Related Experiment Video
Updated: Jan 5, 2026
13:19
Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
9.9K
A Deep Learning Approach for Managing Medical Consumable Materials in Intensive Care Units via Convolutional Neural
Arne Peine1,2, Ahmed Hallawa1,3, Oliver Schöffski4
1Department of Intensive Care Medicine and Intermediate Care, University Hospital Rheinisch-Westfälische Technische Hochschule Aachen, Aachen, Germany.
JMIR Medical Informatics
|October 12, 2019
Summary
Consumabot, a contactless visual recognition system, accurately tracks medical consumables in intensive care units (ICUs) using deep learning. This cost-effective solution improves inventory management and reduces costs by eliminating manual tracking needs.
Area of Science:
- Computer Science
- Medical Informatics
- Artificial Intelligence
Background:
- Intensive care units (ICUs) worldwide utilize high volumes of consumable medical materials, significantly contributing to hospital expenditure.
- Current tracking methods like barcodes or RFID require specialized preparation and high infrastructure investment, hindering accurate consumption prediction and increasing costs.
- There is a critical need for cost-effective, contactless object detection methods for tracking medical consumables in ICUs.
Purpose of the Study:
- To develop and evaluate a contactless visual recognition system for tracking medical consumable materials in ICUs.
- To utilize a deep learning approach on a distributed client-server architecture for real-time object detection and classification.
Main Methods:
- Developed Consumabot, a client-server optical recognition system using the MobileNet convolutional neural network model.
- Trained the system on 20 different ICU materials, with 100 sample images per material.
- Evaluated recognition rates in real-world ICU settings, including unobstructed, 50% covered, and multiple item scenarios.
Main Results:
- Consumabot achieved over 99% reliability after approximately 60 training and 150 validation steps.
- Demonstrated a mean top-1 recognition accuracy of 0.85 (SD 0.11) for unobstructed items.
- Showed acceptable accuracy for 50% covered (0.71) and multiple items (0.78), meeting criteria for contactless, resource-efficient tracking.
Conclusions:
- Consumabot, utilizing a CNN architecture, effectively classifies and registers medical consumables directly into electronic health records.
- The system offers a feasible, contactless solution for tracking ICU consumables, improving efficiency and data accuracy.
- Further development is recommended to address limitations with partially covered items and expand assessment across diverse medical settings.
