Type de stage
Stage de Master 2 / dernière année d’école d’ingénieur
Date ou durée du stage
2022, 5 ou 6 mois
Laboratoires
Sorbonne Université : Laboratoire d’Imagerie Biomédicale et Institut ∂’Alembert
Contexte
Tremendous progress has been made during the last decade in ultrasound (US) medical imaging leading to more quantitative measurements, higher resolution images and novel clinical applications. These advances were triggered by the availability of novel instrumentation (programmable US scanners, probes with high sensitivity) and increasing computational power enabling the use of sophisticated signal processing techniques.
Quantitative ultrasound (US) imaging of cortical bone is an emerging research topic for which we are expecting rapid developments [1,2]. The current quality of bone images is low, preventing the identification of tissue heterogeneities and a proper description of the bone-marrow interface which is of clinical importance. The lower quality of bone ultrasound images, compared to soft tissue images, is due to the limited efficiency of standard image reconstruction techniques to account for the complexity of US propagation in bone (refraction at the soft tissues-bone interface, strong attenuation, coupling of different wave types propagating in a solid).
Reasearchers at LIB have recently obtained the first in vivo images of human cortical bone, enabling a measurement of bone thickness and material properties [2]. However, the images are of relatively poor quality. Novel methods to reconstruct the ultrasound image from radiofrequency signals, including inverse problem and machine learning approches have a high potential to improve the image quality.
References :
[1] Q. Grimal and P. Laugier, “Quantitative Ultrasound Assessment of Cortical Bone Properties Beyond Bone Mineral Density,” IRBM, vol. 40, no. 1. pp. 16–24, 2019.
[2] G. Renaud, P. Kruizinga, D. Cassereau, and P. Laugier, “In vivo ultrasound imaging of the bone cortex,” Phys. Med. Biol., 2018.
Objectif et méthodes
The objective of the project is to propose a framework for bone imaging using an inverse problem (IP) approach based on the matrix form of the imaging operator [3,4]. According to this approach, the interaction of the ultrasound beam with the tissues, and the emission/reception chains are modeled as a linear problem in matrix form. The beamforming operation then consists in determining the strength of scatterers distributed in the imaged medium. While the formulation of the problem is rather straighforward, the calculation of solutions is challenging as it requires a proper regularization of the inverse (optimization) problem. The trainee will investigate various strategries for regularization, in particular those including a priori information on imaging physics.
References :
[3] B. Berthon et al., “Spatiotemporal matrix image formation for programmable ultrasound scanners,” Phys. Med. Biol., vol. 63, no. 3, 2018.
[4] E. Ozkan, V. Vishnevsky, and O. Goksel, “Inverse Problem of Ultrasound Beamforming with Sparsity Constraints and Regularization,” IEEE Trans. Ultrason. Ferroelectr. 65, no. 3, pp. 356–365, 2018.
Missions
The trainee will implement the novel imaging algorithms and evaluate their performance using dedicated ultrasound acquisitions on artificial objects and bones ex vivo an in vivo. The performance of the new algorithms will be compared to conventional beamforming approaches.
Compétences
The successful candidate is expected to have a strong background in signal processing and/or acoustics. He/She should have basic experimental, computer, and writing skills.
Rémunération
Gratifications de stage
Contacts
Quentin GRIMAL (quentin.grimal[at]sorbonne.universite.fr, Laboratoire d’Imagerie Biomédicale)
François OLLIVIER (francois.ollivier[at]sorbonne.universite.fr, Institut ∂’Alembert)