MR Spectroscopy Signal Quantification using Deep Learning
Recrutement: 
Recrutement en cours/passé: 
Recrutement en cours
Periode: 
2017

Scientific context

MR spectrocopy is a medical imaging technique allowing to determine the bio- chemical contents of brain tissue or other organs. From a 1D signal acquired on an MR scanner, the relative concentrations of a given set of metabolites are usually estimated using a nonlinear fit of a parameterized signal, based on a metabolite basis set, to the acquired signal [4, 3]. Deep learning is a machine learning technique widely recognized now for its success for the image recognition problem. It can also be used as an approximation method in accordance with the universal approximation theorem [5, 1, 2]. We will use this property to learn the nonlinear fit procedure: we will learn the function that, given an MR spectroscopic signal, return the relative concentrations.

Objectives

The objective of this internship is to present new solutions for the quantification of MR spectroscopic signals using recent development in machine learning. More specifically, we will use the deep learning frameworks to estimate the amplitude (i.e. the relative concentrations) of all the metabolite signals.

Keywords

deep learning, signal processing, MR spectroscopy, parameter estimation

Skills

Strong knowledge in at least one of the following fields is required:

  • Machine learning (deep learning)
  • Signal processing
  • Applied mathematics;

The available code is written in Matlab and Python.

The successful candidate is expected to be autonomous and show strong motivation and interest in multidisciplinary research (signal processing, machine learning and mr spectroscopy in a medical context).

Applications

Interested applicants are required to send a cover letter, CV and any other relevant documents (reference letter, recent transcripts of marks,...) to: michael.sdika[at]creatis. insa-lyon.fr and helene.ratiney[at]creatis.insa-lyon.fr.

Gratuity ≈ 500 e/mois.

References

[1] G Gybenko. Approximation by superposition of sigmoidal functions. Mathematics of Control,Sinals and Systems, 2(4):303–314, 1989.

[2] Kurt Hornik. Approximation capabilities of multilayer feedforward networks. Neural networks,4(2):251–257, 1991.

[3] H. Ratiney, M. J. Albers, H. Rabeson, and J. Kurhanewicz. Semi-parametric time-domain quantification of hr-mas data from prostate tissue. NMR in Biomedicine, 23(10), 12/2010 2010.

[4] H. Ratiney, M. Sdika, Y. Coenradie, S. Cavassila, D. van Ormondt, and D. Graveron-Demilly. Time-domain semi-parametric estimation based on a metabolite basis set. Nuclear Magnetic Resonance in Biomedicine, 18:1–13, 2005.

[5] Sho Sonoda and Noboru Murata. Neural network with unbounded activations is universal approximator. CoRR, abs/1505.03654, 2015.