BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
De ec ion and Segmen a ion o Re inal Lesions in
Re cam 3 Images Based on Ac i e Con ou s D i en
by S a is ical Local Fea u es
Jan KUBICEK1, Ju aj TIMKOVIC2, Ma ek PENHAKER1, Da id OCZKA1,
Ve onika KOVAROVA1, Alice KRESTANOVA1, Ma in AUGUSTYNEK1, Ma in CERNY1
1Depa men o Cybe ne ic and Biomedical Enginee ing, Facul y o Elec ical Enginee ing and Compu e
Science, VSB–Technical Uni e si y o Os a a, 17. lis opadu 15/2172, 708 33 Os a a-Po uba, Czech Republic
2Clinic o Oph halmology, Uni e si y Hospi al Os a a, 17. lis opadu 1790,
708 52 Os a a-Po uba, Czech Republic
[email p o ec ed], imko ic.j@bluepoin .sk, ma ek.penhak[email p o ec ed], da id.oczk[email p o ec ed],
e onika.ko a o [email p o ec ed], alice.k es ano [email p o ec ed], [email p o ec ed], ma in.ce n[email p o ec ed]
DOI: 10.15598/aeee. 17i2.3045
Abs ac . Clinical e inal image analysis is an impo
aspec o clinical diagnosis in oph halmology. Re inopa-
hy o P ema u i y (ROP) ep esen s one o he mos
se e e e inal diso de s in p ema u ely bo n in an s.
One o he ROP clinical signs is he p esence o e inal
lesions endange ing he ision sys em. Un o una ely,
he s age and p og ess o hese indings is o en only
subjec i ely es ima ed. A p ocedu e such as his is un-
doub edly linked o subjec i e inaccu acies depending
on he expe ience o he oph halmologis . In ou s udy,
a ully au onomous segmen a ion algo i hm o model
e inal lesions ound using Re Cam 3 is p oposed. The
p oposed me hod used a combina ion o e inal image
p ep ocessing and ac i e con ou s o e inal lesion seg-
men a ion. Based on his p ocedu e, a bina y model o
e inal lesions ha allowed e inal lesions o be clas-
si ied om a e inal image backg ound was ob ained.
Ano he impo an aspec o he model was ea u e ex-
ac ion. These ea u es eliably and au oma ically de-
sc ibed he de elopmen s age o an indi idual lesion.
A complex p ocedu e such as his has signi ican impli-
ca ions o oph halmic clinical p ac ice in subs i u ing
manual clinical p ocedu es and imp o ing he accu acy
o ou ine clinical decisions.
Keywo ds
Ac i e con ou , bina y model, ea u e ex ac-
ion, image segmen a ion, Re Cam 3, e inal
lesions.
1. In oduc ion
Re inal image assessmen is a ou ine clinical p oce-
du e. Two medical imaging sys ems a e con en ionally
used in he oph halmology clinical p ac ice. Fundus
came as, which acqui e e inal images in high esolu-
ion, a e used o cap u e indi idual s uc u es in high
con as . The Re Cam 3 sys em is used o cap u e e i-
nal images wi h a esolu ion o 480×640 pixels, which is
subs an ially lowe han a undus came a, bu is, how-
e e , sui able o examina ions on p ema u ely bo n
child en. Ou esea ch exclusi ely belongs o his a ea
[1], [2], [3], [4] and [16].
When a e inal sys em is in es iga ed, se e al majo
s uc u es a e impo an o clinical diagnosis. The
cen e o he e ina con ains he op ic ne e (disc),
which is also a s a ing poin o e inal blood essels.
One o he mos con en ional pa hological indings is
he e inal lesion. Such lesions a e p ima ily a ci cula
shape and da ke in ensi y spec um (Fig. 1) [5], [6],
[7] and [17].
