Generative adversarial network-based photoacoustic image reconstruction from bandlimited and limited-view data

Francis Kalloor Joseph, Aayush Arora, Parimala Kancharla, Mithun Kuniyil Ajith Singh, Wiendelt Steenbergen, Sumohana S. Channappayya

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

14 Downloads (Pure)

Abstract

Ultrasound transducers used in photoacoustic imaging are bandlimited and have a limited detection angle, which degrades the reconstructed image quality. One way to address this problem is to have transducers with multiple frequency bands with acquisition around the sample. This approach is expensive and it is not feasible for systems with a handheld probe using a linear transducer array. In this work, we aim to develop a deep learning method for photoacoustic reconstruction from bandlimited and limited-view data. We have developed a Generative Adversarial Networks (GANs)-based framework conditioned with a photoacoustic measurement for image reconstruction. In this way, the transducer used in the measurement can be incorporated and the generator trying to compensate for the limited data problem. We have developed the model for a handheld photoacoustic system using a linear transducer array with 128 elements having a center frequency of 7MHz and -6dB bandwidth from 4-10 MHz. We trained the network using simulated blood vessel images and tested it on in vivo measurements from the human forearm. We have compared the reconstructed images using the proposed method with the time-reversal on simulated data for detection using a bandlimited and directional transducer and compared it using the ground truth. Further, we compare our results to the in vivo images from the system which uses a delay and sum algorithm. The results from both simulations and experiments show that the proposed approach can remove bandlimited and limited view artifacts and can achieve a better image quality.

Original languageEnglish
Title of host publicationPhotons Plus Ultrasound
Subtitle of host publicationImaging and Sensing 2021
EditorsAlexander A. Oraevsky, Lihong V. Wang
PublisherSPIE Press
ISBN (Electronic)9781510641198
DOIs
Publication statusPublished - 5 Mar 2021
EventPhotons Plus Ultrasound: Imaging and Sensing 2021 - Virtual, Online, United States
Duration: 6 Mar 202111 Mar 2021

Publication series

NameProgress in Biomedical Optics and Imaging - Proceedings of SPIE
Volume11642
ISSN (Print)1605-7422

Conference

ConferencePhotons Plus Ultrasound: Imaging and Sensing 2021
CountryUnited States
CityVirtual, Online
Period6/03/2111/03/21

Keywords

  • artifact removal
  • convolutional neural networks
  • deep learning
  • generative adversarial network
  • image reconstruction
  • Photoacoustic imaging

Fingerprint

Dive into the research topics of 'Generative adversarial network-based photoacoustic image reconstruction from bandlimited and limited-view data'. Together they form a unique fingerprint.

Cite this