Beyond the human eye: Artificial intelligence revolutionizing plasma quality control

Background and Objectives Plasma quality control (QC) in transfusion medicine relies on subjective visual inspections, which vary depending on the technologist's experience and lighting conditions, potentially compromising safety. We developed an automated system to standardize the detection of plasma colour and turbidity abnormalities, and aimed to evaluate its accuracy and reliability in a transfusion setting. Materials and Methods At the Blood Transfusion Center, Khon Kaen University, Thailand, three experienced technologists classified 973 plasma bags. The system was trained on 789 bags (467 normal, 322 abnormal) and tested on 184 bags (145 normal, 39 abnormal) in Phase 1 and 486 bags (287 normal, 199 abnormal) in Phase 2. A sorting machine captured images under controlled lighting, and a deep learning model assessed the quality. Performance was measured using accuracy, sensitivity, specificity and precision, with 95% confidence intervals (CIs). Results The system achieved 87.5% accuracy (95% CI: 82.7%–91.4%) in Phase 1 and 94.7% accuracy (95% CI: 92.3%–96.3%) in Phase 2, with 100% sensitivity (95% CI: 98.1%–100%) and 90.9% specificity (95% CI: 87.1%–93.7%) in Phase 2, ensuring no abnormal units were missed. Cross‐validation on the training set yielded 94.8% accuracy. Conclusion This automated QC system offers a reliable alternative to manual inspection, minimizing errors and reducing wastage (26 units vs. 29–86 with manual methods). Its high sensitivity and integration with a sorting machine support its potential to standardize plasma QC, enhancing patient safety and blood bank efficiency.