The in ensi y spec um o a e inal lesion is an im-
po an ea u e ha allows e inal lesions o be iden-
i ied and also a ec s he e ec i eness o segmen a-
ion. In some cases, he con as in indi idual e i-
nal lesions and o he e inal objec s is low. This low
con as is a majo eason why many segmen a ion
me hods ail, because i is nea ly impossible o ind
he edges o a e inal lesion [8], [9] and [10]. Ano he
signi ican p oblem conce ns he in ensi y spec um o
a e inal lesion. These spec a p ima ily o e lap wi h
e inal blood essels, and i is he e o e no possible
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 194
BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
o apply he mul i egional segmen a ion ha would
sepa a e e inal lesions om he su ounding e inal
s uc u es[11], [12], [13], [14], [18] and [19].
Fig. 1: A e inal image cap u ed wi h Re Cam 3 showing e inal
lesions, indica ed by he blue squa es.
2. Analysis o Pa ien Re inal
Reco ds
Fo he ask o segmen a ion and modeling e inal le-
sions, he Uni e si y Hospi al o Os a a supplied an
ex ensi e da ase o e inal images o es he p oposed
segmen a ion model. All da a was anonymized and had
a uni ied esolu ion o 480×640 pixels. The quali y o
a ce ain image may depend on a physician’s wo k wi h
he e inal p obe as well as o he ac o s, such as ob-
se able di e ences in he sha pness o e inal objec s
and o he image ea u es. The da ase con ained 2797
anonymous e inal image eco ds. Re inal lesions we e
p esen in 1015 o hese images. These images we e
used o es an algo i hm. An example o h ee e inal
images con aining e inal lesions is shown in Fig. 2.
Fig. 2: Example om he e inal da ase showing e inal le-
sions.
3. Re inal Image
P ep ocessing
In o de o enhance he ea u es o e inal lesions, im-
age p ep ocessing was applied in he algo i hm o sep-
a a e RGB channels, equalize his og ams, and apply
b igh ness ans o ma ion and median il e ing.
Fi s , he RGB image da a was o ganized in o in-
di idual in ensi y laye s in o de o highligh e i-
nal lesions in each channel wi h he g ea es con as
(Fig. 3). Empi ically, a combina ion o he g een and
ed channels seemed he op imal comp omise. I was
also appa en ha e inal lesions we e signi ican ly sup-
p essed in he blue channel.
Fig. 3: RGB e inal image sepa a ed in o ed (le ), g een (mid-
dle) and blue ( igh ) channels.
In he ollowing s ep, his og am equaliza ion was ap-
plied. This ope a ion pe o ms a homogenous dis i-
bu ion o in ensi y le els, which is a bene i in p o-
cessing o he low-con as da a. Ou p ocedu e ap-
plied he CLAHE me hod (Con as -Limi ed Adap i e
His og am Equaliza ion). A dis inc ad an age o his
me hod is equaliza ion in smalle , local image segmen s
compa ed o con en ional his og am equaliza ion. An
example using he CLAHE me hod is gi en in Fig. 4.
Fig. 4: Compa ison o he o iginal his og am ( op) and appli-
ca ion o he CLAHE me hod (bo om).
B igh ness ans o ma ion was hen applied. In he
p oposed algo i hm, b igh ness ans o ma ion boos s
he image con as so ha e inal lesions a e displayed
in high con as . An example o he b igh ness ans-
o ma ion is shown in Fig. 5.
In o de o p e en image noise, a median il e was
applied (Fig. 6 and Fig. 7) wi h a 2D ci cula con olu-
ion and 9×9 pixel con olu ion ke nel. Median il e ing
elimina ed high- equency noise, and e inal blood es-
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 195
BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
sels we e mos ly obscu ed om he image. This esul
imp o ed he segmen a ion e ec i eness, as he p es-
ence o blood essels can limi segmen a ion accu acy
because o hei iden ical in ensi y spec a o e inal
lesions.
