1
The 47 h Con e ence o The Canadian Medical and Biological Enginee ing Socie y - La Socié é Canadienne de Génie Biomédical
The 26 h Con e ence o he A lan ic Canada Clinical Enginee ing Socie y
Enhanced Image P ocessing o Implan ed Hyd ogel Sca old Images Using
P opaga ion-Based Imaging Compu ed Tomog aphy
Xiao Fan Ding 1, Zah a Khoz 2, Daniel Chen 1,2,3, Ning Zhu 1,3,4
1 Di ision o Biomedical Enginee ing, Uni e si y o Saska chewan, Saska oon, Canada
2 Depa men o Mechanical Enginee ing, Uni e si y o Saska chewan, Saska oon, Canada
3 Depa men o Chemical and Biological Enginee ing, Uni e si y o Saska chewan, Saska oon, Canada
4 Canadian Ligh Sou ce Inc., Saska oon, Canada
Abs ac — This s udy showed how e ec i e masking o
dense componen s (i.e., bone) in p opaga ion-based imaging
compu ed omog aphy (PBI-CT) scans o biological samples can
enhance he ou comes o deep lea ning denoising echniques.
This was pe o med on ex i o scans o hyd ogel sca olds im-
plan ed in o animal hind limb and supp essing he o e whelm-
ing signal om he bone allowed o clea e and mo e dis inc
isualiza ion o hyd ogel sca olds. This p o ed essen ial o ob-
se ing he in e ac ions o hyd ogel wi hin he physiological en-
i onmen . The de ailed image p ocessing s eps o e o imp o e
he p ac ical applica ion o PBI-CT in issue enginee ing and
egene a i e medicine esea ch.
Keywo ds— Synch o on Imaging, Image P ocessing, Deep
Lea ning, Hyd ogel Sca olds, Ex Vi o
INTRODUCTION
Synch o on adia ion-based p opaga ion based compu ed
omog aphy (PBI-CT) is a phase con as imaging me hod
ha has been shown o be an excellen me hod o s udying
low-densi y hyd ogel sca olds in i o [1]. Howe e , in
complex biological images, i becomes di icul o make ac-
cu a e obse a ions [2].
Fi s ly, he e is an assump ion o homogenei y in he sam-
ple in PBI-μCT which biological samples a e no . The phase
con as o dense componen s (i.e., bone) becomes he domi-
nan signal and obscu es he su ounding issues. Secondly,
he o e whelming signal om he bone also skews he his o-
g am o he images, which poses a challenge when a emp -
ing o use deep lea ning algo i hms asks such as blind de-
noising such as Noise2Noise [5] and Noise2In e se [6].
These algo i hms ely on a balanced his og am o unc ion
op imally, and dis o ions p e en accu a ely deep lea ning o
spa ial ea u es.
Thus, his s udy aimed o de elop he image p ocessing
s eps o mi iga e he limi a ions posed by s ong signal om
bone in PBI-CT o low-densi y hyd ogel sca olds ex i o.
MATERIALS AND METHODS
The image p ocessing s eps aken o supp ess signal om
he bone a e shown in Figu e 1. Fi s , a aw PBI-CT da ase
was econs uc ed using il e ed back p ojec ion (FBP) o wo
sepa a e da ase s: (1) wi h phase con as o imp o e he im-
age con as o he bone o c ea e a bone mask because he
bone con as is o e whelming g ea e han he su ounding
issues, and (2) wi hou phase con as o ha e a measu e o
he g ey alue and noise o he su ounding issues. Second,
his bone mask c ea ed om da a (1) was used o eplace he
bone in da a (2) wi h alue and noise om he measu ed is-
sue. The bone-masked esul images a e hen enhanced
h ough deep-lea ning based blind denoising using
Noise2Noise o Noise2In e se.
2
The 47 h Con e ence o The Canadian Medical and Biological Enginee ing Socie y - La Socié é Canadienne de Génie Biomédical
The 26 h Con e ence o he A lan ic Canada Clinical Enginee ing Socie y
Figu e 1: Image p ocessing s eps aken o mask he
o e whelming signal om he bone s a ing om aw PBI-
CT da ase . The bone-masked esul s a e hen enhanced by
deep-lea ning based blind denoising.
