Page 1
Atlas-based patterns detection. Examples from asynchronous hearts.
Nicolas Duchateau, Mathieu De Craene, Gemma Piella, Bart Bijnens, Alejandro Frangi
Information and Communication Technologies Department, Universitat Pompeu Fabra, Barcelona, Spain .
Adelina Doltra, Etel Silva, Marta Sitges
Hospital Clínic de Barcelona, IDIBAPS, Spain.
Saturday, March 19, 2011
Page 2
2
What is an atlas?
Average anatomy+
Variation around this average
=
Saturday, March 19, 2011
Page 3
Fibers VelocitiesThickness Strain
GeometryPeyrat et al., T
MI 2007
Atlas-based integrated biomarkers
Saturday, March 19, 2011
Page 4
Probabilistic biomarkersEncode normalityP-value = abnormality
Integrated image-based biomarkers
d1 <?> d2 d1 d2
patient
atlas
d = ???
Atlas-based quantification of motion abnormalities
De Craene et al. FIMH 2009
Saturday, March 19, 2011
Page 5
Atlas-based quantification of motion abnormalities
Radial velocity
Long. velocity
p-value (log scale)
(mm/s)
(mm/s)
Atlas
Healthy subjects
Patient to study
Duchateau et al. MEDIA 2011
Saturday, March 19, 2011
Page 6
How to plot p-values?
p-value (log scale)
Duchateau et al. MEDIA 2011
! Temporal evolution at a fixed anatomical point
p-value (log scale)
Red = large abnormality
! Local maps at fixed time t
Saturday, March 19, 2011
Page 7
Base
Apex
Time
IVC Systole Diastole
Blue = Inward (vr<0)Red = Outward (vr>0)
p-value (log scale)
How to plot p-values?
! Spatiotemporal maps of abnormality
Duchateau et al. MEDIA 2011
Saturday, March 19, 2011
Page 8
Patient classification into specific etiologies of HF [Parsai]
" Correction of specific mechanisms of dyssynchrony conditions response
" Predicitive value of specific classes
• Septal flash [Parsai]• Septal rebound stretch [De boeck]• Apical transverse motion [Voigt]
Why quantifying abnormalities? CRT context
C1= Intra DYS
C2= Inter DYS
C3= Long AV
C4= Short AV
C5= Other
Response rate per
classvs. Global response
rate ?
Parsai et al., EHJ 2009
De Boeck et al., EJHF 2009
Voigt et al., EHJ 2009
Saturday, March 19, 2011
Page 9
Fig.3: Septal flash mechanismWhat is a “septal flash” ?
Healthy volunteer CRT candidate with SF
4
Parsai, Bijnens et al., EHJ 2009
Saturday, March 19, 2011
Page 10
2D echo, 4-chamber view
14
Patient population
21 Healthy volunteers
! 60 frames/s0.24 x 0.24 mm2
88 candidates OFF / ON / FU (11±2 months)EF < 35%, QRS duration > 120ms, and (or) NYHA class III-IV
! 30 frames/s0.24 x 0.24 mm2
CRT response:Clinical 6min walking test increase " 10% or NYHA class reduction " 1 point
Echocardiographic LV end-systolic volume reduction " 15%
(in alive patients without heart transplantation)
Saturday, March 19, 2011
Page 11
IVC Systole Diastole
OFF
PV-maps vs visual inspection: agreement?
Mid inferoseptal
!" #$%!"!" !! "
#$%!" ! #$
Atlas-based
M-m
ode
Cohen’s Kappa = 0.88Observed agreement = 0.94
Saturday, March 19, 2011
Page 12
IVC Systole Diastole
OFF
[Bleeker] Response rate with the current guidelines:
0.7 (clinical response) 0.5 (echo response)
" Ejection fraction <35%
" QRS duration >120ms
" and/or NYHA classification III-IV)
Predictive value of SF at baseline?
RespondersResponders Response rate
Response rate
SensitivitySensitivity SpecificitySpecificity
Total Clin. Echo. Clin. Echo. Clin. Echo. Clin. Echo.
CRT 88 72 53 0.82 0.60
SF (Visual) 55 48 44 0.87 0.80 0.67 0.83 0.56 0.69
SF (Atlas) 60 52 44 0.87 0.73 0.72 0.83 0.50 0.54
Bleeker et al. AJC 2006
Saturday, March 19, 2011
Page 13
CRT #9Septal flash
CRT #8Septal flash
CRT #12Left-right
interaction
Blue = Inward (vp<0)Red = Outward (vp>0)
IVC Systole Diastole
OFF Follow-upLocal p-value
(log scale)
Monitor CRT outcome
Saturday, March 19, 2011
Page 14
! Extend to 3D imaging modalities (tMRI)
Future work
3D tMRI image provided by KCL, London.Saturday, March 19, 2011
Page 15
15
! Work with the full strain tensor information
Non-responder to CRT
OFF Follow-up
Future work
3D Ultrasound images provided by Hospital Clinic, BCN.Saturday, March 19, 2011
Page 16
16
! Work with the full strain tensor information
Responder to CRT
OFF Follow-up
Future work
3D Ultrasound images provided by Hospital Clinic, BCN.Saturday, March 19, 2011
Page 17
Conclusions! Atlas-based patterns detection
! Generic concept, extendable to other pathological patterns and image modalities
! Automatic and reproducible analysis ! Information available at any spatiotemporal location! Comparison OFF/Follow-up performed at the same
location! Intrinsic notion of abnormality
! Application to CRT and Septal Flash detection! Automatic/Visual detection in good statistical agreement! Quantify patient evolution at follow-up
! How closer the patient gets to “normality”! Applied to a large (88) population of patients
Saturday, March 19, 2011
Page 18
Acknowledgements
! Projects! euHeart. FP7 european project! cvREMOD. Spanish project for technology transfer
! Institutions ! Universitat Pompeu Fabra! Hospital Clínic de Barcelona
Saturday, March 19, 2011