Abstract Number: PB0967
Meeting: ISTH 2021 Congress
Background: Shear-mediated platelet adhesion is essential to initiating clot formation in vascular diseases and prosthetic cardiovascular devices. However, sparse and noisy raw in vitro image data continues to hamper validation of predictive computational models of platelet adhesion under flow.
Aims: To determine if adhesion dynamics is age-specific and intracellular Ca2+-dependent, by applying a novel machine learning (ML)-guided approach for accurate image analysis of flowing adult and cord platelets.
Methods: Gel-filtered platelets, prepared from blood drawn from consenting healthy adult volunteers or cord blood obtained from neonates delivered via Caesarean sections under Stony Brook University IRB-approved protocols, were diluted to 150,000/μl and perfused at wall shear stress of 30 dyne/cm2 through 100 µg/ml vWF-coated microchannels, with adhesion events captured at 1000 fps. Platelets were also pre-treated with 20 μM BAPTA-AM to evaluate intraceullar Ca2+ dependence. A semi-unsupervised learning system (SULS) classified platelet morphology from DIC microscope images, from which geometric parameters and rolling direction were calculated. Rotational angles and velocities fit to a modified Jeffery orbit model were compared across the age and Ca2+ treatment groups using two-sample t-tests.
Results: SULS accurately predicted moving platelet boundaries (Fig. 1A), with false prediction area of 0.728 μm2. For both adult and cord platelets, we observed distinct periods characterizing longer lift-off from and shorter reattachment to the vWF surface (p<0.05, Fig. 2B-C). Cord platelets (n=21) flip non-significantly faster than adult platelets (n=70, p>0.05, Fig. 2D). Intracellular Ca2+-depleted cord platelets (n=3) show a 1.16-fold increase in peak rotational speed compared to untreated cord platelets (p>0.05, Fig. 2D).
Conclusions: Our integrated ML-microscopy approach allows accurate segmentation of flipping platelets, showing heterogeneity of platelet motion during adhesion and possible dependence on age and intracellular Ca2+ availability. This framework bridges sparse in vitro data and multiscale computational models, which may predict physiologically significant platelet dynamics beyond the capabilities of current imaging technology.
To cite this abstract in AMA style:Sheriff J, Wang P, Zhang P, Zhang Z, Bahou W, Deng Y, Bluestein D. Machine Learning-guided Analysis of Adult and Cord Platelet Adhesion Dynamics [abstract]. Res Pract Thromb Haemost. 2021; 5 (Suppl 2). https://abstracts.isth.org/abstract/machine-learning-guided-analysis-of-adult-and-cord-platelet-adhesion-dynamics/. Accessed December 6, 2023.
« Back to ISTH 2021 Congress
ISTH Congress Abstracts - https://abstracts.isth.org/abstract/machine-learning-guided-analysis-of-adult-and-cord-platelet-adhesion-dynamics/