The Importance of Objective Methods for the Evaluation of Deep Learning Image Processing Devices

Intro

This is an ongoing project investigating the efficacy of generative adversarial networks in removing noise from medical scans, and the degree to which these networks' methods of constructing realistic high-quality xrays from noisy input undermines necessary objectivity in processing raw medical image data.

Overview

References

1. Denoising Autoencoders
2. Generative Adversarial Networks
3. Network Optimization
4. Simulating Image Noise

---

Denoising Autoencoders

Tutorials

• Tensorflow Intro to Autoencoders
• Keras Denoising Autoencoder

Research

• Medical Image Denoising Using Convolutional Denoising Autoencoders [paper] [code]
• Learning Deep Representations Using Convolutional Auto-Encoders with Symmetric Skip Connections [paper] [code]
• Hyperspectral X-Ray Denoising: Model-Based and Data-Driven Solutions [paper]

---

Generative Adversarial Networks (GAN)

Tutorials

• Tensorflow DCGAN
• Tensorflow ESRGAN

Research

• Low Dose CT Image Denoising Using a Generative Adversarial Network with Wasserstein Distance and Perceptual Loss [paper] [code]
• Digital Radiography Image Denoising Using a Generative Adversarial Network [paper]

---

Network Optimization

1. VGG Loss [paper] [code] [docs]

---

Simulating X-ray Image Noise

Research

• A Technique for Simulating the Effect of Dose Reduction on Image Quality in Digital Chest Radiography [paper]