Preventing Syringe Swap Errors with an Attention-Based CRNN: A Real-Time Mobile AI Solution

Patrick Reichherzer, Vanessa Restrepo Schild, Ollie Dare, Steven Webster-Edge

Abstract: Medication errors, particularly syringe swaps, pose a serious patient safety risk in operating theatres. We present a mobile app that replaces the manual selection of pre-printed syringe labels with an automated, image-based printing system. The real-time workflow captures an ampule image and prints a verified label using a two-stage approach: initial optical character recognition (OCR), followed by a lightweight Convolutional Recurrent Neural Network (CRNN) with attention for refined text extraction. This design minimizes computational load for offline use on mobile devices while maintaining accuracy under poor imaging conditions. We detail the system’s implementation and CRNN architecture, highlighting its potential to reduce syringe swap errors and improve patient safety.

External IDs:dblp:conf/aime/ReichherzerSDW25