Advancing Medical Imaging: A Comprehensive Synthetic Dataset for Infant Brain MRI Analysis

Accurate diagnosis and treatment planning are consistently challenging tasks in the medical field, especially when reliant on image diagnosis. The availability of sufficient and high-quality datasets becomes crucial in addressing this challenge. To tackle this issue, synthetic datasets can play a vital role in augmenting the number of samples while maintaining good quality. This study focuses on utilizing Deep Convolutional GANs (DCGANs) and Wasserstein GANs (WGANs) to generate synthetic datasets, specifically for Infant Brain MRI imaging. The Human Connectome Project (dHCP) dataset, comprising 800 samples categorized into four classes, was used for this study. Several key metrics, including the Similarity Index (SSI), Inception Score (IS), and Fréchet Inception Distance (FID), were employed to evaluate the performance of the generated synthetic datasets. The Inception Score was calculated for each class individually as well as for the overall synthetic dataset. For the normal class (first class), DCGAN achieved an Inception Score of lS.45, while WGAN scored 15.13. In the second class, DCGAN scored 19.70 compared to WGAN's 16.21. Similarly, in the third class, DCGAN scored lS.65 and WGAN 15.65, and in the abnormal class (fourth class), DCGAN scored 17.58 against WGAN's 16.00. The overall synthetic dataset scored 19.29 for DCGAN and 16.47 for WGAN. As the higher Inception Scores across different classes and for the overall dataset denoted that DCGAN performed well compare to WGAN in generating synthetic images for the Infant Brain MRI dataset.