New ESC method cuts ultrasound image registration cost 3.5x

A research team led by Jipeng Yan introduced Extremum Seeking Control (ESC) to replace traditional gradient descent (GD) in parametric image registration for ultrasound localization microscopy (ULM). ULM requires tissue motion correction to achieve super-resolution, but explicit gradient evaluation in each iteration is computationally expensive. ESC avoids this by integrating perturbed and demodulated similarity metrics across iterations, approximating gradient information without differentiation. Using simulated ground-truth motions from a beating ex vivo porcine heart dataset, ESC matched GD's registration accuracy and convergence while cutting per-iteration cost by 3.5×.

The method was embedded in a two-stage motion correction pipeline: affine registration for global motion and B-spline registration for local deformation. Applied to ULM imaging of a beating ex vivo porcine heart with 2.4 MHz diverging-wave ultrasound, ESC achieved a spatial resolution of 219 μm—well below the half-wavelength diffraction limit of 321 μm. This demonstrates ESC as a practical, computationally efficient alternative for high-quality super-resolution ultrasound imaging, with potential for real-time clinical applications.