Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Synthetic Generation of Dermatoscopic Images with GAN and Closed-Form Factorization

Rohan Reddy Mekala, Frederik Pahde, Simon Baur, Sneha Chandrashekar, Madeline Diep, Markus Wenzel, Eric L. Wisotzky, Galip Ümit Yolcu, Sebastian Lapuschkin, Jackie Ma, Peter Eisert, Mikael Lindvall, Adam Porter, Wojciech Samek

TL;DR

This work proposes an innovative unsupervised augmentation solution that harnesses Generative Adversarial Network (GAN) based models and associated techniques over their latent space to generate controlled semiautomatically-discovered semantic variations in dermatoscopic images.

Abstract

In the realm of dermatological diagnoses, where the analysis of dermatoscopic and microscopic skin lesion images is pivotal for the accurate and early detection of various medical conditions, the costs associated with creating diverse and high-quality annotated datasets have hampered the accuracy and generalizability of machine learning models. We propose an innovative unsupervised augmentation solution that harnesses Generative Adversarial Network (GAN) based models and associated techniques over their latent space to generate controlled semiautomatically-discovered semantic variations in dermatoscopic images. We created synthetic images to incorporate the semantic variations and augmented the training data with these images. With this approach, we were able to increase the performance of machine learning models and set a new benchmark amongst non-ensemble based models in skin lesion classification on the HAM10000 dataset; and used the observed analytics and generated models for detailed studies on model explainability, affirming the effectiveness of our solution.

Synthetic Generation of Dermatoscopic Images with GAN and Closed-Form Factorization

TL;DR

This work proposes an innovative unsupervised augmentation solution that harnesses Generative Adversarial Network (GAN) based models and associated techniques over their latent space to generate controlled semiautomatically-discovered semantic variations in dermatoscopic images.

Abstract

In the realm of dermatological diagnoses, where the analysis of dermatoscopic and microscopic skin lesion images is pivotal for the accurate and early detection of various medical conditions, the costs associated with creating diverse and high-quality annotated datasets have hampered the accuracy and generalizability of machine learning models. We propose an innovative unsupervised augmentation solution that harnesses Generative Adversarial Network (GAN) based models and associated techniques over their latent space to generate controlled semiautomatically-discovered semantic variations in dermatoscopic images. We created synthetic images to incorporate the semantic variations and augmented the training data with these images. With this approach, we were able to increase the performance of machine learning models and set a new benchmark amongst non-ensemble based models in skin lesion classification on the HAM10000 dataset; and used the observed analytics and generated models for detailed studies on model explainability, affirming the effectiveness of our solution.
Paper Structure (20 sections, 11 figures, 3 tables)

This paper contains 20 sections, 11 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Our Approach
  3. Background
  4. Unsupervised Transformation Pipeline
  5. GAN Training
  6. Latent-Space Factorization
  7. GAN Inversion
  8. Identify Relevant Transformations
  9. Classifier Training Enhancement With Synthetic Data
  10. Experimental Setup
  11. Model, Task, and Training
  12. Results and Discussion
  13. Transformations Developed
  14. Evaluation
  15. Model Analysis with Explainable AI
  16. ...and 5 more sections

Figures (11)

  • Figure 1: Transformation development pipeline
  • Figure 2: Samples of synthetically generated skin lesions.
  • Figure 3: Two pairs of original (left in each case) and HyperStyle inverted images (right in each case).
  • Figure 4: Examples of original images and their generated/synthetic images after the transformations have been applied to the original images.
  • Figure 5: Predictive performance (recall) of the synthetically augmented models compared to the baseline model.
  • ...and 6 more figures