Liver Electrical Impedance Tomography for Early Identification of Fatty Infiltrate in Obesity

Liver Electrical Impedance Tomography for Early Identification of Fatty Infiltrate

in Obesity

Chih

Chiang Chang

, Zi

Yu Huang

, Shu

Fu Shih

, Yuan Luo

, Arthur Ko

ingyu Cui

Susana

Cavallero

Swarna

Das

Gail Thames

, Alex Bui

, Jonathan

Jacobs

,7,

, P

ivi

Pajukanta

, Holden Wu

Chong

Tai

, Zhaoping Li

, and Tzung K. Hsiai

Department of Bioengineering, University of California

Los Angeles, Los Angeles, CA

Department of Medical Engineering,

California Institute of Technology, Pasadena, CA

Department of Radiological Sciences, David Geffen School of Medicine at UCLA, Los Angeles,

Department of Biomedical Engineering, Southern University of Science and Technology,

Shenzhen, Guangdong, China

Department of Medicine, David Geffen School of Medicine at UCLA, Los Angeles, CA

Center for Human Nutrition, David Geffen School of Medicine at UCLA, Los Angeles, CA

ivision

of Digestive Diseases, David Geffen School of Medicine at UCLA, Los Angeles,

Greater Los Angeles VA Healthcare System, Los Angeles, CA

Department of Human Genetics, David Geffen School of Medicine at UCLA, Los Angeles, CA

Institute for Precision Health, David Geffen School of Medicine at UCLA, Los Angeles, C

+Both authors contributed equally.

Corresponding author:

Tzung K. Hsiai, M.D., Ph.D.

Department of Medicine and Bioengineering, UCLA

Medical Engineering, Caltech

10833 Le Conte Ave., CHS17

054A

Los Angeles, CA 90095

1679

Email:

Thsiai@mednet.ucla.edu

Phone: 310

268

3839

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

Abstract

Non

lcoholic

fatty liver disease (NAF

D) is

ndemic

developed countries

and

one of the

most common causes of cardio

metabolic

diseases

overweight/

obese individuals

While liver

biopsy or magnetic resonance imaging (MRI)

he current gold

standard to diagnose NAFLD

the former is prone to bleeding

and the latter is costly

hereby demonstrated liver

lectrical

impedance tomography (EIT)

non

invasive and portable detection

method for

fatty infiltrate.

enrolled

subjects

(

female

and 4

males

;

27 to 74 year

old

)

undergo

liver

MRI scans

followed by

EIT measurement

via

multi

electrode array

The

liver

MRI

scans

provided subject

specific

a priori

knowledge

of the

liver boundary conditions for segmentation and

EIT

reconstruction

, and

the

D multi

echo

MRI data

quantif

ied

liver

proton

density fat fraction

(

PDFF

)

as a

recognized

reference

standard

for

validating

liver fat

infiltrate

Using

acquired

voltage data and

the reconstruction algorithm for

the

EIT imaging,

compute

the

absolute

conductivity distribution

abdomen

in 2

Correlation analyses

were performed to

compare the

individual EIT conductivity vs. MRI PDFF with their demogr

aphics in terms of gender, BMI (



), age (years), waist circumference (cm), height (cm), and weight (kg).

Our results indicate that

EIT conductivity



)

and

liver MRI for PDFF

were

not correlated with the demographics

whereas

the

decrease in

EIT conductivity

was correlated with the

increase in

I PDFF

(

0.69

= 0.003

)

Thus, EIT conductivity

holds promise for developing

non

invasive,

portable,

and quantitative

method to

detect

fatty liver disease

Key Words:

Fatty Liver, E

lectri

cal Impedance Tomography (EIT),

Magnetic Resonance Imaging

(

MRI

proton

density fat fraction

(PDFF)

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

Introduction

besity is the major risk

actor

ssociated

with

the development

of nonalcoholic fatty liver disease

(NAFLD)

, affecting more than a third of

American

. adults

, and

he prevalence of severe obesity

(BMI



35 kg



)

continu

ing

rise

nationwide

NAFLD

is now one of

the most common causes

cirrhosis

requiring

liver transplantation

in the Western world

2,3

clinical challenge in the

management of NAFLD resides in non

invasively detecting fatty

liver

(i.e.

