Evaluation of Blood Sample Collection Tubes Using Deep Learning
Universidad Autónoma de Baja California, Ignacio Franco-Alucano, Julián I. Aguilar-Duque, Universidad Autónoma de Baja California, Yolanda Angelica Baez-Lopez, Universidad Autónoma de Baja California, Jorge Limón-Romero, Universidad Autónoma de Baja California, María Marcela Solís Quinteros, Universidad Autónoma de Baja California, Diego Tlapa, Universidad Autónoma de Baja California · Revista Mexicana de Ingeniería Biomédica · 2025
Phlebotomy is a procedure to obtain blood samples, mainly for laboratory clinical analysis. The amount of blood, tube identification, and the use of the appropriate tube are characteristics that the health professional visually inspects. Being a manual activity, the possibility of error is latent and can affect quality, workflow, and efficiency. Despite the advancement of industry 4.0 technologies, including artificial intelligence (AI), there is little evidence of applications in clinical laboratories. This study aims to evaluate the suitability of using deep learning (DL) in inspecting tubes with blood samples. Specifically, three architectures, YOLOv5, YOLOv7, and YOLOv8, are tested to detect six classes, including cap color and the presence of labels. The highest precision performance was presented by the YOLOv8 model, obtaining a precision of 0.927 in detection, which shows a high capacity to inspect important characteristics in the phlebotomy service. Therefore, being DL is a suitable alternative to assist health professionals in inspection activities. Future work includes expanding the number of images in a balanced manner.