Artefact Removal (Physiological)

Rasmus M. Birn, Ph.D.

Overview

• The Problem

– Cardiac Fluctuations

– Respiration

• Solutions

– And potential pitfalls

– Improvements?

• Issues to consider

– Should we correct for physiological noise?

Functional MRI

t

t

Task-related

Neuronal

Activity

Physiologic noise(cardiac, respiration)

Subject motion

Scanner instabilities

Image reconstruction errors

Spontaneous

Neuronal

Activity

Hemodynamicsresting state

Typical task-related activation study

t

task

t

4

Resting-state Functional Connectivity

resting state

t

Rasm

us

M. B

irn

Resting-state Functional Connectivity

t

resting state

t

Rasm

us

M. B

irn

7

components

Independent Component Analysis

=

space (voxels)

com

pon

ents

time

time

space

x

Component #1

Component #2

…time

Sig

nal

time

Sig

nal

time (s)

%S

…

Physiological noise

0 40 80 120 160 200 240 280 320

time (s)

Cardiac

M.S. Dagli et al., NeuroImage 9, 1999

Respiration

time

00 TIME = 110 s

DF=90o400

200

B0

Field Map

Figure courtesy of J. Bodurka

Physiological Noise

B.P. Poncelet et al., Radiology, 185:645-651, 1992.

The brain is not a rigid body

Physio. fluctations can occur at low frequencies

Variations in breathing during rest DBOLD signal

PETCO2

BOLD (GM)

BOLD (WM)

R.G. Wise et al., NeuroImage 21, 2004

DPETCO2 correlated w/ DBOLD

Resting fluctuations in respiration

RVT = Respiration

Volume per Time

50 100 150 200 250 300 350

0

6

|Z|

group (n=11)

RVT-related changes

RVT

Effects of Heart Rate

C. Chang et al., Neuroimage 2009

Heart Rate

Birn et al. (unpublished)

Tools to correct for physiological noise

• Filtering - Works only if TR is short enough (< 400ms)

• IMPACT (Chuang et al., 2001) – if TR short enough

• Retrospective correction (k-space) (X. Hu et al., 1995)

• RETROICOR (Glover et al., 2000)

• CORSICA (V. Pelbarg, et al., 2007)

• PESTICA (E.B. Beall, et al., 2007)

• RVTcor (R.M. Birn, et al., 2006)

• RVHRcor (C. Chang, et al., 2009)

• ANATICOR (H-J. Jo, et al., )

• APPLECOR, PEARCOR (M Marx, et al., 2013)

• PSTCor (J.S. Anderson, et al., 2011)

• CompCor (Y. Behzadi, et al., 2007)

• FIX (L. Griffanti, et al., 2014)

• ICA-AROMA (###, et al., 2015)

Can filtering (<0.1Hz) reduce cardiac noise?

0 5 10 15 20

time (s)

Sig

na

l

Cardiac fluctuation (1.1 Hz) Sampled every TR (2s)

0 0.5 1 1.5 2Frequency (Hz)

0

0.1

0.2

0.3

Po

we

r

• Not always: Aliasing

Correction of physiological noise

RETROICOR (G. Glover et al., Magn. Reson. Med. 44, 2000.)

Pulse Oximeter

0 1 2 3 4 5 6 7 8 9 10

time (s)

Respiration

MR Images

fc

fr

sin( fc )

cos( fc )

sin( 2fc )

cos( 2fc )

sin( fr )

cos( fr )

sin( 2fr )

cos( 2fr )

Additional

Regressors:

Correction of physiological noise

time

Sig

nal

Reshuffle the data based on its

cardiac or respiration phase

0 1 2 3 4 5 6

Sig

na

l

phase

Correction: Cardiac + Respiration

X. Hu et al.,

Magn. Res. Med. 1995

At what stage should physio correction be done?

• Preprocessing

– Motion correction

– Physiological noise correction

– Slice time correction

– Nuisance Regression

– Spatial Smoothing

– Convert to percent signal change

– Temporal filter

• Define ROI (seed)

• Average EPI time course over ROI

• RegressionT.B. Jones et al., Neuroimage 42, 2008

RVHRcor (Respiration Volume + Heart Rate)

C. Chang et al., Neuroimage 2009

Nuisance Regression

H.J. Jo et al., NeuroImage 52, 2010

The effect of global signal regression

The global signal is

highly correlated w/

RVT

-1

-0.5

0

0.5

1

CC

: R

VT

~ g

lobal sig

nal

without global regression

with global regression

Connectivity maps

0.5

0.3

-0.1

0

(separate study, n = 18)

Global Signal Regression

K. Murphy, et al., Neuroimage, 2008

Can introduce anti-correlations

Global Signal Regression

Z.S. Saad, et al., Brain Connectivity, 2012

Can alter group differences …under certain conditions

Nuisance Regression

H.J. Jo et al., NeuroImage 52, 2010

PSTCor (Phase-shifted Soft Tissue Corr.)

J.S. Anderson et al.,

HBM 32, 2011

Connectivity maps

retroicor rvtcor rvhr wm/csf wm/csf+

global

No Corr.

Connectivity maps

retroicor rvtcor rvhr wm/csf wm/csf+

global

No Corr.

Differences in PCC connectivity vs. no corr.

ric rvtcor rvhr wm/csf wm/csf+

global

R2 (vs no correction)

0

0.45

R2

ric rvtcor rvhr wm/csf wm/csf+

global

RRF varies across the brain

-6s

-20s

#

-20 -10 0 10 20 30 40

RRF

GAM

shift (s)

Histogram of latencyGAM

time (s)0 10 20 30 40

RRF

a

b c

Optimal latency of Respiration Response Function

(RRF)

0

0.04

0.08

0.12

R2

Birn et al., 2008

Subject Specific physiological models

M. Falahpour et al., Neuroimage 2013

“FIX” (FMRIB’s ICA-based X-noisifier)

L. Griffanti, et al., Neuroimage 2014

ICA-AROMA

ICA-AROMA

R.H.R. Pruim et al., NeuroImage 112 (2015)

Should we perform physio corrections?

• Reduces fluctuations related to heart beat

and respiration

– Reduce false positives

– Reduce false negatives

• Some physiological fluctuations are

associated with neuronal activity

– E.g. Heart rate variability

This is still an open question in the field