simple

hepatic

stea

tosis) and monitoring disease progression to steatohepatitis (hepatic inflammation), fibrosis

(liver sca

ring), and ultimately cirrhosis

4,5

While liver biopsy remains the gold standard for

diagnosis of NAFLD, it carries

substantial risk of bleeding

and is confounded by sampling bias

and inter

observer variability

While liver

MRI

proton

density fat fraction (PDFF) is recognized as

the non

invasive reference standard for validating liver fat infiltrate

7,8

is costly for underserved

communities

. Wh

ile

ultrasound is

non

invasive, it is

limited by spatial resolution and operator

dependency

9,10

Thus, there remains an unmet clinical need to develop a non

invasive and

economic method

that is operator

independent and portable for early detection of fatty

liver

disease.

this end

demonstrated

in prior study

the theoretical

and experimental

basis of electrical

impedance tomography (EIT)

for

measuring

liver fat content in the New Zealand White Ra

bit

model of atherosclerosis and fatty liver disease

By virtue of

tissue

specific

electrical

conductivity,

fatty infiltrate in

the

liver was characterized by its

frequency

dependent electrical

impedance

(Z)

response to applied alternating current (AC)

At low

frequencies

, the cell membranes

impede

the current flow, resulting in high conductivity, whereas at high frequency, they serve as the

mperfect capacitors

resulting in

tissue

and fluid

dependent impedance

. Thi

s impedimetric

property enables the development of

multi

electrode array to measure tissue

specific

conductivity, morphology, and changes in

volume in response to

changes in

cardiac

output

lung capacity

host of literature

has

demonstrated the application of EIT for functional

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

studies of the brain, cardiac

stroke volume

, and respiratory ventilation (

transthoracic impedance

pneumography

)

In this context

we reconstruct

liver frequency

dependent conductivity distribution with

the

multi

electrode array

acquired voltage data to demonstrate liver EIT

Applying the multi

electrode array, we performed EIT voltage measurements

by biasing elec

trical currents (at 2

mA and 50

250 kHz) to the upper abdomen

The currents penetrate

the body to varying depths

and the resulting boundary voltages

were

acquired

the electrodes.

In response to the applied

alternating

current

(AC)

, muscle and blood

are

more conductive than fat, bone, or lung tissue due

to the varying free ion content.

20,21

Fat

free tissue

such as skeletal muscle

carries high water

(~73%) and electro

lyte (ions and proteins)

conte

allow

ing

for efficient electrical conductivity



)

, whereas fat

infiltrated tissue

such as fatty liver (steatosis)

is anhydrous

, resulting in a

reduction in

conductivity

This impedimetric property

provides the theoretical principle to apply

the portable liver

EIT for

the

early identification of fatty liver

infiltrate with

translational implications

for

the

prevention of

liver fibrosis and major adverse coronary events (MACE).

Unlike EIT for

cardiopulmonary function

focusing on the differential

conductivity

addressed the

ill

posed inverse problem to demonstrate the absolute liver conductivity in 2

e recruited

overweight/

obese

subject

s to undergo liver 3T MRI scans, fol

lowed by

voltage

measurements

via the

flexible multi

electrode array (

Swisstom AG, Switzerland)

for EIT

MRI

images

were

acquire

to provide subject

specific

a priori

knowledge

the

liver geometry for

performing liver

segmentation and

position

ing

solv

the

inverse problem

EIT

reconstruction

ther

ompared

the

subject

specific EIT conductivity

with

the

liver

MRI

proton

density fat

fraction

PDFF

as a reference standard

for validating

fatty liver

infil

rate

Next, w

e performed

the

rson’s c

orrelation analyses

between the

EIT

liver

conductivity

and

demographic parameters,

and

also

performed the correlation analyses between

MRI PDFF

and

demographics

parameters

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

in terms of gender, BMI (



ge (years), waist circumference (cm), height (cm), and weight

(kg)

