Abstract: Objective The hypoxic environment of plateau regions significantly affects the quality of stored red blood cells (RBCs), and conventional assessment methods are inadequate for rapid, accurate, and dynamic quality monitoring. This study developed a deep learning-based image recognition model for evaluating red blood cells (RBCs) storage leisions, systematically compared the morphological evolution of RBCs under different storage conditions (normoxia versus hypoxia), and explored its potential application in transfusion support at high altitude. Methods RBC units stored under normoxic (21% O2) or hypoxic (8% O2) conditions were collected together with units obtained from Beijing (≈500 m) and Lhasa (≈3 600 m). RBC images were acquired every week to construct a time-series dataset. A nine-class RBC morphological recognition model was established based on the YOLOV5s algorithm and the morphological index (MI) and smooth disc cell percentage (SDC%) were introduced as quality assessment parameters to compare the progression of storage leisions in RBCs under different storage conditions and from different geographic origins. Results In the normoxic storage group, MI began to decline significantly from week 3 onward; by week 5, MI had decreased by 21.08% and SDC% by 31.33%. In contrast, the hypoxic storage group showed declines of only 13.40% in MI and 20% in SDC%, with statistically significant differences between groups (P<0.01). RBCs stored at high altitude exhibited significantly slower morphological deterioration than those stored in the plains from week 2 onward. At week 5, MI in the plateau group was 83.96%, significantly higher than 76.61% in the plains group, suggesting that the hypoxic environment at high altitude helps preserve stored RBC morphology. Conclusion This study achieved a dynamic, deep learning-based assessment of storage lesions in plateau RBCs. The proposed MI and SDC% metrics enable quantitative evaluation of RBC morphological deterioration at high altitude and offer advantages including high throughput, noninvasiveness, and transferability. This model provides intelligent technical support for rapid quality testing of blood products in plateau regions.

Key words: High-altitude red blood cells, Storage lesion, Deep learning, Morphological index, Hypoxic storage