GAN-Based Data Augmentation Approach for Improving Deep Learning-Based Healthcare Diagnosis

Abstract

Although the rapid development of deep learning methods has greatly enhanced healthcare diagnosis systems, large, diverse and high-quality medical datasets are often limited. However, there are still some considerable difficulties in creating reliable diagnostic models: data scarcity, class imbalance, and privacy concerns. To improve deep learning (DL) healthcare diagnosis, this study introduces a Generative Adversarial Network (GAN) based data augmentation approach. The proposed approach is based on using GANs to generate realistic synthetic medical data that can be used to augment existing datasets without compromising the clinical characteristics. This can help to enhance the correctness, resilience, and capability to generalize of deep learning models by expanding the diversity of datasets and balancing underrepresented classes. The augmented data is combined with traditional training data and assessed on standard metrics of diagnostic performance including accuracy, precision, recall and F1-score. The experimental results show that the GAN-generated samples can be used to effectively enhance the performance of the developed models, especially in the case of low medical data. The overall proposed framework offers a potential solution for creating more dependable and scalable AI-powered healthcare diagnosis systems, which would help address the issues of data scarcity and patient privacy.