SSDM:A Self Supervised Diffusion Model For Lung Anomaly Detection Using Chest X-rays

Self supervised learning is emerging very fastly in computer vision tasks, which address the scarcity of annotated medical images. We introduce a self-supervised approach for anomaly detection using the Diffusion Probabilistic Model, where the model is trained exclusively on normal chest X-rays and serves as a baseline for identifying anomalies. We implemented a U-net-based Gaussian diffusion model(SSDM) that adds noise in an iterative manner to images and learns to generate them in reverse. Early detection of disease is critical for medical diagnosis. The empirical results show that the model we put forward has great accuracy in detecting the lung anomaly and thereby helps to diagnose disease at an early stage.