Fig. 5: Compa ison o a e inal image be o e applying b igh -
ness ans o ma ion (le ) and a e b igh ness ans o -
ma ion ( igh ).
Fig. 6: A e inal image be o e applying median il a ion (le )
and a e median il a ion wi h a 9×9 ke nel ( igh ).
Fig. 7: Compa ison o he RoI (Region o In e es ) o he e inal
lesion wi hou il a ion (le ) and a e median il a ion
( igh ).
4. De o mable Segmen a ion
Model
In his sec ion, a segmen a ion model o e inal lesions
[15] is in oduced. The geome ical pa ame e s in his
model can be changed acco ding o a p ede ined num-
be o segmen a ion s eps (i e a ions).
We use he unc ional ene gy, ep esen ing he in ensi y
dis ibu ion o indi idual pixels:
ELGDF =RΩELGDF
xdx
RΩPN
i=1 −ω(x−y) log pi,x(I(y))dydx,
(1)
whe e pi,x(I(y)) ep esen s he p obabili y densi y
unc ion in he a ea Ωi,ω(x−y) ep esen s he weigh ed
unc ion and −log p ep esen s he ansi ion om min-
imiza ion o he desi ed maximiza ion.
I is assumed ha he image egion Ωcan be sep-
a a ed in o wo a eas: he image backg ound and he
o eg ound, in a o m o a ma hema ical bina y model.
Using he Hea iside unc ion H, he ene gy unc ion
may be exp essed:
ELGDF
x(φ, u1(x), u2(x), σ1(x)2, σ2(x)2
=−Rω(−y) log px,i(I(y)M1(φ(y))dy
−Rω(x−y) log px,i(I(y)M2(φ(y))dy,
(2)
whe e M1(φ(y))=H(φ(y)) and M2(φ(y))=1−H(φ(y)),
and he ene ge ic unc ion can be ew i en as:
ELGDF (φ, u1, u2, σ2
1, σ2
2)
=RΩELGDF
x(φ, u1(x), u2(x), σ2
1(x), σ2
2(x))dx,
(3)
The le el-se unc ion is egula ized in o de o achie e
an accu a e con ou e olu ion. This egula iza ion p e-
en s dis o ion o he ac i e con ou om he dis ance
gi en:
P(φ) = Z1
2(|Oφ(x)| −1)2dx. (4)
The le el-se unc ion egula ion is consequen ly done
by es ic ion o i s leng h:
L(φ) = Z|OH(φ(x)) |dx. (5)
The en i e ene ge ic po en ial may be ew i en as:
F(φ, u1, u2, σ2
1, σ2
2)
=ELGDF (φ, u1, u2, σ2
1, σ2
2) + υL(φ) + µP (φ),
(6)
whe e υ, µ > 0 ep esen he weigh ed cons an s. P ac-
ically, he Hea iside unc ion may be app oxima ed by
he smoo hing unc ion de ined as:
Hε=1
21 + 1
2 an−1x
ε.(7)
The de i a ion o such a unc ion is:
δε=H0
ε(x) = 1
π
ε
ε2+x2.(8)
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 196
BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
App oxima ion o he ene gy unc ional is gi en by:
F(φ, u1, u2, σ2
1, σ2
2)
=ELGDF (φ, u1, u2, σ2
1, σ2
2) + υLε(φ) + µP (φ).
