MIDL is one of the premiere conferences that covers a wide range of deep learning topics for medical imaging. Lucas's work focuses on developing efficient neural fields to model blood flow using CT perfusion imaging for acute ischemic stroke patients, surpassing even professional software in clinical practice. His approach enables radiologists to more accurately assess damaged brain regions.

Check the awards and link to the abstract here

The neural fields learn a representation of the CT perfusion data. Another neural field learns the blood flow, blood transit time, and delay, guided by a physics-informed loss function originating from a differential equation. With hash-encoding coordinate embeddings and meta-learned neural field initializations, the method is accelerated to achieve a compute time fast enough for clinical practice.
The neural fields learn a representation of the CT perfusion data. Another neural field learns the blood flow, blood transit time, and delay, guided by a physics-informed loss function originating from a differential equation. With hash-encoding coordinate embeddings and meta-learned neural field initializations, the method is accelerated to achieve a compute time fast enough for clinical practice.