An AI-enabled tool for quantifying overlapping red blood cell sickling dynamics in microfluidic assays

Area of Science:

• Biophysics
• Computational Biology
• Hematology

Background:

• Accurate quantification of sickle cell dynamics is crucial but challenging in dense cell populations due to overlap and aggregation.
• Traditional methods like dilution reduce statistical power and do not eliminate clustering in microfluidic sickling assays.
• Longitudinal studies require tracking large cell populations to capture cumulative, history-dependent changes.

Purpose of the Study:

• To develop an automated deep learning framework for quantifying red blood cell (RBC) populations in dense suspensions.
• To enable robust analysis of RBC morphological transitions under various biophysical conditions, overcoming limitations of cell overlap.
• To establish a scalable computational platform for investigating cellular biomechanics and assessing therapeutic efficacy.

Main Methods:

• Utilized AI-assisted annotation (Roboflow) and deep learning (nnU-Net) for segmentation and classification of RBCs in time-lapse microscopy.
• Integrated a watershed algorithm to separate overlapping cells, enhancing quantification accuracy in dense suspensions.
• Trained the model with limited labeled data, demonstrating high segmentation performance and addressing scarcity of manual annotations.

Main Results:

• The AI framework accurately quantifies RBC populations across varying densities, predicting temporal evolution of the sickle cell fraction.
• Achieved high segmentation performance, effectively handling challenges of cell overlap and limited manual annotations.
• Demonstrated potential to more than double experimental throughput in dense cell suspensions and capture drug-dependent sickling behavior.

Conclusions:

• The AI-driven framework provides a scalable and reproducible computational platform for investigating RBC biomechanics.
• This approach enables quantitative tracking of dynamic changes in RBC morphology, revealing distinct mechanobiological signatures.
• Facilitates more efficient assessment of therapeutic efficacy in microphysiological systems for sickle cell disease research.