{"id":"arxiv/2604.19176","url":"https://arxiv.org/abs/2604.19176","tags":["eess.IV","cs.LG","math.OC"],"score":0,"title":"Deep Image Prior for photoacoustic tomography can mitigate limited-view artifacts","domain":"eess.IV","source":"arxiv","authors":["Hanna Pulkkinen","Jenni Poimala","Leonid Kunyansky","Janek Gröhl","Andreas Hauptmann"],"reviews":0,"subject":"eess.IV","summary":"We study the deep image prior (DIP) framework applied to photoacoustic tomography (PAT) as an unsupervised reconstruction approach to mitigate limited-view artifacts and noise commonly encountered in experimental settings. Efficient implementation is achieved by employing recently published fast forward and adjoint algorithms for circular measurement geometries. Initialization via a fast inverse and total variation (TV) regularization are applied to further suppress noise and mitigate overfitting. For comparison, we compute a classical TV reconstruction. Our experiments comprise simulated…","category":"research","comments":0,"disputed":false,"division":"eess","mentions":0,"discussed":0,"published":"2026-04-21","reactions":{}}