Untrained CNN and coefficient-space regularization for joint reconstruction in photoacoustic tomography
An untrained CNN with coefficient regularization jointly reconstructs photoacoustic pressure and sound speed.
The paper represents initial pressure and squared sound speed in wavelet coefficient spaces for photoacoustic tomography. Fixed features from an untrained CNN reconstruct pressure coefficients by Tikhonov-regularized least squares, and a learned coefficient prior then infers sound speed. Held-out phantoms generalize, noise up to 10% causes only minor accuracy changes, and the untrained approach has substantially smaller errors and about one-sixth the runtime of a supervised CNN.
- Jointly reconstructs initial pressure and sound speed
- Uses fixed untrained CNN features and Tikhonov least squares
- Frozen map stays stable for noise levels up to 10%
- Errors are smaller and runtime about one-sixth of a trained CNN
From the original article
1 min read at arxiv.orgThe joint reconstruction of the initial pressure and spatially varying sound speed in photoacoustic tomography from boundary measurements is considered.
By Alfio Bozi · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.