Automatic detection of body packing in abdominal X-ray images

Objectives: It is difficult and time-consuming to detect narcotic drugs concealed within human bodies by visual assessment of X-ray images . This study aimed to describe a method for automatic detection of concealed drugs based on abdominal X-ray transmission images. Methods: 43 X-ray images from suspects during border inspection are used in this study. Image features were extracted from both the directional fractal dimension and the gray level to form feature vectors. An unbalanced data preprocessing method was implemented before training to minimize the false alarms. A support vector machine model was applied for training and final classification. Results: The proposed method yielded a 97.7% classification accuracy and a 96.1% sensitivity in 43 image cases, which were superior to the method without unbalanced data process resulting in a 95.3% classification accuracy and a 92.3% sensitivity. Conclusions: This fast, computer-based X-ray transmission image processing method can be used in the field inspection for rapid, automatic detection of body packing.