Structure Unbiased Adversarial Model for Medical Image Segmentation

by   Tianyang Zhang, et al.

Generative models have been widely proposed in image recognition to generate more images where the distribution is similar to that of the real images. It often introduces a discriminator network to discriminate original real data and generated data. However, such discriminator often considers the distribution of the data and did not pay enough attention to the intrinsic gap due to structure. In this paper, we reformulate a new image to image translation problem to reduce structural gap, in addition to the typical intensity distribution gap. We further propose a simple yet important Structure Unbiased Adversarial Model for Medical Image Segmentation (SUAM) with learnable inverse structural deformation for medical image segmentation. It consists of a structure extractor, an attention diffeomorphic registration and a structure & intensity distribution rendering module. The structure extractor aims to extract the dominant structure of the input image. The attention diffeomorphic registration is proposed to reduce the structure gap with an inverse deformation field to warp the prediction masks back to their original form. The structure rendering module is to render the deformed structure to an image with targeted intensity distribution. We apply the proposed SUAM on both optical coherence tomography (OCT), magnetic resonance imaging (MRI) and computerized tomography (CT) data. Experimental results show that the proposed method has the capability to transfer both intensity and structure distributions.


page 1

page 2

page 3

page 4

page 6

page 7

page 8


SkrGAN: Sketching-rendering Unconditional Generative Adversarial Networks for Medical Image Synthesis

Generative Adversarial Networks (GANs) have the capability of synthesizi...

Large-Kernel Attention for 3D Medical Image Segmentation

Automatic segmentation of multiple organs and tumors from 3D medical ima...

Joint segmentation and discontinuity-preserving deformable registration: Application to cardiac cine-MR images

Medical image registration is a challenging task involving the estimatio...

Privacy Preserving for Medical Image Analysis via Non-Linear Deformation Proxy

We propose a client-server system which allows for the analysis of multi...

Two-stage MR Image Segmentation Method for Brain Tumors based on Attention Mechanism

Multimodal magnetic resonance imaging (MRI) can reveal different pattern...

ToothSegNet: Image Degradation meets Tooth Segmentation in CBCT Images

In computer-assisted orthodontics, three-dimensional tooth models are re...

Dimensionality Reduction for Improving Out-of-Distribution Detection in Medical Image Segmentation

Clinically deployed segmentation models are known to fail on data outsid...

Please sign up or login with your details

Forgot password? Click here to reset