Convolutional neural network-based ambient light-independent panel digit surveillance technique for infusion pumps

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Patient Safety Technology

Background:

• Medication errors from incorrect infusion pump programming pose significant patient safety risks.
• Continuous monitoring of infusion pump settings is crucial for accurate medication administration.
• Existing surveillance methods may not adequately address user- programming errors.

Purpose of the Study:

• To develop and evaluate a novel surveillance technique for infusion pumps using a convolutional neural network (CNN).
• To continuously monitor variations in infusion pump panel digits to detect potential programming errors.
• To enhance patient safety and reduce medication administration errors.

Main Methods:

• Development of a CNN model for analyzing infusion pump panel digits from frame images.
• Extraction of Region of Interest (ROI) images under varying ambient lighting conditions.
• Experimental evaluation of the CNN model's accuracy and performance in continuous monitoring tests.

Main Results:

• The CNN model achieved high accuracies of 99.9% for training and validation datasets.
• Continuous monitoring demonstrated an overall accuracy of 95.5% for volume recognition.
• The system maintained high accuracy (98.2% day, 90.0% night) under different lighting conditions.

Conclusions:

• The proposed surveillance technique effectively monitors infusion pump operations, significantly improving patient safety.
• This AI-driven approach is expected to reduce medication errors and alleviate burdens on healthcare professionals.
• The technology holds promise for enhancing the safety of long-term infusion therapies.