The signal- o-noise a io (SNR) and con as - o-noise a io
(CNR) we e used o assess he image esul s wi h and wi hou
bone masking. In addi ion, he pe cep ion-based image qual-
i y e alua o (PIQE), a no- e e ence image quali y sco e used
o image quali y assessmen pa o MATLAB Image P o-
cessing Toolbox was used. PIQE is use ul o image assess-
men when he e a e no a p io i e e ences such as he de-
noising esul s.
Image acquisi ion ook used synch o on adia ion-based
p opaga ion-based imaging mic ocompu ed omog aphy
(SR-PBI-µCT) pe o med a he 05ID-2 beamline o Cana-
dian Ligh Sou ce (CLS) using a 30 keV monoch oma ic
beam, a 1.5 m sample- o-de ec o dis ance, a Hamama su
AA-60 beam moni o , a 500 µm LuAG scin illa o , and a Ha-
mama su O ca Flash 4 came a wi h an e ec i e pixel size o
13 µm. CT econs uc ion used he UFO oolki [3] wi h he
anspo o in ensi y equa ion phase e ie al algo i hm [4].
RESULTS AND DISCUSSION
Figu e 2 shows he esul s om di e en image p ocessing
app oaches as well as a g ey alue p o ile ac oss he hyd ogel
ma e ial. The bone masking g ea ly enhanced he denoising
capabili y o Noise2Noise, which was shown o be g ea e
han Noise2In e se. This was shown no only in he images
bu also in he p o ile, which only he bone masking plus
Noise2Noise could e eal he cha ac e is ic edge enhance-
men o PBI-CT.
Figu e 2: Compa ison o di e en image p ocessing me hods
o isualizing hyd ogel sca old implan ed in o a hind
limb. The scale ba ep esen s 1 mm.
Quan i a i e image measu es a e shown in Table 1. Deep
lea ning denoising me hods, Noise2In e se and
Noise2Noise, a e capable o g ea ly imp o ing bo h he SNR
and CNR compa ed o FBP as a baseline. Especially wi h
Noise2Noise when he signal om he bone is supp essed.
The SNR inc eases d ama ically, indica ing he me hod e -
ec i ely educes noise and enhances he cla i y o he image
ea u es. Highe CNR alues sugges imp o ed isibili y o
meaning ul s uc u es agains he backg ound. Noise2Noise
wi hou bone p oduced he lowes PIQE sco e which shows
ha his esul is pe cei ed as ha ing be e highes quali y.
Table 1: Quan i a i e measu es o image quali y image
p ocessing me hods.
Figu e 3 shows he p og ession o a hyd ogel sca old im-
plan a ion o ne e egene a ion in a hind limb samples, as
obse ed nondes uc i ely h ough PBI-CT and p ocessed
h ough bone masking and Noise2Noise. These esul s show
changes in sca old s uc u e and densi y wi hin a p o ec i e
spi al shaped PCL s en o e wo days inside he animal
body. This longi udinal imaging p o ides insigh s in o he in-
e ac ions be ween he implan and su ounding physiologi-
cal en i onmen .
Figu e 3: PBI-µCT scans o a hindlimb samples con-
aining hyd ogel sca old implan a longi udinal ime
poin s. (a) Immedia ely a e implan a ion o se a baseline.
(b) 2 days a e implan a ion. Scale ba s ep esen 1 mm.
A baseline shown in Figu e 3 (a), he sca old emained
wi hin he PCL s en wi h dis inc s ands isible. Howe e ,
a e 2 days, he sca old has become swollen, squeezed be-
ween he PCL s en , and less dense as indica ed by a lowe
g ey alue shown in Figu e 3 (b). The dec ease in g ey alue
o e he 2 days e lec he e ol ing in e ac ion be ween he
sca olds wi hin he physiological en i onmen which could
no p e iously be obse ed due o he s ong signal om he
bone and noise.
FBP
Noise2In-
e se wi h
bone
Noise2In-
ese wi h-
ou bone
Noise2N
oise wi h
bone
Noise2No
ise wi h-
ou bone
SNR
29.21 ±
1.97
42.61 ±
0.25
71.33±
0.58
110.32 ±
31.22
173.06 ±
40.61
CNR
0.54 ±
0.16
0.85 ±
0.19
1.25±
0.29
1.13 ±
0.25
2.19 ±
0.60
PIQE
7.61 ±
0.69
7.96 ±
1.06
7.85 ±
0.78
6.47 ±
0.96
5.44 ±
0.41
3
The 47 h Con e ence o The Canadian Medical and Biological Enginee ing Socie y - La Socié é Canadienne de Génie Biomédical
The 26 h Con e ence o he A lan ic Canada Clinical Enginee ing Socie y
CONCLUSION
This s udy showed ha supp essing he signal om bone
in PBI-CT can enhance he esul s o deep lea ning denoising
echniques. Using SNR, CNR, and PIQE, denoising esul s
wi hou he bone consis en ly showed imp o ed esul s com-
pa ed o denoising esul s wi h he bone. This allowed clea e
isualiza ion o hyd ogel sca olds ex i o and c ucial o ob-
se ing he dynamic in e ac ions o hyd ogel sca olds wi hin
he physiological en i onmen . The e ined imaging ap-
p oach p omises o imp o e he p ac ical applica ion o PBI-
CT in issue enginee ing and egene a i e medicine esea ch
ACKNOWLEDGEMENTS
This wo k was suppo ed by he Uni e si y o Saska che-
wan G adua e Schola ship and NSERC INSPIRE G adua e
Fellowship, and he NSERC Disco e y g an . Pa o all o
he esea ch desc ibed in his pape was pe o med a he Ca-
nadian Ligh Sou ce, a na ional esea ch acili y o he Uni-
e si y o Saska chewan, which is suppo ed by he Canada
Founda ion o Inno a ion (CFI), he Na u al Sciences and
Enginee ing Resea ch Council (NSERC), he Na ional Re-
sea ch Council (NRC), he Canadian Ins i u es o Heal h Re-
sea ch (CIHR), he Go e nmen o Saska chewan, and he
Uni e si y o Saska chewan.
CONFLICT OF INTEREST
The au ho s decla e ha hey ha e no con lic o in e es .
REFERENCES
[1] X. Duan, N. Li, X. Chen, and N. Zhu, “Cha ac e iza ion o Tissue Sca -
olds Using Synch o on Radia ion Mic ocompu ed Tomog aphy Im-
aging,” Tissue Eng Pa C Me hods, ol. 27, no. 11, pp. 573–588, No .
2021, doi: 10.1089/ en. ec.2021.0155.
[2] L. Ning e al., “Nonin asi e Th ee-Dimensional in Si u and in Vi o
Cha ac e iza ion o Biop in ed Hyd ogel Sca olds Using he X- ay
P opaga ion-Based Imaging Technique,” ACS Appl Ma e In e aces,
ol. 13, no. 22, pp. 25611–25623, 2021, doi: 10.1021/acsami.1c02297.
[3] T. Fa agó e al., “To u : a as , e sa ile and use - iendly image p o-
cessing oolki o compu ed omog aphy,” J Synch o on Radia , ol.
29, no. 3, pp. 1–12, May 2022, doi: 10.1107/S160057752200282X.
[4] D. Paganin, S. C. Mayo, T. E. Gu eye , P. R. Mille , and S. W. Wilkins,
“Simul aneous phase and ampli ude ex ac ion om a single de ocused
image o a homogeneous objec ,” J Mic osc, ol. 206, no. 1, pp. 33–
40, 2002, doi: 10.1046/j.1365-2818.2002.01010.x.
[5] J. Leh inen e al., “Noise2Noise: Lea ning Image Res o a ion wi hou
Clean Da a,” Ma . 2018, [Online]. A ailable:
h p://a xi .o g/abs/1803.04189
[6] A. A. Hend iksen, D. M. Pel , and K. J. Ba enbu g, “Noise2In e se:
Sel -Supe ised Deep Con olu ional Denoising o Tomog aphy,”
IEEE T ans Compu Imaging, ol. 6, pp. 1320–1335, 2020, doi:
10.1109/TCI.2020.3019647.
