anne-sophie montcuquet 1,2, lionel herve 1, jean-marc dinten 1, jérôme i. mars 2 1 leti / lisa –...

1
Anne-Sophie MONTCUQUET 1,2 , Lionel HERVE 1 , Jean-Marc DINTEN 1 , Jérôme I. MARS 2 1 LETI / LISA – CEA, Minatec, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France. 2 GIPSA-Lab / Dept Images – Signal, 961 Rue de la Houille Blanche, BP 46, 38402 Saint Martin d'Hères, France. [email protected] Introductio n Fluorescent imaging in diffusive media is an emerging imaging modality for medical applications: injected fluorescent markers (in multiplexing, several specific markers are used) bind specifically to targeted compounds, like carcinoma. The region of interest is illuminated with near infrared light and the emitted back fluorescence is analyzed to localize the fluorescence sources. For medical diagnostic application, thick media have to be investigated: as the fluorescence signal gets exponentially weak with the light travel distance, any disturbing signal - such as biological tissues autofluorescence - may be a limiting factor. To remove these unwanted contributions, or to separate different fluorescent markers, a spectroscopic approach and a blind source separation method are explored. We present in this poster a feasibility experiment on an optical phantom in which a marked tumor is simulated. We show how an NMF unmixing preprocessing eradicates the autofluorescence signal of the phantom and allows to get more accurate 3-D reconstructions of the specific marker by Fluorescence Diffuse Optical Tomography (FDOT). Fluorescent imaging Image CEA Non-negative Matrix Factorization Therapeutic window Formal statement NMF applied to spectroscopy Given a non-negative matrix , find non-negative matrices and such that: (P stands for the number of fluorescent sources to unmix) Challenge We want to unmix several fluorescence spectra: A spectroscopic approach is chosen. We do not have much information about the fluorescence spectra : A blind source separation method is required. Conclusion Algorithm Multiplicative update rules Update of A: 1. Initialization of matrices A (constant) and S (spectra models) with A 0 and S 0 > 0 2. Minimization of the cost function F Update, in turn, of A and S Algorithm steps Fluorescent probes location Non-negative Matrix Factorization: a blind sources separation method applied to optical fluorescence spectroscopy and multiplexing The use of red light limits the biological tissues absorption Injected fluorescent markers bind specifically to a given molecule Experimental set-up Update of S: V The fluorescence signal is collected along a line of N x detectors by a spectrometer coupled with a CCD camera: a N x x N λ acquisition is measured. A translation stage, covering N y steps, is then used to get a scanning of the whole object. Feasibility experiment Results (1/2) Autofluorescen ce (PPIX) ICG-LNP Results (2/2) NMF decomposition gives two distinct fluorescence spectra . An original regularized NMF algorithm is used. . Experiments were performed ex vivo on optical phantoms to assess the capacity of NMF to unmix overlapping specific fluorescence and autofluorescence spectra. . The NMF algorithm is also suitable for in vivo experiments. . Spectrally resolved acquisitions combined to NMF processing successfully separate different fluorescent markers or filter different fluorescence contributions of interest from measurements impaired by autofluorescence. . NMF preprocessing improves FDOT reconstructions of specific fluorescent markers distributions by removing the disturbing fluorescence signals Intensity data Forward model: finite volume method *Andor technologies * * 690 nm

Upload: justina-caldwell

Post on 06-Jan-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Anne-Sophie MONTCUQUET 1,2, Lionel HERVE 1, Jean-Marc DINTEN 1, Jérôme I. MARS 2 1 LETI / LISA – CEA, Minatec, 17 rue des Martyrs, 38054 Grenoble Cedex

Anne-Sophie MONTCUQUET 1,2, Lionel HERVE 1, Jean-Marc DINTEN 1, Jérôme I. MARS 2

1 LETI / LISA – CEA, Minatec, 17 rue des Martyrs, 38054 Grenoble Cedex 9, France. 2GIPSA-Lab / Dept Images – Signal, 961 Rue de la Houille Blanche, BP 46, 38402 Saint Martin d'Hères, France.

[email protected]

IntroductionFluorescent imaging in diffusive media is an emerging imaging modality for medical applications: injected fluorescent markers (in multiplexing, several specific markers are used) bind specifically to targeted compounds, like carcinoma. The region of interest is illuminated with near infrared light and the emitted back fluorescence is analyzed to localize the fluorescence sources. For medical diagnostic application, thick media have to be investigated: as the fluorescence signal gets exponentially weak with the light travel distance, any disturbing signal - such as biological tissues autofluorescence - may be a limiting factor.

To remove these unwanted contributions, or to separate different fluorescent markers, a spectroscopic approach and a blind source separation method are explored. We present in this poster a feasibility experiment on an optical phantom in which a marked tumor is simulated. We show how an NMF unmixing preprocessing eradicates the autofluorescence signal of the phantom and allows to get more accurate 3-D reconstructions of the specific marker by Fluorescence Diffuse Optical Tomography (FDOT).

Fluorescent imaging

Image CEA

Non-negative Matrix Factorization

Therapeutic window

Formal statement

NMF applied to spectroscopy

Given a non-negative matrix , find non-negative matrices

and such that:

(P stands for the number of fluorescent sources to unmix)

Challenge

We want to unmix several fluorescence spectra:

A spectroscopic approach is chosen.

We do not have much information about the fluorescence spectra :

A blind source separation method is required.

Conclusion

Algorithm

Multiplicative update rules

Update of A:

1. Initialization of matrices A (constant) and S (spectra models) with A0 and S0 > 0

2. Minimization of the cost function F

Update, in turn, of A and S

Algorithm steps

Fluorescent probes location

Non-negative Matrix Factorization: a blind sources separation method applied to optical fluorescence spectroscopy and multiplexing

The use of red light limits the biological tissues absorption

Injected fluorescent markers bind specifically to a given molecule

Experimental set-up

Update of S:

V

The fluorescence signal is collected along a line of Nx detectors by a spectrometer coupled with a CCD camera: a Nx x Nλ acquisition is measured. A translation stage, covering Ny steps, is then used to get a scanning of the whole object.

Feasibility experiment Results (1/2)

Autofluorescence (PPIX)

ICG-LNP

Results (2/2)NMF decomposition gives two distinct fluorescence

spectra

. An original regularized NMF algorithm is used.

. Experiments were performed ex vivo on optical phantoms to assess the capacity of NMF to unmix overlapping specific fluorescence and autofluorescence spectra.

. The NMF algorithm is also suitable for in vivo experiments.

. Spectrally resolved acquisitions combined to NMF processing successfully separate different fluorescent markers or filter different fluorescence contributions of interest from measurements impaired by autofluorescence.. NMF preprocessing improves FDOT reconstructions of specific fluorescent markers distributions by removing the disturbing fluorescence signals

Intensity data

Forward model: finite volume method

*Andor technologies

**

690 nm