Deep-learning optical flow for measuring velocity fields from experimental data.

Area of Science:

• Biophysics
• Soft Matter Physics
• Fluid Dynamics

Background:

• Microtubule-based active nematics exhibit complex fluid flows.
• Quantifying these flows is crucial for understanding active matter systems.
• Traditional methods like Particle Image Velocimetry (PIV) face limitations in dense or complex biological samples.

Purpose of the Study:

• To evaluate Deep Learning-based Optical Flow (DLOF) for quantifying active nematic flows.
• To compare DLOF performance against PIV under varying sample labeling densities.
• To establish DLOF as a robust tool for biophysical fluid flow measurement.

Main Methods:

• Deep convolutional neural networks were used for DLOF feature extraction from video frames.
• Flow velocity was validated using semi-automated particle tracking and passive tracer beads.
• DLOF and PIV methods were applied to microtubule-based active nematic samples with different labeling densities.

Main Results:

• DLOF provided more accurate velocity fields than PIV for densely labeled samples.
• DLOF successfully resolved flow details along the nematic director, overcoming PIV's contrast limitations at high densities.
• For sparsely labeled samples, DLOF yielded comparable results to PIV but with higher resolution.

Conclusions:

• DLOF is a superior method to PIV for quantifying active nematic flows, especially in dense conditions.
• DLOF offers higher resolution and overcomes limitations of PIV in complex biophysical systems.
• This study validates DLOF as a versatile and accurate tool for fluid flow analysis in active, soft, and biological matter.