Infectious medical waste characterization using X-ray transmission with Machine learning

In Germany, around 425,400 Mg of specially healthcare-related waste is generated annually, of which 2.4% (10,100 Mg) is infectious (EWC 18 01 03*). Its hazardous nature creates organizational and safety challenges for healthcare facilities and disposal companies, while composition and recycling potential remain largely unknown. However, standardized regulations enable systematic investigation. Aim of this study is to evaluate the potential of X-ray transmission combined with machine learning (ML) for material identification and recycling of infectious waste, while simultaneously assessing the previously unknown composition of this waste stream. To this end, n = 6,179 waste containers with infectious waste from German healthcare sector were systematically analysed at a waste disposal company in North Rhine-Westphalia under field trial conditions. For each container (30-l, 50-l or 60-l), two X-ray images from different angles, its mass, and three external images (including labels identifying the waste-generating facility type) were recorded. Eight content classes were defined and used to train a neural network for automatic content classification. Among all waste containers (58.7 Mg; 0.6% of annual German infectious waste), 50-l containers dominated (60%). Laboratory waste (67%) and medium-weight materials (57%) were the most frequent content classes, while metallic objects were rare and mainly occurred in heterogeneous batches. Containers with laboratory waste showed the highest share of homogeneous batches (89.5%) and the second-highest median mass (9.64 kg) after homogeneous batches (10.14 kg). Heterogeneous batches predominated in hospitals, whereas homogeneous batches dominated in laboratories. Overall, the study shows that X-ray transmission with ML reliably identifies homogeneous (F1 Score = 1), potentially recyclable fractions and provides the first quantitative characterization of infectious waste in Germany.