Generalization of Deep Neural Networks for Chest Pathology Classification in X-Rays Using Generative Adversarial Networks

Salehinejad, Hojjat; Valaee, Shahrokh; Dowdell, Tim; Colak, Errol; Barfett, Joseph

doi:10.1109/icassp.2018.8461430

Cited by 167 publications

(136 citation statements)

References 15 publications

Supporting

Mentioning

133

Contrasting

Unclassified

Order By: Relevance

“…The sigmoid activation function (Equation )

f (x) = \frac{1}{1 + {normale}^{- x}}

is an S‐shaped curve having values between 0 and 1. The sigmoid function is favorable as it is easy to understand but the disadvantages are its slow convergence and saturation of gradients (Salehinejad, Sankar, Barfett, Colak, & Valaee, ). The Hyperbolic Tangent function‐ Tanh (Equation ) has an output

f true(x true) = \frac{1 - {normale}^{- 2 x}}{1 + {normale}^{- 2 x}}

that ranges between −1 and 1.…”

Section: Methodsmentioning

confidence: 99%

Artificial neural network‐based modeling of snow properties using field data and hyperspectral imagery

Haq

Ghosh

Rahaman

et al. 2019

Natural Resource Modeling

View full text Add to dashboard Cite

This study attempts to model snow wetness and snow density of Himalayan snow cover using a combination of Hyperspectral image processing and Artificial Neural Network (ANN). Initially, a total of 300 spectral signature measurements, synchronized with snow wetness and snow density, were collected in the field. The spectral reflectance of snow was then modeled as a function of snow properties using ANN. Four snow wetness and three snow density models were developed. A strong correlation was observed in near‐infrared and shortwave‐infrared region. The correlation analysis of ANN modeled snow density and snow wetness showed a strong linear relationship with field‐based data values ranging from 0.87–0.90 and 0.88–0.91, respectively. Our results indicate that an Artificial Intelligence (AI) approach, using a combination of Hyperspectral image processing and ANN, can be efficiently used to predict snow properties (wetness and density) in the Himalayan region. Recommendations for resource managers Snow properties, such as snow wetness and snow density are mainly investigated through field‐based survey but rugged terrains, difficult weather conditions, and logistics management issues establish remote sensing as an efficient alternative to monitor snow properties, especially in the mountain environment. Although Hyperspectral remote sensing is a powerful tool to conduct the quantitative analysis of the physical properties of snow, only a few studies have used hyperspectral data for the estimation of snow density and wetness in the Himalayan region. This could be because of the lack of synchronized snow properties data with field‐based spectral acquisitions. In combination with Hyperspectral image processing, Artificial Neural Network (ANN) can be a useful tool for effective snow modeling because of its ability to capture and represent complex input‐output relationships. Further research into understanding the applicability of neural networks to determine snow properties is required to obtain results from large snow cover areas of the Himalayan region.

show abstract

“…The sigmoid activation function (Equation )

f (x) = \frac{1}{1 + {normale}^{- x}}

f true(x true) = \frac{1 - {normale}^{- 2 x}}{1 + {normale}^{- 2 x}}

that ranges between −1 and 1.…”

Section: Methodsmentioning

confidence: 99%

Artificial neural network‐based modeling of snow properties using field data and hyperspectral imagery

Haq

Ghosh

Rahaman

et al. 2019

Natural Resource Modeling

View full text Add to dashboard Cite

show abstract

“…GANs and their variants have been applied in different architectures in continuing efforts to improve the accuracy and effectiveness of image classification. The GAN framework has been utilized in numerous works as a more generic approach to generating realistic training images that synthetically augment datasets in order to combat overfitting; e.g., for synthetic data augmentation in liver lesions [7], retinal fundi [9], histopathology [12], and chest X-rays [28]. Calimeri [24] to generate images of three liver lesion classes to synthetically augment the limited dataset and improve the performance of CNN for liver lesion classification.…”

Section: Related Workmentioning

confidence: 99%

Multi-adversarial Variational Autoencoder Networks

Imran

Terzopoulos

2019

2019 18th IEEE International Conference on Machine Learning and Applications (ICMLA)

View full text Add to dashboard Cite

The unsupervised training of GANs and VAEs has enabled them to generate realistic images mimicking real-world distributions and perform image-based unsupervised clustering or semi-supervised classification. Combining the power of these two generative models, we introduce Multi-Adversarial Variational autoEncoder Networks (MAVENs), a novel network architecture that incorporates an ensemble of discriminators in a VAE-GAN network, with simultaneous adversarial learning and variational inference. We apply MAVENs to the generation of synthetic images and propose a new distribution measure to quantify the quality of the generated images. Our experimental results using datasets from the computer vision and medical imaging domains-Street View House Numbers, CIFAR-10, and Chest X-Ray datasets-demonstrate competitive performance against state-of-the-art semi-supervised models both in image generation and classification tasks.

show abstract

“…GANs are also being used to help segmentation networks to work well across different modalities, e.g CycleGAN proposed by Zhu et al [25] to segment brain images in both CT and MRI [22]. Salehinejad et al [18] use DCGAN to synthesize X-ray chest scans with under-represented diseases and to classify lung diseases. The motivation of the work from Salehinejad et al is most similar to ours but instead of generating images from scratch, ScarGAN simulates diseases on scans of healthy patients.…”

Section: Related Workmentioning

confidence: 99%

ScarGAN: Chained Generative Adversarial Networks to Simulate Pathological Tissue on Cardiovascular MR Scans

Lau¹,

Hendriks

Lieman-Sifry³

et al. 2018

Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support

View full text Add to dashboard Cite

Medical images with specific pathologies are scarce, but a large amount of data is usually required for a deep convolutional neural network (DCNN) to achieve good accuracy. We consider the problem of segmenting the left ventricular (LV) myocardium on late gadolinium enhancement (LGE) cardiovascular magnetic resonance (CMR) scans of which only some of the scans have scar tissue. We propose ScarGAN to simulate scar tissue on healthy myocardium using chained generative adversarial networks (GAN). Our novel approach factorizes the simulation process into 3 steps: 1) a mask generator to simulate the shape of the scar tissue; 2) a domain-specific heuristic to produce the initial simulated scar tissue from the simulated shape; 3) a refining generator to add details to the simulated scar tissue. Unlike other approaches that generate samples from scratch, we simulate scar tissue on normal scans resulting in highly realistic samples. We show that experienced radiologists are unable to distinguish between real and simulated scar tissue. Training a U-Net with additional scans with scar tissue simulated by ScarGAN increases the percentage of scar pixels correctly included in LV myocardium prediction from 75.9% to 80.5%.

show abstract

Generalization of Deep Neural Networks for Chest Pathology Classification in X-Rays Using Generative Adversarial Networks

Cited by 167 publications

References 15 publications

Artificial neural network‐based modeling of snow properties using field data and hyperspectral imagery

Artificial neural network‐based modeling of snow properties using field data and hyperspectral imagery

Multi-adversarial Variational Autoencoder Networks

ScarGAN: Chained Generative Adversarial Networks to Simulate Pathological Tissue on Cardiovascular MR Scans

Contact Info

Product

Resources

About