Leveraging 2D object detection models for accurate biomatter threat identification in 3D CT scans of baggage

The detection of biomatter threats in baggage is a critical task for ensuring biosecurity, especially at international borders. This study introduces a novel methodology for converting 3D computed tomography (CT) volumetric data into 2D representations to enable efficient object detection using state-of-the-art 2D models. Systematic time-series feature engineering techniques were applied to vertical variation sequences from 3D data, transforming complex volumetric structures into compact and informative 2D projections. The proposed method leverages Light Gradient Boosting Machine and Random Forest (RF) estimators, combined with dimensionality reduction techniques such as Principal Component Analysis (PCA) and Partial Least Squares Regression, to select the most discriminative features. Evaluations using the YOLOv10l object detection framework demonstrated high detection accuracy, achieving a mean average precision of 0.851 with RF and PCA. Despite challenges such as class imbalance and computational trade-offs, the methodology offers a scalable, efficient, and highly accurate approach to biomatter detection. This study not only addresses limitations in current 3D image analysis techniques but also highlights potential applications in medical imaging and industrial inspection.