iWaste: Video-Based Medical Waste Detection and Classification
Junbo Chen, Jeffrey Mao, Cassandra L. Thiel, Yao Wang
Abstract
Waste auditing is important for effectively reducing the medical waste generated by resource-intensive operating rooms. To replace the current time-intensive and dangerous manual waste auditing method, we propose a system named iWASTE to detect and classify medical waste based on videos recorded by a camera-equipped waste container. In this pilot study, we collected a video dataset of 4 waste items (gloves, hairnet, mask, and shoecover) and designed a motion detection based preprocessing method to extract and trim useful frames. We propose a novel architecture named R3D+C2D to classify waste videos by combining features learnt by 2D convolutional and 3D convolutional neural networks. The proposed method obtained a promising result (79.99% accuracy) on our challenging dataset.Clinical Relevance-iWaste enables consistent and effective real-time monitoring of solid waste generation in operating rooms, which can be used to enforce medical waste sorting policies and to identify waste reduction strategies.