Photoacoustic Imaging with Conditional Priors from Normalizing Flows

TitlePhotoacoustic Imaging with Conditional Priors from Normalizing Flows
Publication TypeConference
Year of Publication2021
AuthorsRafael Orozco, Ali Siahkoohi, Gabrio Rizzuti, Tristan van Leeuwen, Felix J. Herrmann
Conference NameNeural Information Processing Systems (NeurIPS)
Month12
Keywordsconditional prior, deep image, MAP, NIPS, normalizing flow, Photoacoustic, Variational Inference
Abstract

Photoacoustic imaging is a biomedical imaging technique based on the photoacoustic effect. It leverages the interplay between optics and acoustics as a mean to circumvent the limitations of imaging modalities relying on single-type physics. Light beams generated by a pulsed laser can penetrate biological tissues by several centimeters, and are absorbed based on oxygen saturation or hemoglobin concentration. While optical absorption is in principle an ideal parameter for medical imaging (e.g., with respect to the detection of cancerous tissue), strong scattering imposes important limitations in its imaging resolution. Ultrasonics, on the other hand, can theoretically provide resolution of medical diagnostic value, but produce images of mechanical properties whose contrasts are not sensitive. In photoacoustics, optical and acoustic effects are combined to gain the best of both worlds. Under conditions of thermal and stress confinement, thermal energy can efficiently build up in biological tissues, which in turn undergo thermal expansion and effectively act as a spatially distributed acoustic source. In photoacoustic imaging, the actual object of interest is the induced source, as it is directly related to optical absorption and can be recovered with a relatively higher resolution than pure optical imaging, based on the acquired ultrasonic data.

Notes

(NIPS, virtual)

URLhttps://slim.gatech.edu/Publications/Public/Conferences/NIPS/2021/orozco2021NIPSpicp/deep_inverse_2021.html
Presentation
URL2
Software
Citation Keyorozco2021NIPSpicp