Machine learning cell segmentation from intravital microscopy to determine mechanical properties of stored transfused erythrocytes in vivo

Background and Objective: Red blood cell (RBC) deformability is a key determinant of blood mechanical and fluid properties, as well as microcirculatory flow. Human RBCs, approximately 8 μm in diameter, must repeatedly deform to pass through much narrower capillaries where most of the oxygen exchange occurs. Even slight reductions in RBC deformability can markedly affect functional capillary density and microvascular oxygen delivery. This study compares the mechanical properties and deformability of stored RBCs with those of native RBCs to assess the impact of stored blood transfusions on microvascular function. Methods: Capillaries in the rat cremaster muscle were imaged before and after transfusion with two-week-old stored RBCs via high-speed intravital microscopy with a total magnification of 1800x. A Python script employing Cellpose (1), a deep learning-based segmentation algorithm, was developed to extract RBC shapes from each video frame. Discrete deformation gradient tensors were computed under a small-strain approximation using a weighted least-squares method on clusters of edge points from consecutive Cellpose masks. Cauchy stress and strain tensors, along with mechanical properties, were derived for both native and stored RBCs pre- and post-transfusion. Results: RBCs stored for two weeks exhibited significantly higher shear modulus, Young’s modulus, and bulk modulus compared to native cells. No statistically significant differences were observed between stored and native RBCs in average, maximum, or minimum wall shear stress. Conclusions: The integration of machine learning-based cell segmentation and an automated Python workflow greatly increased both the number of cells that could be analyzed and the efficiency with which they could be processed compared with manual methods. Previously, in vivo RBC deformation studies were limited by uncertainty in tracking individual cell groups. The present approach enables high-throughput group-level comparisons with sufficient statistical power to detect significant differences between pre- and post-transfusion conditions. The findings demonstrate that RBC storage impairs mechanical performance and heightens the risk of microhemodynamic dysfunction or capillary plugging following transfusion of stored blood. Moreover, the versatility of the Cellpose tool and its integration within this mathematical framework provide broad potential for future applications, including large-scale analyses of oxygen diffusion relative to endothelial exposure in RBCs under diverse physiological conditions, as well as studies of deformation and adhesion behavior in other cell types, such as leukocytes. Reference: 1. C. Stringer, et al. Cellpose: a generalist algorithm for cellular segmentation. Nat. Methods, 18 (1) (2021), pp. 100-106 This abstract was presented at the American Physiology Summit 2026 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.