Adaptation of multiscale function extension to inexact matching. Application to the mapping of individuals to a learnt manifold

28/08/2013
OAI : oai:www.see.asso.fr:2552:5101
DOI :

Résumé

Adaptation of multiscale function extension to inexact matching. Application to the mapping of individuals to a learnt manifold

Métriques

621
206
12.4 Mo
 application/pdf
bitcache://4c7f059e324bead8659d009c4e987df6be2b0b45

Licence

Creative Commons Aucune (Tous droits réservés)

Sponsors

Sponsors scientifique

logo_smf_cmjn.gif

Sponsors financier

logo_gdr-mia.png
logo_inria.png
image010.png
logothales.jpg

Sponsors logistique

logo-minesparistech.jpg
logo-universite-paris-sud.jpg
logo_supelec.png
Séminaire Léon Brillouin Logo
logo_cnrs_2.jpg
logo_ircam.png
logo_imb.png
<resource  xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
                xmlns="http://datacite.org/schema/kernel-4"
                xsi:schemaLocation="http://datacite.org/schema/kernel-4 http://schema.datacite.org/meta/kernel-4/metadata.xsd">
        <identifier identifierType="DOI">10.23723/2552/5101</identifier><creators><creator><creatorName>Nicolas Duchateau</creatorName></creator><creator><creatorName>Mathieu De Craene</creatorName></creator><creator><creatorName>Marta Sitges</creatorName></creator><creator><creatorName>Vicent Caselles</creatorName></creator></creators><titles>
            <title>Adaptation of multiscale function extension to inexact matching. Application to the mapping of individuals to a learnt manifold</title></titles>
        <publisher>SEE</publisher>
        <publicationYear>2013</publicationYear>
        <resourceType resourceTypeGeneral="Text">Text</resourceType><dates>
	    <date dateType="Created">Thu 10 Oct 2013</date>
	    <date dateType="Updated">Mon 25 Jul 2016</date>
            <date dateType="Submitted">Wed 19 Sep 2018</date>
	</dates>
        <alternateIdentifiers>
	    <alternateIdentifier alternateIdentifierType="bitstream">4c7f059e324bead8659d009c4e987df6be2b0b45</alternateIdentifier>
	</alternateIdentifiers>
        <formats>
	    <format>application/pdf</format>
	</formats>
	<version>9911</version>
        <descriptions>
            <description descriptionType="Abstract"></description>
        </descriptions>
    </resource>
.

SEE-GSI’13 conference – 30/08/2013 N. Duchateau1, M. De Craene2, M. Sitges1,V. Caselles3 1 Hospital Clínic – IDIBAPS - Universitat de Barcelona (ES) 2 Philips Research, Medisys, Suresnes (FR) 3 Universitat Pompeu Fabra, Barcelona (ES) How to quantify cardiac (ab)normal motion? Healthy volunteer CRT candidate with intra- ventricular dyssynchrony How to quantify cardiac (ab)normal motion? Duchateau et al. Med Image Anal (2011) d = ??? Atlas of motion Healthy subjects Patient to study Then… how to compare dyssynchrony patterns? d = ??? a) Manifold-learning from training setLowdimension Coordinatespace Highdimension Ambientspace Swiss roll (P=3) 2D embedding (M=2) 2D embedding (e.g.) N+1 images, P pixels Synthetic dataset CRT dataset ISOMAP =Tenenbaum et al. Science (2000) Duchateau et al. Med Image Anal (2012) b) Individual comparison to the whole population Duchateau et al. Med Image Anal (2012) b) Individual comparison to the whole population Analogy with: “pre-image problem” = Kwok andTsang, IEEETrans Neur Netw (2004) “denoising auto-encoders” -  Hinton and Salakhutdinov, Science (2006) -  Bengio et al. arXiv:1206.5538v2 (2012) b) Individual comparison to the whole population Duchateau et al. Med Image Anal (2012) Problem: -  Often no explicit formulation of the mappings Solution: > Formulation of a two-steps interpolation problem + combination of f and g to complete mapping to the manifold: •  Etyngier et al. NIPS (2007) •  Gerber et al. Med Image Anal (2010) Mapping points to the manifold Combination of f and g to complete mapping to the manifold Remark: reconstruction error Exact / inexact interpolation Also known as: •  Ridge regression •  Nyström extension •  Out-of-sample extension = Bengio et al. NIPS (2004) Exact / inexact interpolation Also known as: •  Ridge regression •  Nyström extension •  Out-of-sample extension = Bengio et al. NIPS (2004) Choice of optimal kernel scale Blaschko. Machine learning course @Philips (2013) Choice of optimal kernel scale First solution: locally adapted kernel bandwidth Duchateau et al. Med Image Anal (2012) Second solution: multiscale extension = Coifman and Lafon,ACHA (2006) Framework of “diffusion maps” to reach density invariance: -  For the manifold learning = diffusion maps -  For the out-of-sample extension = geometric harmonics For the out-of-sample extension: Geometric harmonics = Coifman and Lafon,ACHA (2006) Liu et al. CVIU (2012) In the manifold learning process: Diffusion distance / diffusion maps In our case = dyadic scales Second solution: multiscale extension Framework of “diffusion maps” to reach density invariance Second solution: multiscale extension Geometric harmonics = Coifman and Lafon,ACHA (2006) Bermanis et al.ACHA (2006) Geometric harmonics = Coifman and Lafon,ACHA (2006) Taken from: Bermanis et al.ACHA (2006) Second solution: multiscale extension Analogy with wavelets Mallat. Book WaveletTour (1998) Contribution: multiscale + inexact matching Contribution: multiscale + inexact matching Contribution: multiscale + inexact matching Contribution: multiscale + inexact matching http://www.texasheartinstitute.org Back to the application: CRT candidates Training = set of patients with Septal Flash (SF) Testing = new subjects (healthy or not) d = ??? Back to the application: CRT candidates Back to the application: CRT candidates Varying bandwidth Duchateau et al. Med Image Anal (2012) + FIMH (2013) Back to the application: CRT candidates Varying bandwidth Duchateau et al. Med Image Anal (2012) + FIMH (2013) Multiscale extension Present work Conclusions New patient Normality Known (?) pathologies •  Modeling a pathological pattern as deviation from normality •  Non-linear analysis •  Comparison of motion patterns in their whole (and not just peak or time-to-event measurements) •  Contributions: 1.  Interpolation problem now well-solved 2.  Combination of multiscale + inexact matching (and associated practical questions about parameters tuning) 3.  A-posteriori check of the varying bandwidth kernel low bias }  Previous collaborators: ◦  Universitat Pompeu Fabra, Barcelona (ES) G Piella, BH Bijnens ◦  University of Sheffield (UK) AF Frangi ◦  Hospital Clínic, Barcelona (ES) A Doltra, E Silva, MA Castel, L Mont, J Brugada }  Projects: ◦  MICINN (MTM2012-3077) ◦  GRC (2009-SGR-773) ◦  GenCat (ICREA Academia) Acknowledgements Proceedings: LNCS 8085:578-86, 2013. Thanks !!! Any questions…?