GAN-based one dimensional medical data augmentation

Abstract

With the continuous development of human life and society, the medical field is constantly improving. However, modern medicine still faces many limitations, including challenging and previously unsolvable problems. In these cases, artificial intelligence (AI) can provide solutions. The research and application of generative adversarial networks (GAN) are a clear example. While most researchers focus on image augmentation, there are few one-dimensional data augmentation examples. The radiomics feature extracted from RT and CT images is one-dimensional data. As far as we know, we are the first to apply the WGAN-GP algorithm to generate radiomics data in the medical field. In this paper, we input a portion of the original real data samples into the model. The model learns the distribution of the input data samples and generates synthetic data samples with similar distribution to the original real data, which can solve the problem of obtaining annotated medical data samples. We have conducted experiments on the public dataset Heart Disease Cleveland and the private dataset. Compared with the traditional method of Synthetic Minority Oversampling Technique (SMOTE) and common GAN for data augmentation, our method has significantly improved the AUC and SEN values under different data proportions. At the same time, our method has also shown varying levels of improvement in ACC and SPE values. This demonstrates that our method is effective and feasible.

Multichannel One-Dimensional Data Augmentation with Generative Adversarial Network

“…Ref. [ 21 ] applied WGAN-GP for one-dimensional medical data augmentation to solve the problem of obtaining annotated medical data samples. Ref.…”

“…Similar to [ 8 ], they converted multichannel one-dimensional EEG data to images, which increases computational overhead and possibly leads to information loss. Moreover, the proposed methods in [ 8 , 20 , 21 ] cannot be applied directly to the multi-channels’ one-dimensional data augmentation task, which would otherwise help the classifier to extract more information across channels for each instance.…”

“…Most of these works, however, centered on computer vision-related tasks. A few works on one-dimensional data include that of [ 8 , 20 , 21 ]. A comprehensive survey on time series data augmentation was also conducted by [ 22 ].…”

Two-Stage Input-Space Image Augmentation and Interpretable Technique for Accurate and Explainable Skin Cancer Diagnosis

“…The concept of GAN-based augmentation refers to the use of generative adversarial networks (GANs) for the purpose of producing synthetic data samples that can be used to augment an existing dataset [10]. This technique is particularly useful in cases where the original dataset is small or imbalanced, as it can help to increase the size of the dataset and balance the class distribution.…”