Light In The Black: An Evaluation of Data Augmentation Techniques for COVID-19 CT's Semantic Segmentation

TL;DR

This paper systematically evaluates twenty data augmentation techniques to enhance the training of neural networks for COVID-19 CT image segmentation, highlighting the effectiveness of spatial transformations.

Contribution

It provides an extensive analysis of data augmentation methods for COVID-19 CT segmentation, with over 3,000 experiments across five datasets, identifying the most promising techniques.

Findings

Spatial transformations significantly improve segmentation performance

Data augmentation enhances neural network training robustness

Over 3,000 experiments validate the effectiveness of specific techniques

Abstract

With the COVID-19 global pandemic, computer-assisted diagnoses of medical images have gained much attention, and robust methods of Semantic Segmentation of Computed Tomography (CT) became highly desirable. Semantic Segmentation of CT is one of many research fields of automatic detection of COVID-19 and has been widely explored since the COVID-19 outbreak. In this work, we propose an extensive analysis of how different data augmentation techniques improve the training of encoder-decoder neural networks on this problem. Twenty different data augmentation techniques were evaluated on five different datasets. Each dataset was validated through a five-fold cross-validation strategy, thus resulting in over 3,000 experiments. Our findings show that spatial level transformations are the most promising to improve the learning of neural networks on this problem.