Detection of Body Packs in Abdominal CT scans Through Artificial Intelligence; Developing a Machine Learning-based...

Area of Science:

• Forensic Medicine
• Medical Imaging
• Artificial Intelligence

Background:

• Identifying individuals smuggling illegal substances internally is crucial for forensic medicine and toxicology.
• Current methods for detecting internal body packing can be challenging and require advanced imaging techniques.

Purpose of the Study:

• To develop and evaluate a novel artificial intelligence (AI) based diagnostic method for real-time detection of body packs in abdominal CT scans.
• To improve the accuracy and efficiency of identifying individuals smuggling contraband via internal body packing.

Main Methods:

• A cross-sectional study utilizing 888 abdominal CT scans.
• Development of a machine learning model using RetinaNet with a modified neck for object detection.
• Incorporation of oriented bounding boxes (angled Bbox) in the training dataset to enhance detection accuracy.

Main Results:

• The proposed Body Packs Detection (BPD) model achieved a mean average precision (mAP) of 86.6% at an IoU of 0.5.
• The model demonstrated a mAP of 45.6% across various IoU thresholds (0.5-0.95).
• Achieved a Recall value of 58.5%, outperforming standard RetinaNet in detecting body packs.

Conclusions:

• Deep learning networks can effectively identify body packs in abdominal CT scans.
• Integrating object shape and variability is key for AI in healthcare diagnostics.
• Specialized datasets curated by experts are essential for training effective AI models for specific detection tasks like body packing.