Computer aided detection of surgical retained foreign object for prevention

Area of Science:

• Medical Imaging
• Surgical Safety
• Artificial Intelligence in Medicine

Background:

• Surgical retained foreign objects (RFOs) lead to significant patient morbidity and mortality, costing billions annually in preventable medical expenses.
• Radiographic detection of RFOs is suboptimal, with current accuracy rates around 59%.
• There is a critical need for improved technologies to prevent and detect RFOs during surgical procedures.

Purpose of the Study:

• To develop and evaluate a novel system combining a 3D gossypiboma micro tag (μTag) with a computer-aided detection (CAD) system for enhanced RFO detection.
• To enable the CAD system to operate in high specificity mode for real-time intraoperative use by surgeons and in high sensitivity mode for radiologist review.

Main Methods:

• A dataset of 1800 cadaver images was created, featuring 3D μTags in random orientations alongside other surgical objects.
• A CAD system was developed, incorporating modules for μTag enhancement, segmentation, feature analysis, classification, and detection.
• The CAD system was trained and validated on subsets of the cadaver image data.

Main Results:

• In high specificity mode (for surgeons), the CAD system achieved 79.5% sensitivity with 0.003 false positives per image on the test set.
• In high sensitivity mode (for radiologists), the CAD system achieved 90.2% sensitivity with 0.23 false positives per image on the test set.
• Performance on the training set showed 81.5% sensitivity (high specificity) and 96.1% sensitivity (high sensitivity).

Conclusions:

• This study presents the first use of a 3D μTag for consistent 2D radiographic projection regardless of orientation.
• It is the first CAD system designed to detect man-made objects against complex anatomical backgrounds.
• The developed CAD system demonstrates effective performance in both high specificity and high sensitivity modes for RFO detection.