Lens-Free Shadow Imaging-Based Cell Classification with Deep Learning for Improved CDC Crossmatching

초록

The complement-dependent cytotoxicity (CDC) assay is the criterion (gold) standard for detecting anti-human leukocyte antigen antibodies. Conventional cell classification techniques like microscopy, flow cytometry, and ELISA are constrained by limited accuracy, slow processing times, and potential subjectivity. This study explores the potential of lens-free shadow imaging technology (LSIT) integrated with deep learning for enhanced cell classification in CDC crossmatching. LSIT's label-free, high-throughput, and cost-effective nature makes it suitable for large-scale cell classification tasks. We used YOLO (you only look once) to quickly identify and categorize objects and SVM (support vector machine) to classify cells based on shadow parameters obtained by LSIT. This method offers an exciting new way to process sensor data and utilize artificial intelligence. The YOLO-based convolutional neural network model achieved an accuracy of 0.9712, an IOU of 0.9074, and an overall processing time of under 10 seconds, significantly outperforming traditional methods in accuracy and efficiency. Integrating LSIT, YOLO, and SVM presents a novel and promising approach for cell classification in CDC crossmatching, overcoming the limitations of conventional methods to offer improved accuracy, efficiency, and potential for automation. This study can potentially revolutionize cell classification in CDC crossmatching, leading to more accurate and efficient diagnoses.