consistency loss, and performing reverse propagation to update parameters of the encoder and the auxiliary decoder; the method specifically comprises the following steps:
A. sequentially inputting the unlabeled images into the encoder and the main decoder to obtain the blood vessel segmentation result of the unlabeled main decoder
Figure BDA0003563567560000091
B. Will have no effect onRandom transformation of label image
Figure BDA0003563567560000092
And converting the image without label
Figure BDA0003563567560000093
Inputting the data into an encoder to obtain the random transformation coding characteristics without labels
Figure BDA0003563567560000094
Figure BDA0003563567560000095
For making a random change
Figure BDA0003563567560000096
The operation function of (1); random transformation
Figure BDA0003563567560000097
Specifically, random grid transformation and random rotation transformation are adopted for random transformation; wherein the random grid transformation adopts 2 × 2 and 3 × 3 grids, and the random rotation range is [ -180 °,180 °]The specific effect is shown in figure 2;
C. random transformation coding features without labels
Figure BDA0003563567560000098
Inputting the result into an auxiliary decoder to obtain a blood vessel segmentation result of the label-free random transformation auxiliary decoder
Figure BDA0003563567560000099
D. For the obtained label-free random transformation, assisting the decoder to segment the blood vessel
Figure BDA00035635675600000910
Performing inverse transformation to obtain the result of inverse transformation vessel segmentation of the unlabeled random transformation auxiliary decoder
Figure BDA00035635675600000911
E. Based on unlabeled random transform, assisting decoder to inversely transform vessel segmentation result
Figure BDA00035635675600000912
And the unlabeled main decoder blood vessel segmentation result
Figure BDA00035635675600000913
Calculating a consistency loss based on the image transformation; specifically, the consistency loss L based on image transformation is calculated by adopting the following formulad(XU；θe,θm,θa)：
Figure BDA00035635675600000914
In the formula XUIs a no-label image; thetaeIs a parameter of the encoder; thetamIs a parameter of the primary decoder; thetaaIs a parameter of the secondary decoder; mse () is the mean square error function; x is the number ofUIs a currently computed label-free image;
Figure BDA00035635675600000915
for making a random change
Figure BDA00035635675600000916
The operation function of (1);
Figure BDA00035635675600000917
for making a random change
Figure BDA00035635675600000918
The inverse transform operation function of (1);
Figure BDA00035635675600000919
an encoding operation function for the encoder;
Figure BDA00035635675600000920
is a decoding operation function of the primary decoder;
Figure BDA00035635675600000921
is a decoding operation function of the secondary decoder;
F. random transformation coding features without labels
Figure BDA0003563567560000101
Inverse transformation is carried out to obtain the coding characteristics of label-free random transformation inverse transformation
Figure BDA0003563567560000102
And inputting the result into an auxiliary decoder to obtain a blood vessel segmentation result of the unlabeled random transformation inverse transformation auxiliary decoder
Figure BDA0003563567560000103
G. Assisting a decoder vessel segmentation result according to label-free random transformation inverse transformation
Figure BDA0003563567560000104
And the unlabeled main decoder blood vessel segmentation result
Figure BDA0003563567560000105
Calculating consistency loss based on the characteristic disturbance; specifically, the consistency loss L based on the characteristic disturbance is calculated by the following formulaf(XU；θe,θm,θa)：
Figure BDA0003563567560000106
In the formula XUIs a no-label image; thetaeIs a parameter of the encoder; thetamIs a parameter of the primary decoder; thetaaIs a parameter of the secondary decoder; mse () is the mean square error function; x is the number ofUIs a currently computed label-free image;
Figure BDA0003563567560000107
for making a random change
Figure BDA0003563567560000108
The operation function of (1);
Figure BDA0003563567560000109
