Generative and Autoencoder Models for Large-Scale Mutivariate Unsupervised Anomaly Detection

Advancing medical imaging with GAN-based anomaly detection

“…Technically, GANs are highlighted for their exceptional capability to create synthetic data that is indistinguishable from real data, which is critical for improving anomaly detection accuracy in medical imaging [26], [27]. Moreover, our research diverges from other studies by utilizing a diverse array of datasets rather than depending solely on the MURA dataset [3], [28]. This approach not only broadens the scope of our findings but also ensures a thorough evaluation of GAN-based anomaly detection across various medical imaging fields.…”

“…This approach not only broadens the scope of our findings but also ensures a thorough evaluation of GAN-based anomaly detection across various medical imaging fields. Additionally, unlike studies that focus on a single modality [3], [24], [28], our research incorporates multiple imaging modalities, such as radiographs and CT scans. This multi-modal strategy increases the robustness of our results.…”

“…Presently, the primary impediment to integrating AI-based solutions into healthcare systems is their limited generalization capacity. Anomaly detection [3], [5] grapples with generalization challenges. In the broader context, ensuring "trustworthiness" in AI algorithms necessitates considering a multitude of factors.…”

“…The intersection of artificial intelligence (AI) and medical imaging has led to groundbreaking advancements in disease diagnosis and ealthcare [1], [2]. Among the various AI methodologies, generative adversarial networks (GANs) have emerged as a powerful tool for the detection of anomalies [3]- [5]. GANs [6], which were originally designed for image generation, have demonstrated remarkable potential in capturing subtle irregularities in medical data, often imperceptible to the human eye [7].…”