Following Bonferroni correction for multi

testing,

correlation analyses revealed that

liver EIT

conductivity



)

and

MRI PDFF w

ere

not correlated with these demographics

;

however,

the

EIT

liver conductivity

map

was negatively correlated with

the

MRI PDFF. This inverse correlation

between

the

EIT

liver

conductivity and MRI PDFF

holds promises for developing non

invasive

and portable liver EIT for early detection of silent

fatty

liver content

in the

healthy

overweight/

obese individuals

Results

Schematic

work

flow illustrate

the steps to compare and validate the EIT reconstruction

with MRI

The

recruitment of subjects followed the guidelines of the Human Subjects Protection

Committee of UCLA, described in the method section. For the workflow and schematic setup

(

Fig 1

), each s

ubject would undergo a different series of MRI scans, including a 30

min MRI

multi

echo imaging to acquire the PDFF map of the liver. Next, EIT measurement with 32

electrodes attached to the subject’s abdominal region was performed right after the MRI scan

obtain the corresponding liver fat measurement. The average PDFF of the liver, as well as the

EIT conductivity of the liver, were then quantified for validation and comparison.

Comparison between

MRI multi

echo imaging and EIT images

Liver MRI images provide

a priori

geometric information to reconstruct 2D EIT images. This

information includes the boundary conditions for

1) the abdominal cross

section, 2) the

peripheral tiss

s consisting of the skin, subcutaneous fat, and the rib

, a

nd 3) the

liver

in the

abdomen. With this information,

liver

EIT

inverse problem

was solved as described in

the

ethods

section

For each subject, EIT liver conductivity and MRI PDFF were compared with

the corresponding BMI value

(

Table

)

. Also, the

representative

abdomen

MRI i

ges

for

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

anatomy and

PDFF, liver segmentation (annotation), and

liver

EIT conductivity

maps



)

were

compared (

Fig.

). We observed that

the MRI PDF

and

EIT

liver

conductivity w

ere

not

correlated

with the magnitude of BMI.

Despite

a negative correlation with EIT

the

MRI PDFF for

Subject 17 with

a relatively lower BMI (BMI = 27.1 Kg



, PDFF =

6.2%, EIT = 0.3243 S



)

was higher than

that of

Subject 3 with a high BMI value (BMI = 39.0 Kg



, PDFF = 3.82%,

conductivity = 0.3296 S



We also noted

that MRI PDFF for

ubject

with

low BMI value

(BMI = 27.9 Kg



, MRI PDFF =

3.62

%, EIT =

0.3473



) was lower than

that of

ubject 10

with a high BMI values

(BMI = 34.3 Kg



, MRI PDFF = 16.44%, EIT = 0.3007 S



Despite

similar BMI

(27.1



vs. 27.9



the

percentages of

MRI PDFF

ubject 17

was

around two

times higher than that of Subject

11 (

6.20

vs.

3.62

Notably

, the MRI PDFF for

ubject 10 (BMI = 34.3 Kg



) was

more than

times

higher than that of

ubject 3 (BMI = 39.0



). These inconsistent

comparisons

suggest

that BMI

may

not be the ideal i

ndex for

predicting

the levels of fatty liver infiltrate

, and Pearson’s correlation analyses were performed in

the ensuing results

conductivity

vs. MRI PDFF

Using the MRI PDFF and EIT conductivity data from

Table 1

, we

performed the Pe

rson’s

correlation analyses and

identified

whether the magnitude of BMI correlates with the percentage

of MRI PDFF or EIT conductivity. We demonstrated that the correlation betw

een BMI and MRI

PDFF (

0.037,

=0.89

)

and

the correlation

between BMI and EIT (

)

were statistically insignificant (

Fig.

). However, the

confidence interval

plot

revealed

statistically significant

negative correlation between EIT and MRI PDFF

(

0.69,

0.003

)

(

Fig.

This

finding suggests

that

EIT conductivity

may be

used as an index for

non

invasive detec

tion

method

quantify human liver fatty

infiltrate.

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

Correlation analys

with the

demographic parameters, MRI PDFF, and EIT conductivity

demonstrate

liver

EIT for

identification

fatty liver

infiltrate

in the

enrolled

subjects

(BMI >

25)

, we performed correlation

analyses

with

demographic parameters including

age, waist

circumference, height, and weight

, respectively

(

Table

). We compared the correlation

coefficients between MRI PDFF and the demographic parameters (

Fig.