for making a random change
Figure BDA00035635675600001010
The inverse transform operation function of (1);
Figure BDA00035635675600001011
an encoding operation function for the encoder;
Figure BDA00035635675600001012
is a decoding operation function of the primary decoder;
Figure BDA00035635675600001013
is a decoding operation function of the secondary decoder;
H. performing back propagation through a gradient descent algorithm, and updating parameters of the encoder and the auxiliary decoder;
s6, repeating the steps S3-S5 until the condition that the model training is finished is reached, and obtaining a final blood vessel segmentation model of the OCTA image; in specific implementation, the respective segmentation results of the main decoder and the auxiliary decoder are averaged to obtain a final segmentation result;
and S7, performing blood vessel segmentation on the actual OCTA image by adopting the blood vessel segmentation model of the final OCTA image obtained in the step S6.
The following examples are combined to compare the effect of the segmentation method of the present invention with that of the existing method:
comparing the method provided by the invention with the existing segmentation method: the full-supervision method uses all labels of a training set for training, and other methods use 3.3% of labeled data, and the segmentation result, the accuracy of the group Truth, the Dice coefficient and the error discovery rate are used as evaluation criteria; the specific segmentation results are shown in table 1:
TABLE 1 schematic table of OCTA vessel segmentation results
Segmentation method Accuracy (%) Dice coefficient (%) Error discovery Rate (%)
Full-supervision method 91.21 75.11 19.97
Semi-supervised method 90.70 73.48 17.49
The method of the invention 91.34 75.61 15.99
As can be seen from Table 1, the method of the present invention is superior to the existing two different methods in three indexes, including the fully supervised method, and more accurate segmentation results are obtained through the constraint of double consistency. Fig. 3 compares the results of the semi-supervised method and the method of the present invention, which are the original image, the group Truth, the semi-supervised method result and the method result from left to right, respectively, and it can be seen from the figure that the method of the present invention performs better on the continuity of the segmented blood vessels.
Fig. 4 is a schematic method flow diagram of the imaging method of the present invention: the imaging method comprising the blood vessel segmentation method of the OCTA image, provided by the invention, specifically comprises the following steps:
(1) obtaining an OCTA image;
(2) segmenting the blood vessels in the OCTA image acquired in the step (1) by adopting the blood vessel segmentation method of the OCTA image;
(3) according to the segmentation result obtained in the step (2), marking and carrying out secondary imaging on the blood vessel segmentation result obtained in the step (1);
(4) and outputting the OCTA image with the blood vessel mark to finish the final imaging process.
The imaging method provided by the invention can be applied to the existing OCTA image equipment; in particular application, the imaging method of the invention is integrated into the equipment; when the equipment works, firstly, fundus images of inspectors are acquired according to the prior art, and the existing OCTA images are obtained; then, the device segments the blood vessels in the obtained OCTA image (at this time, the blood vessel segmentation and marking are not finished) by adopting the blood vessel segmentation method of the OCTA image provided by the invention; then, according to the segmentation result, highlighting and marking the blood vessel on the OCTA image (such as marking by using a highlighted color) and carrying out secondary imaging; the secondary imaging can obtain an OCTA image with the blood vessel prominent mark; and finally, outputting the obtained OCTA image with the blood vessel protrusion mark, thereby finishing the final imaging work. At this time, the existing OCTA image apparatus becomes a multifunctional OCTA image apparatus with vessel segmentation and labeling.

