Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Fundus blood vessel image segmentation adversarial sample generation method and segmentation network security evaluation method

A blood vessel image and anti-sample technology, applied in the field of medical image processing, can solve suboptimal problems and achieve the effect of verifying security

Inactive Publication Date: 2019-11-26
NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
View PDF3 Cites 8 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Goodfellow et al. proposed a fast gradient sign method (Fast Gradient Sign Method, FGSM) in 2015 to generate adversarial samples. This method calculates the gradient of the loss function of each element, and then moves a small step based on the gradient descent direction, although This approach is fast, but using only a single direction based on a linear approximation of the loss function often leads to suboptimal results

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Fundus blood vessel image segmentation adversarial sample generation method and segmentation network security evaluation method
  • Fundus blood vessel image segmentation adversarial sample generation method and segmentation network security evaluation method
  • Fundus blood vessel image segmentation adversarial sample generation method and segmentation network security evaluation method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0044] The invention discloses a method for generating an adversarial sample for fundus blood vessel image segmentation, which includes the following steps:

[0045] Step 1, establish fundus blood vessel image segmentation network f;

[0046] In this embodiment, the fundus blood vessel image segmentation network is a U-Net network, such as figure 2 As shown, it includes four lower convolutional layers D, four upper convolutional layers U and one sigmoid operation layer; one lower convolutional layer contains two convolutional layers and one pooling layer; one upper convolutional layer contains two Convolutional layers, a deconvolutional layer, and a concatenation operation.

[0047] In the present invention, on the original U-Net network structure, the convolution operation is set with kernelsize as 3 and padding as 1, which ensures that the size of the image remains unchanged after the convolution operation. The pooling operation selects the maximum pooling method, and the...

Embodiment 2

[0074] In this embodiment, an adversarial example is generated through a general perturbation. A general perturbation is a fixed perturbation, unlike one per image, a general perturbation can be applied to all original images and fool their pre-trained segmentation models.

[0075] Let X represent the fundus blood vessel image in the pixel space R d distribution within , f is the segmentation model with high accuracy on X. The goal of the general perturbation is to find a fixed pattern u∈R d , so that for satisfy:

[0076] f(x+u)≠f(x), s.t.||u|| p ≤σ(5)

[0077] The parameter σ is the preset threshold, which means that the general disturbance u performs L p Upper bound for norm constraints. In order for the disturbance to be imperceptible to the human eye, σ needs to be small enough. Such as Figure 5 As shown, it is a schematic diagram of the model framework for generating adversarial samples for fundus blood vessel image segmentation in this embodiment.

[0078] B...

Embodiment 3

[0085] In this embodiment, the Dice coefficient before and after the attack is used to evaluate the security of the fundus blood vessel image segmentation network, as shown in Table 1 for non-target attacks.

[0086] Table 1 Dice coefficient after non-target attack on U-Net network (initial value is 68.5%)

[0087]

[0088] It can be seen from Table 1 that compared with the output of the previous segmentation model and the Dice coefficient of manually marked pictures of 68.5%, after the attack, the Dice coefficient drops significantly, indicating that the segmentation network has made a significant difference to the input image with disturbances invisible to the human eye. The very different segmentation results indicate that the U-Net network used for segmentation is very vulnerable to counter-perturbations and is extremely vulnerable to attacks. After being attacked, the Dice coefficient drops sharply. For example in L ∞ When = 20, the Dice coefficient drops from 68.5% t...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a fundus blood vessel image segmentation adversarial sample generation method and a segmentation network security evaluation method. The fundus blood vessel image segmentationadversarial sample generation method comprises the steps: 1, establishing a fundus blood vessel image segmentation network; 2, collecting an original fundus blood vessel image, marking blood vessels in the collected image, and constructing a training sample to train a fundus blood vessel image segmentation network; 3, constructing a generative disturbance network to generate an adversarial sampleimage; 4, inputting the generated adversarial sample image into a trained fundus blood vessel image segmentation network to obtain a segmentation result, calculating an objective function, and updating parameters of a generated disturbance network by minimizing the objective function to obtain an optimized generated disturbance network; and 5, generating disturbance by using the optimized disturbance generation network, and adding the disturbance to the original fundus blood vessel image to obtain an adversarial sample image. According to the fundus blood vessel image segmentation adversarialsample generation method, the adversarial sample image which is indistinguishable with the human eyes of the original image can be obtained, and the obtained generative disturbance network can betterlearn the characteristics of the segmentation network.

Description

technical field [0001] The invention belongs to the technical field of medical image processing, and in particular relates to a method for generating an adversarial sample for fundus blood vessel image segmentation and a segmentation network security evaluation method. Background technique [0002] In recent years, deep learning algorithms driven by new network structures and big data advancements have shown amazing high performance in many artificial intelligence systems, such as image recognition and semantic segmentation. The application of deep learning in clinical medicine is also very exciting. In medical diagnosis, deep learning algorithms seem to have reached the same level as doctors in radiology, pathology, dermatology and ophthalmology. In 2018, the U.S. Food and Drug Administration (FDA) approved the first autonomous AI medical diagnosis system and said they are actively developing a new regulatory framework to foster innovation in this area. [0003] However, S...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/10
CPCG06T2207/20081G06T2207/20084G06T2207/30041G06T7/10
Inventor 张道强徐梦婷张涛李仲年
Owner NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com