)

Following the

Bonferroni correction for multi

testing, the

relations

with

age (

0.13,

= 0.64

waist circumference (

0.23,

= 0.4

height (

0.59,

= 0.016

)

and weight

(

0.41,

= 0.12

)

were

statistically insignificant

. We further compared the correlation

coefficients between

liver

EIT and demographic parameters

(

Fig.

The correlation with

age (

0.1,

= 0.71

), waist circumference (

0.05,

= 0.8

height (

0.63,

0.0092

)

and

weight (

0.19,

= 0.47

) were statistically insignificant

Thus,

hese analyses corroborate that BMI and

other

demographic parameters were not

correlated

with liver fat infiltrate

our

overweight/

obese

subjects.

Discussion

invasive

and cost

effective

monitoring of fatty liver disease

remains

unmet clinical

need for

the early identification of cardiometabolic disorders

While

liver biopsy or magnetic

resonance imaging (MRI)

have been

performed to

detect

non

alcoholic fatty liver disease

(

NAFL

)

, the

risk of complications, sampling errors and cost

limit its clinical application

for

the

general population

. We hereby demonstrated the

development of

liver EIT as a

non

invasive and

portable

detection

method

for

quantify

ing

liver fat content

e recruited 19

overweight/

obese

adults

with

BMI > 25 Kg



to undergo liver MRI scan

e performed the

individual liver

EIT

measurements

with a multi

electrode array

, and we

used the anatomic

information to

address

the

inverse problem

for

reconstruct

ing

the

subject

specific

EIT

conductivity map

We performed

correlation analyses

liver EIT vs

MRI PDFF

in relation to the individual demographics

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

our best knowledge,

this is

the first

demonstration of

statistically significant negative correlation

etween

EIT

acquired

liver conductivity

and

MRI

quantified PDFF.

EIT has been applied

clinical medicine

ver the past two decades

iagnostic

EIT was

developed

for pulmonary

function

and

lung capacity

For instance

espiratory monitoring was

exhibit

ed by transthoracic impedance pneumography

26,27

, and

the

cardiac output (CO) and

stroke volume (SV) measurements

were

demonstrated via myocardial motion and blood

volume

,respectively

28,29

EIT

has

also

been

applied for

assess

ing

conductivity in breast tissue

and brain

Using the

multi

electrode configuration, we

obtain

voltage from the surface of the

abdomen by injection of AC current, to reconstruct

the EIT

conductivity

map

inside the liver

While EIT has been extensively studied

the nonlinear forward and inverse models for

reconstructing

the

EIT

conductivity map

remain a computational challenge. The ill

posed inverse

problem

introduces issues of existence, uniqueness, and instability

of the solution

he non

linear inverse model for EIT reconstruction requires

a priori

knowl

edge

of the

anatomic

boundaries to enhance the spatial resolution for

establishing

the absolute conductivity value

improv

the EIT reconstruction,

investigators

integrated EIT

with other imaging

modalities

including

registration with

MRI

introduction of

ultrasonic vibration

the

target tissue in the presence of

the

magnetic field.

integration

ould generate inductive

currents within the

liver

result

ing

in higher spatial resolution

, thus

obviating the need for

a priori

knowledge of

the

object

geometry and location

needed

for EIT

reconstruction

Researchers have also applied o

ther approaches

enhanc

the algorithm

for

solv

ing

the

ill

posed inverse

problem.

or instance,

particle swarm optimization (PSO)

was applied

via

paradigm shift from the conventional Gauss

Newton methods for fast convergence and high

spatial resolution

to solve

EIT

43,44

ecent studies have applied machine learning

, including

Convolut

ional

Neural Networks

to solve the non

linear ill

posed inverse problem for accurate

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

EIT reconstruction

45,46

. Hamilton

et al

obtained

absolute EIT images by combining the D

bar

method

with subsequent processing

using

onvolutional

eural

etworks (CNN)

technique

for

sharpening EIT reconstruction

et al

utilized

eep

eural

etworks (DNN) to directly obtain

a nonlinear r

elationship between the one

dimensional boundary voltage and the internal

conductivity

Experimentally

the accuracy

of EIT reconstruction

may

be further

enhanced

increasing

the electrode arrays at multiple level

around

the abdomen. This multi

level

configuration w

ould be able

inject

the currents

to and

record the voltages

from

the

entire

liver

As a result, 3

reconstruct

ion of a

liver conductivity distributio

would be

realized

As a corollary, w

e compared the liver anatomy with MRI PDFF distributions from a

representative 3

D rendering

(

Fig S

)