Claims (10)

1. A blood vessel segmentation method of an OCTA image comprises the following steps:
s1, obtaining a retina OCTA image data set, and dividing the retina OCTA image data set into a labeled image and a non-labeled image;
s2, constructing an encoder and a double decoder so as to form a segmentation model; the dual decoder includes a primary decoder and a secondary decoder;
s3, selecting a plurality of images with labels and images without labels from the images obtained in the step S1;
s4, sequentially inputting the images with the labels into the encoder and the main decoder to obtain a blood vessel segmentation result, calculating loss of a supervision part, and performing reverse propagation to update parameters of the encoder and the main decoder;
s5, inputting the label-free image into an encoder, inputting the output result of the encoder into a main decoder or an auxiliary decoder to obtain a blood vessel segmentation result, calculating consistency loss, and performing reverse propagation to update parameters of the encoder and the auxiliary decoder;
s6, repeating the steps S3-S5 until the condition that the model training is finished is reached, and obtaining a final blood vessel segmentation model of the OCTA image;
and S7, performing blood vessel segmentation on the actual OCTA image by adopting the blood vessel segmentation model of the final OCTA image obtained in the step S6.
2. The method of segmenting blood vessels in an OCTA image as claimed in claim 1, wherein the step S1 is performed by acquiring a retinal OCTA image dataset and dividing the retinal OCTA image dataset into a labeled image and a non-labeled image, in particular by acquiring a retinal OCTA image dataset and dividing the retinal OCTA image dataset into a labeled image
Figure FDA0003563567550000011
And unlabelled images
Figure FDA0003563567550000012
Wherein the content of the first and second substances,
Figure FDA0003563567550000013
for the i-th image with a label,
Figure FDA0003563567550000014
for the tag carried by the ith tagged image,
Figure FDA0003563567550000015
for the ith unlabeled image, NLFor the total number of tagged images, NUThe total number of unlabeled images.
3. The method of segmenting blood vessels in an OCTA image according to claim 2, wherein the encoder and the dual decoder are constructed in step S2, thereby forming a segmentation model; the dual decoder includes a primary decoder and a secondary decoder; specifically, an encoder and a double decoder are constructed, so that a segmentation model is formed; the dual decoder includes a primary decoder and a secondary decoder; the parameter of the encoder is thetaeThe parameter of the main decoder is thetamThe parameter of the auxiliary decoder is thetaa。
4. The method for segmenting blood vessels of an OCTA image according to claim 3, wherein the step S4 of inputting the labeled images into the encoder and the main decoder in sequence to obtain the result of segmenting blood vessels, calculating the loss of the supervised part, and performing back propagation to update the parameters of the encoder and the main decoder comprises the following steps:
sequentially inputting the images with the labels into an encoder and a main decoder to obtain a blood vessel segmentation result;
supervision part loss LsDefined as cross entropy pixel by pixel as a loss function; cross entropy loss function LCE(XL,YL；θe,θm) Is defined as
Figure FDA0003563567550000021
In the formula XLFor images with labels, YLFor labels corresponding to the images with labels, θeBeing a parameter of the encoder, thetamIs the parameter of the main decoder, h is the height of the picture, w is the width of the picture, c is the number of classes, yLIs a real label, log is a logarithm taking 2 as a base number,
Figure FDA0003563567550000022
as a function of the decoding operation of the primary decoder,
Figure FDA0003563567550000023
as a function of the encoding operation of the encoder, xLLabeling the sample;
the parameters of the encoder and the main decoder are updated by back propagation through the gradient descent algorithm.
5. The method of segmenting blood vessels in an OCTA image according to claim 4, wherein the step S5 includes inputting the unlabeled image into the encoder, inputting the output result of the encoder into the primary decoder or the secondary decoder to obtain the segmentation result of the blood vessels, calculating the consistency loss, and performing back propagation to update the parameters of the encoder and the secondary decoder, and specifically includes the following steps:
A. sequentially inputting the unlabeled images into the encoder and the main decoder to obtain the blood vessel segmentation result of the unlabeled main decoder
Figure FDA0003563567550000024
B. Randomly transforming unlabeled images
Figure FDA0003563567550000031
And converting the image without label
Figure FDA0003563567550000032
Inputting the data into an encoder to obtain the random transformation coding characteristics without labels
Figure FDA0003563567550000033
Figure FDA0003563567550000034
Is a random transformation
Figure FDA0003563567550000035
The operation function of (1);
C. random transformation coding features without labels
Figure FDA0003563567550000036
Inputting the data into an auxiliary decoder to obtain a label-free random transformation auxiliary solutionEncoder vessel segmentation results
Figure FDA0003563567550000037
