Super-Resolution Image Reconstruction Using Diffuse Source Models

Abstract

Image reconstruction is central to many scientific fields, from medical ultrasound and sonar to computed tomography and computer vision. While lenses play a critical reconstruction role in these fields, digital sensors enable more sophisticated computational approaches. A variety of computational methods have thus been developed, with the common goal of increasing contrast and resolution to extract the greatest possible information from raw data. This paper describes a new image reconstruction method named the Diffuse Time-domain Optimized Near-field Estimator (dTONE). dTONE represents each hypothetical target in the system model as a diffuse region of targets rather than a single discrete target, which more accurately represents the experimental data that arise from signal sources in continuous space, with no additional computational requirements at the time of image reconstruction. Simulation and experimental ultrasound images of animal tissues show that dTONE achieves image resolution and contrast far superior to those of conventional image reconstruction methods. We also demonstrate the increased robustness of the diffuse target model to major sources of image degradation, through the addition of electronic noise, phase aberration, and magnitude aberration to ultrasound simulations. Using experimental ultrasound data from a tissuemimicking phantom containing a 3 mm diameter anechoic cyst, the conventionally reconstructed image has a cystic contrast of −6.3 dB whereas the dTONE image has a cystic contrast of −14.4 dB.

Single image super-resolution of medical ultrasound images using a fast algorithm

“…CONCLUSION This paper studied a new super-resolution algorithm for ultrasound images. By exploring the properties of the decimation matrix in the Fourier domain, we were able to calculate an analytical solution of the super-resolution problem with an ℓ2-norm regularizer and were able to embed this analytical solution into an ADMM framework for a more general ℓp-norm regularizer (p ∈ [1,2]). Due to the implementation of the analytical solution, the proposed method allowed to reduce the computation time comparing with the classical method.…”

“…However, US images suffer from a relatively low contrast, reduced spatial resolution and low signal-to-noise ratio. Even though advances in ultrasonic hardware have improved the resolution of US images during the last 15-20 years, e.g., [1,2], post-processing techniques enhancing US image resolution are still appealing due to the physical limitations of device-based solutions. In an ultrasound system, there are frequency (RF) data, in phase/quadrature (IQ) data and B-mode image (also called displayed image) [3], whose relationships are shown in Fig.…”

Ultrasound applications to medicine part i: physical principles.

“…[102] Knowing the speed of ultrasound in a medium, one can calculate the depth of reflection points correspond to generate echoes, if it can find the time that it takes to return each echo produced by a change in tissue [103]. Above is the principle on which teams are based diagnostic imaging reconstruction with ultrasound [104]. In the human body the ultrasound velocity is between 1,500 and 1,660 / , being in the bones of 3.360 / [108].…”

Motion estimation-based image enhancement in ultrasound imaging

“…In US, image quality can be improved in either pre-or post-processing. The former is usually achieved through the modernization of the US scanners, for instance by using high frequency transducers (at the cost of limited penetration depth [4]), backprojection image recovery methods [5] or by designing a proper adaptive beamforming (ABF) algorithm such as Diffuse Time-domain Optimized Near-field Estimator (dTONE) [6] to replace the conventional delay and sum beamforming (at the cost of high computational load). Unfortunately, such techniques lead to tremendous instrumentation constraints that hinder the experimental reproducibility.…”