SinGAN-Seg: Synthetic training data generation for medical image segmentation

@article{Thambawita2022SinGANSegST,
  title={SinGAN-Seg: Synthetic training data generation for medical image segmentation},
  author={Vajira Lasantha Thambawita and Pegah Salehi and Sajad Amouei Sheshkal and S. Hicks and Hugo L.Hammer and Sravanthi Parasa and Thomas de Lange and Paal Halvorsen and M. Riegler},
  journal={PLoS ONE},
  year={2022},
  volume={17}
}
Analyzing medical data to find abnormalities is a time-consuming and costly task, particularly for rare abnormalities, requiring tremendous efforts from medical experts. Therefore, artificial intelligence has become a popular tool for the automatic processing of medical data, acting as a supportive tool for doctors. However, the machine learning models used to build these tools are highly dependent on the data used to train them. Large amounts of data can be difficult to obtain in medicine due… 
PolypConnect: Image inpainting for generating realistic gastrointestinal tract images with polyps
TLDR
The PolypConnect pipeline, which can convert non-polyp images into polyp images to increase the size of training datasets for training, is proposed, which shows a 5 .
Synthetic data for unsupervised polyp segmentation
TLDR
This work produces realistic synthetic images using a combina-tion of 3D technologies and generative adversarial networks and uses zero annotations from medical professionals in the pipeline, achieving promising results on real polyp segmentation datasets.

References

SHOWING 1-10 OF 73 REFERENCES
SinGAN: Learning a Generative Model From a Single Natural Image
We introduce SinGAN, an unconditional generative model that can be learned from a single natural image. Our model is trained to capture the internal distribution of patches within the image, and is
HyperKvasir, a comprehensive multi-class image and video dataset for gastrointestinal endoscopy
TLDR
The HyperKvasir dataset is presented, the largest image and video dataset of the gastrointestinal tract available today and can play a valuable role in developing better algorithms and computer-assisted examination systems not only for gastro- and colonoscopy, but also for other fields in medicine.
On Aliased Resizing and Surprising Subtleties in GAN Evaluation
TLDR
This paper identifies and characterize variations in generative modeling development pipelines, provides recommendations based on signal processing principles, and releases a reference implementation to facilitate future comparisons.
DeepFake electrocardiograms: the key for open science for artificial intelligence in medicine
TLDR
Generative adversarial networks capable of generating realistic synthetic DeepFake 12-lead 10-sec electrocardiograms (ECGs) using adversarial neural networks on normal ECGs from two population studies are presented.
Differential privacy in health research: A scoping review
TLDR
More development, case studies, and evaluations are needed before differential privacy can see widespread use in health research, and diminished accuracy in small datasets is problematic.
DeepSynthBody: the beginning of the end for data deficiency in medicine
TLDR
DeepSynthBody is presented, a novel framework that overcomes some of the inherent restrictions and limitations of medical data by using deep generative adversarial networks to produce synthetic data with characteristics similar to the real data, so-called DeepSynth (deep synthetic) data.
DivergentNets:Medical Image Segmentation by Network Ensemble
TLDR
This work proposes a segmentation model named TriUNet composed of three separate UNet models, and combines an ensemble of well-known segmentation models into a model called DivergentNets to produce more generalizable medical image segmentation masks.
...
...