Enhancing Diagnostic Accuracy Through Generative AI And Synthetic Data Generation For Robust Medical Imaging

Abstract

Medical imaging diagnosis is critically constrained by the scarcity of annotated datasets, class imbalance, and limited generalization of deep learning models. This paper presents a Generative Adversarial Network (GAN)-based framework that synthesizes high-fidelity medical images to augment training corpora for Convolutional Neural Network (CNN) classifiers. The proposed architecture combines a conditional GAN generator G(z|c) with a discriminator D(x|c), whose adversarial minimax objective drives convergence toward realistic synthetic distributions. Augmented datasets reduce overfitting and improve sensitivity, specificity, and F1-score across disease categories. Experimental validation on benchmark datasets demonstrates classification accuracy improvements of 4.3–7.8 percentage points over baseline non-augmented models. Comparison with state-of-the-art augmentation strategies confirms the superiority of GAN-based synthesis. Results establish the proposed framework as a scalable, data-efficient solution for clinical decision-support systems.