D. The obtained label-free random transformation is assisted with the vessel segmentation result of the decoder
Figure FDA0003563567550000038
Performing inverse transformation to obtain the result of inverse transformation vessel segmentation of the unlabeled random transformation auxiliary decoder
Figure FDA0003563567550000039
E. Based on unlabeled random transform, assisting decoder to inversely transform vessel segmentation result
Figure FDA00035635675500000310
And the unlabeled main decoder blood vessel segmentation result
Figure FDA00035635675500000311
Calculating a consistency loss based on the image transformation;
F. random transformation coding features without labels
Figure FDA00035635675500000312
Inverse transformation is carried out to obtain the coding characteristics of label-free random transformation inverse transformation
Figure FDA00035635675500000313
And inputting the result into an auxiliary decoder to obtain a blood vessel segmentation result of the unlabeled random transformation inverse transformation auxiliary decoder
Figure FDA00035635675500000314
G. Assisting the decoder vessel segmentation result according to the label-free random transformation inverse transformation
Figure FDA00035635675500000315
And no markSigned master decoder vessel segmentation result
Figure FDA00035635675500000316
Calculating consistency loss based on the characteristic disturbance;
H. and performing back propagation through a gradient descent algorithm, and updating parameters of the encoder and the auxiliary decoder.
6. The method of segmenting blood vessels in OCTA image as claimed in claim 5, wherein the random transformation in step B
Figure FDA00035635675500000317
Specifically, random grid transformation and random rotation transformation are adopted for random transformation; wherein the random grid transformation adopts 2 × 2 and 3 × 3 grids, and the random rotation range is [ -180 °,180 °]。
7. The method of segmenting blood vessels in an OCTA image as claimed in claim 6, wherein the step E of calculating the consistency loss based on the image transformation specifically adopts the following formula to calculate the consistency loss L based on the image transformationd(XU；θe,θm,θa)：
Figure FDA0003563567550000041
In the formula XUIs a no-label image; thetaeIs a parameter of the encoder; thetamIs a parameter of the primary decoder; theta.theta.aIs a parameter of the secondary decoder; mse () is the mean square error function; x is the number ofUIs a currently computed label-free image;
Figure FDA0003563567550000042
for making a random change
Figure FDA0003563567550000043
The operation function of (1);
Figure FDA0003563567550000044
is a random transformation
Figure FDA0003563567550000045
The inverse transform operation function of (1);
Figure FDA0003563567550000046
an encoding operation function for the encoder;
Figure FDA0003563567550000047
a decoding operation function for the primary decoder;
Figure FDA0003563567550000048
is a function of the decoding operation of the secondary decoder.
8. The method of segmenting blood vessels in OCTA image as claimed in claim 7, wherein the step G of calculating the consistency loss based on the characteristic disturbance specifically adopts the following formulaf(XU；θe,θm,θa)：
Figure FDA0003563567550000049
In the formula XUIs a no-label image; thetaeIs a parameter of the encoder; thetamIs a parameter of the primary decoder; thetaaIs a parameter of the secondary decoder; mse () is the mean square error function; x is a radical of a fluorine atomUIs the currently computed label-free image;
Figure FDA00035635675500000410
for making a random change
Figure FDA00035635675500000411
The operation function of (1);
Figure FDA00035635675500000412
for making a random change
Figure FDA00035635675500000413
The inverse transform operation function of (1);
Figure FDA00035635675500000414
an encoding operation function for the encoder;
Figure FDA00035635675500000415
is a decoding operation function of the primary decoder;
Figure FDA00035635675500000416
is a function of the decoding operation of the secondary decoder.
9. The method of segmenting blood vessels in an OCTA image according to claim 8, wherein the final segmentation result is obtained by averaging the segmentation results of the primary decoder and the secondary decoder in the final blood vessel segmentation model of the OCTA image in step S6.
10. An imaging method including the vessel segmentation method of the OCTA image according to any one of claims 1 to 9, comprising the steps of:
(1) obtaining an OCTA image;
(2) segmenting blood vessels in the OCTA image acquired in the step (1) by adopting the blood vessel segmentation method of the OCTA image according to any one of claims 1-9;
(3) according to the segmentation result obtained in the step (2), marking and carrying out secondary imaging on the blood vessel segmentation result obtained in the step (1);
(4) and outputting the OCTA image with the blood vessel mark to finish the final imaging process.
CN202210303004.1A 2022-03-24 2022-03-24 Blood vessel segmentation method and imaging method of OCTA image Pending CN114693928A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210303004.1A CN114693928A (en) 2022-03-24 2022-03-24 Blood vessel segmentation method and imaging method of OCTA image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210303004.1A CN114693928A (en) 2022-03-24 2022-03-24 Blood vessel segmentation method and imaging method of OCTA image

