ML Informed Ultrasound Registration

work with Christrian Herz

Helped him successfully train network for registering cardiac ultrasound

Next step: help him evaluate registration results, compare to other methods

next step: try in ITK-SNAP

Transformer Registration

Found architecture that trains transformer for image registration in 2-D

Key insight: linear patch embedding works for input, but output needs learnable smoothing after linear decoding.

next step: visualize attention maps

Inverse Consistent by Construction Registration

Back to Reports