Title: WGAN-GP-Based Synthetic Augmentation of Fahr Disease CT Images for Improved Calcification Detection
Authors: Mourad Henchiri
Volume: 9
Issue: 5
Pages: 17-23
Publication Date: 2025/05/28
Abstract:
Fahr disease is a rare neurodegenerative disorder characterized by bilateral intracranial calcifications predominantly affecting the basal ganglia. Accurate detection of these calcifications on computed tomography (CT) scans is critical for diagnosis but is challenged by the limited availability of annotated imaging data. In this study, we employ Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to generate realistic synthetic CT images of Fahr disease, thereby augmenting the scarce dataset. This augmented dataset is used to train a convolutional neural network (CNN) for automated calcification detection. Experimental results demonstrate that training the CNN solely on real CT images achieves a detection accuracy of approximately 78%, whereas augmenting the training data with WGAN-GP-generated synthetic images improves accuracy to 88%, with corresponding increases in sensitivity and specificity from 75% to 85% and 80% to 90%, respectively. These findings highlight that WGAN-GP-based synthetic image augmentation effectively mitigates data scarcity and substantially enhances the performance of automated detection models for Fahr disease calcifications.