Publications (1)

Publication Number Publication Date
CN114693928A true CN114693928A (en) 2022-07-01

Family

ID=82138551

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210303004.1A Pending CN114693928A (en) 2022-03-24 2022-03-24 Blood vessel segmentation method and imaging method of OCTA image

Country Status (1)

Country Link
CN (1) CN114693928A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115620163A (en) * 2022-10-28 2023-01-17 西南交通大学 Semi-supervised learning deep cut valley intelligent identification method based on remote sensing image

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115620163A (en) * 2022-10-28 2023-01-17 西南交通大学 Semi-supervised learning deep cut valley intelligent identification method based on remote sensing image

Similar Documents

Publication Publication Date Title
CN110390351B (en) Epileptic focus three-dimensional automatic positioning system based on deep learning
CN107766874B (en) Measuring method and measuring system for ultrasonic volume biological parameters
CN113420826B (en) Liver focus image processing system and image processing method
CN113205535B (en) X-ray film spine automatic segmentation and identification method
Wu et al. Registration of longitudinal brain image sequences with implicit template and spatial–temporal heuristics
CN113724876A (en) Intra-stroke hospital complication prediction model based on multi-mode fusion and DFS-LLE algorithm
Liu et al. Multimodal MRI brain tumor image segmentation using sparse subspace clustering algorithm
Saidy et al. Chest X-ray image segmentation using encoder-decoder convolutional network
CN111738997A (en) Method for calculating new coronary pneumonia lesion area ratio based on deep learning
US11600379B2 (en) Systems and methods for generating classifying and quantitative analysis reports of aneurysms from medical image data
CN115512110A (en) Medical image tumor segmentation method related to cross-modal attention mechanism
Cetin et al. Multiple sclerosis lesion detection in multimodal MRI using simple clustering-based segmentation and classification
CN114693928A (en) Blood vessel segmentation method and imaging method of OCTA image
Zhao et al. IFT-net: Interactive fusion transformer network for quantitative analysis of pediatric echocardiography
CN111986216B (en) RSG liver CT image interactive segmentation algorithm based on neural network improvement
Chandra et al. CCsNeT: Automated Corpus Callosum segmentation using fully convolutional network based on U-Net
CN113222979A (en) Multi-map-based automatic skull base foramen ovale segmentation method
CN112200810A (en) Multi-modal automated ventricular segmentation system and method of use thereof
CN104537711B (en) A kind of white matter fiber parameterized model construction method based on healthy population
CN106709921B (en) Color image segmentation method based on space Dirichlet mixed model
Bauer et al. Computer-aided analysis of airway trees in micro-CT scans of ex vivo porcine lung tissue
US8675941B2 (en) Method for visualization of airway anatomical changes
CN115565671A (en) Atrial fibrillation auxiliary analysis method based on cross-model mutual teaching semi-supervision
CN113298754B (en) Method for detecting control points of outline of prostate tissue
Rahmany et al. A priori knowledge integration for the detection of cerebral aneurysm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination