Baggage screening in airports is a cornerstone of airport security measures. The advent of computer vision technologies in recent years has led to the development of several automated systems for identifying security threats in baggage scans. However, existing methods struggle with cluttered and occluded baggage items, and the scarcity of high-quality labeled data hampers their performance. Hence, in this paper, we propose a novel semi-supervised anomaly detection framework that leverages topological features for self-supervised learning. By utilizing persistence diagrams and persistence images, our approach captures the structural properties of the data, enhancing threat detection capabilities. The proposed framework employs a clustering-based self-supervised pretext task to generate pseudo labels, effectively utilizing both labeled and unlabeled data. Our framework was rigorously tested on the SIXray public dataset, achieving 86.23% accuracy and 86.05% F1-score, outperforming state-of-the-art semi-supervised methods.