In e na ional Con e ence on Bio ab ica ion 2023, Sep embe 17 h -20 h 2023, Saska oon, Canada
Au oma ic 3D Segmen a ion o Hyd ogel Sca olds
Based on PBI-µCT
X. F. Ding1, X. Duan1, N. Li1, D. Chen1,2, and N. Zhu1,3,4
1 Di ision o Biomedical Enginee ing, Uni e si y o Saska chewan, Saska oon, SK, Canada
2 Depa men o Mechanical Enginee ing, Uni e si y o Saska chewan, Saska oon, SK, Canada
3 Canadian Ligh Sou ce Inc., Saska oon, SK, Canada
4 Depa men o Chemical and Biological Enginee ing, Uni e si y o Saska chewan, Saska oon, SK, Canada
Keywo ds — Hyd ogel sca old, Synch o on, Compu ed omog aphy, Segmen a ion
INTRODUCTION
Hyd ogel sca olds a e a p omising bioma e ial used in issue
enginee ing and egene a i e medicine. Sca olds can be
cons uc ed using 3D biop in ing echniques which allow o
in ica e a chi ec u es a he si e o inju y [1]. Howe e , 3D
segmen a ion o hyd ogel sca olds emains a challenge due o
he low-densi y o hyd ogels exhibi ing poo image con as .
One p omising me hod is o use synch o on adia ion (SR)
p opaga ion-based imaging (PBI) mic ocompu ed omog aphy
(µCT). This me hod shows he phase shi o X- ays be ween
sample and de ec o wi h s ong edge enhancemen .
Quan i a i ely, his phase shi can be calcula ed using phase
e ie al (PR) algo i hm which con e s he edge enhancemen
in o a ea whe e he X- ay p opaga ed h ough he sample [2]
bu will educe sha pness be ween ac oss bounda ies.
Al e na i ely, image denoising, e.g., Noise2In e se (N2I) [3],
can be used o maximize he edge enhancemen bu does no
ep esen he a ea. Thus, SR-PBI-µCT da a can be p ocessed
in complemen a y ways. A con olu ional neu al ne wo k
(CNN) can be ained o lea n he complemen a y a ea and
edge in o ma ion which allow o mo e accu a ely ep esen ed
esul han ei he me hod alone. This p o ides e icien 3D
segmen a ion wi hou manual inpu o p e-exis ing e e ence.
MATERIALS AND METHODS
The 3D p in ed hyd ogel sca olds we e composed o 3% w/
algina e. Sca olds we e scanned in i o using SR-PBI-µCT
pe o med a he 05ID-2 beamline a he Canadian Ligh
Sou ce (CLS). The X- ay ene gy was 30 keV, he sample- o-
de ec o dis ance was 1.5 m, and he e ec i e pixel size was
13 μm. Combining he a ea in o ma ion om PR and edge
in o ma ion om N2I was accomplished by aining a CNN in
a U-Ne a chi ec u e. The combined ou pu was au oma ically
segmen ed by applying a g ey alue h eshold. This esul was
compa ed o a popula semi-au oma ic segmen a ion me hod,
Biomedisa, as well as a ully manual segmen a ion conside ed
o be g ound u h by calcula ing he Dice simila i y
coe icien .
RESULTS AND DISCUSSION
A ypical hyd ogel sca old is shown in Figu e 1 which has
been p ocessed using phase e ie al, noise2in e se, and he
p oposed combined me hod. The signal- o-noise a io (SNR)
and con as - o-noise a io (CNR) we e calcula ed o each
image as a quan i a i e measu e o image quali y wi h a g ea e
alue mo e ad an ageous o segmen a ion. No only does he
combined me hod o e he g ea es image con as , bu he los
edge om PR has been eco e ed which allows o
segmen a ion shown in a ep esen a i e slice. The p oposed
me hod demons a es a 6% inc ease in Dice coe icien o he
g ound u h o e Biomedisa. The pink/g een egions ep esen
he o e /unde es ima ions, espec i ely, by each me hod.
Figu e 1. SR-PBI-µCT da a p ocessed using (a) phase e ie al, (b)
noise2in e se, and (c) he p oposed combined me hod. The scale ba
ep esen s 1 mm. Manual (d), semi-au oma ic (e), and au oma ic ( )
segmen a ions wi h Dice coe icien s.
CONCLUSIONS
The p oposed segmen a ion me hod yields compa a i ely
accu a e esul s o exis ing me hods wi hou he ime and
a en ion equi ed o manual inpu .
REFERENCES
[1] Ning L. e al., ACS Appl. Ma e . In e aces. 13, 25611-
25623, 2021
[2] Paganin D., e al., J. Mic osc. 205(P 1):33-40, 2002
[3] Hend iksen A. A., Pel D. M., Ba enbu g K., J. IEEE T ans
Compu . Imaging. 6:1320-1335, 2020
ACKNOWLEDGEMENTS
This wo k was suppo ed by NSERC Disco e y g an s. All o
he esea ch desc ibed in his a icle was pe o med a he
Canadian Ligh Sou ce.
