eduzhai > Applied Sciences > Engineering >

CNN-Based Image Reconstruction Method for Ultrafast Ultrasound Imaging

  • king
  • (0) Download
  • 20210507
  • Save

... pages left unread,continue reading

Document pages: 24 pages

Abstract: Ultrafast ultrasound (US) revolutionized biomedical imaging with itscapability of acquiring full-view frames at over 1 kHz, unlocking breakthroughmodalities such as shear-wave elastography and functional US neuroimaging. Yet,it suffers from strong diffraction artifacts, mainly caused by grating lobes,side lobes, or edge waves. Multiple acquisitions are typically required toobtain a sufficient image quality, at the cost of a reduced frame rate. Toanswer the increasing demand for high-quality imaging from single-shotacquisitions, we propose a two-step convolutional neural network (CNN)-basedimage reconstruction method, compatible with real-time imaging. A low-qualityestimate is obtained by means of a backprojection-based operation, akin toconventional delay-and-sum beamforming, from which a high-quality image isrestored using a residual CNN with multi-scale and multi-channel filteringproperties, trained specifically to remove the diffraction artifacts inherentto ultrafast US imaging. To account for both the high dynamic range and theradio frequency property of US images, we introduce the mean signed logarithmicabsolute error (MSLAE) as training loss function. Experiments were conductedwith a linear transducer array, in single plane wave (PW) imaging. Trainingswere performed on a simulated dataset, crafted to contain a wide diversity ofstructures and echogenicities. Extensive numerical evaluations demonstrate thatthe proposed approach can reconstruct images from single PWs with a qualitysimilar to that of gold-standard synthetic aperture imaging, on a dynamic rangein excess of 60 dB. In vitro and in vivo experiments show that trainingsperformed on simulated data translate well to experimental settings.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...