. The 3

EIT conductivity map was reconstructed

with the aid of

the MRI multi

echo sequence

a priori

knowledge

(

Fig S

). The high

fat

region in the MRI PDFF map (red dashed box) was

also

detected by the EIT

demonstrating

lower

conductivity. The 3

D MRI and EIT analyses further

support

the negative correlation

betw

een MRI PDFF and EIT

liver

conductivity.

he 3

D EIT conductivity map

reveals

the

inhomogeneous fat distribution as

supported by

the

MRI 3

D rendering image

(

Fig S

)

. With

additional

scanning along the z

direction, a precise mapping could be reconstructed

unveil

the

detail

the heterogeneous fat distribution.

While

MRI images

provide

the

a priori

knowledge

to solve the ill

posed inverse problem for

EIT reconstruction,

alternative methods to provide

such information

would allow for

low

cost

liver EIT screening for the general population

A strain

‐

displacement conversion method for

reconstructing the deformed shape of the object with the boundary conditions was proposed

Luo

et al

This method would potentially p

rovide the outer abdomen boundary information

embedding the positional sensors

the

EIT sensor belt.

Another method is to apply

differential

EIT

which

has the potential to identify the

tissue

specific conductivity

f the two frequencies are

correctly selected

, it is possible

to differentiate the fatty tissue from

the

non

fatty tissues or

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint

organs

by virtue of

fat

tissue

specific

electrical properties

are distinguished from

other

tissues

organs

(

Table S1

)

48,49

thus providing the peripheral layer boundaries information

Furthermore, u

sing

large number of MRI image database, we may correlate the

liver

and

peripheral layer boundar

ies with the waist circumferences. This correlation would

provide

calibration curve

between

the boundary conditions

and

demographic parameter. Thus, the

foremen

tioned

method

would be the future area of research to obviate the need for the

MRI

acquired

a priori

knowledge to improve

EIT reconstruction

To assess whether the preexisting medical conditions would influence the absolute EIT

conductivity, we

included the two subjects with electrolytes abnormities (n=18)

We observed

that the correlation value decreased from

R =

0.69 (

= 0,003, n=16) to R=

0.21 (

0.4

, n=18

(

Fig S2A

)

. We further excluded 2 subjects with anemia, and we noted that the correlation

improved from

R =

0.69 (

= 0,003, n=16) to R =

0.7

0 (

0.0049

, n=14)

(

Fig S2B

)

. These

results were consistent with the

impedimetric property underlying

the composition of

the liver in

the setting of pre

existing conditions

associated electrolyte disturbance. In this case, the

presence of leukemia, renal failure, and anemia disrupted the homeostasis of the organ

systems; thus, altering the liver conductivity. Thus, our funda

mental and experimental analyses

pave the way for defining our exclusion criteria for future subject enrolment.

In summary, we enrolled

overweight/

obese subjects to undergo MRI scan

and liver EIT

easurement

to generate

the

EIT conductivity map. We

demonstrate

that the increase in liver

EIT conductivity is correlated with

decrease in MRI PDFF. As a corollary, we

demonstrated

that the 3

D EIT conductivity map also revealed the heterogeneous distribution of fatty gradient

as evidenced by the 3

D MRI

PDFF. Our correlation analyses support

that subject

specific

EIT offers a non

invasive and portable method for

the

detection of hepatic

fat

infiltrate

;

thereby,

proving

translational

basis

for

developing liver EIT suitable for operator

independent

cost

identification and monitoring of fatty liver

disease.

The copyright holder for this preprint

this version posted December 22, 2020.

;

https://doi.org/10.1101/2020.12.21.423854

doi:

bioRxiv preprint