(9)
Minimiza ion o he ene gy unc ional is done using he
g adien descen low equa ion:
∂φ
∂ =−δε(φ)(e1−e2) + δε(φ)di Oφ
|Oφ|
+µO2(φ)−di Oφ
|Oφ|,
(10)
whe e
e1(x) = ZΩ
ω(y−x)log(σ1(y)) + (u1(y)−I(x))2
2σ1(y)2dy
(11)
e2(x) = ZΩ
ω(y−x)log(σ2(y)) + (u1(y)−I(x))2
2σ2(y)2dy
(12)
The segmen a ion me hod is d i en by con olling
pa ame e s de e mining he low and smoo hness o he
segmen a ion p ocess:
•n- he numbe o i e a ions,
•∆ - he ime s ep o he cu e shi ,
•µ- a cons an ensu ing as minimal de ia ion,
•α- a cons an p o iding weigh o he image en-
e gy,
•υ- a cons an a ec ing he con ou leng h,
•ε- his alue gi es a wid h o he Di ac impulse,
ensu ing quicke mo emen o he ini ial cu e,
•σ- is a pa ame e o he Ke nel unc ion, pa ially
compensa ing o inhomogenei y o he in ensi y
image domain,
•λ1- de e mines he weigh o he a ea inside he
cu e du ing segmen a ion,
•λ2- de e mines he weigh o he a ea ou side he
cu e du ing segmen a ion.
5. Re inal Lesion Modeling
In his sec ion, segmen a ion esul s and building a seg-
men a ion model a e in oduced. A model o a e inal
lesion can be c ea ed om he ini ial cu e aced in-
side a lesion. Since e inal lesions a e ci cula in shape,
a ough app oxima ion may be done wi h an ini ial ci -
cle o smalle adius han he lesion.
A c i ical pa ame e o ac i e con ou e olu ion is
he numbe o i e a ions. I he numbe o i e a-
ions was unde es ima ed, he ac i e con ou would no
each he eal bo de s o he lesion. Howe e , i he
segmen a ion p ocess was o e es ima ed, he segmen-
a ion cu e would end o sp ead ou side he lesion
egion. We empi ically se he numbe o i e a ions a
300. Figu e 8 and Fig. 9 illus a e he e olu ion o he
cu e on he lesion’s RoI and he en i e image a ea,
espec i ely.
Fig. 8: E olu ion o he ac i e con ou wi h 300 i e a ions on
he lesion RoI.
Fig. 9: Mul iple de ec ion o h ee e inal lesions a e 300 i e -
a ions in he en i e e inal image a ea.
The second pa o he segmen a ion p ocess is bina-
iza ion. The a ea con aining lesions needs o be di -
e en ia ed om he image backg ound. The e olu ion
o an ac i e con ou is linked o he ene gy map. This
map is able o classi y he ene gy o he ac i e con ou .
The ene gy inside he con ou has nega i e ene gy,
while he a ea ou side has posi i e ene gy (Fig. 10).
F om he ene gy h eshold, a e inal lesion model can
be buil (Fig. 10 and Fig. 11).
The ollowing ou pu (Fig. 11) shows a si ua ion wi h
mul iple lesions p esen .
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 197
BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
Fig. 10: P ocess o building a e inal lesion model: he ac i e
con ou a e 300 i e a ions (le ), he ene gy map (cen-
e ), and he bina y model ( igh ).
Fig. 11: A p ep ocessed e inal image (le ) and a bina y ma h-
ema ical model o mul iple lesions ( igh ).
6. Tes s and Quan i a i e
Compa ison
As men ioned abo e, he segmen a ion p ocedu e de-
pends on se e al aspec s o de e mine he e ec i eness
and accu acy o segmen a ion.
The numbe o i e a ions gi es in o ma ion abou
how many imes he ac i e con ou can change i s
shape du ing he segmen a ion p ocess. Figu e 12
shows a compa ison o di e en i e a ion se ings. Fig-
u e 12(a) shows ha adjacen e inal blood essels may
cause signi ican p oblems when segmen ing e inal le-
sions. I a la ge numbe o i e a ions is selec ed, he
ac i e con ou ends o sp ead in o e inal blood es-
sels, as hei in ensi y spec a a e nea ly iden ical o
e inal lesions.
Fig. 12: Demons a ion o he e ec o i e a ions on con ou
modeling: (a) o e -segmen a ion – 400 i e a ions, (b)
unde -segmen a ion – 50 i e a ions, and (c) op imal
se ings – 300 i e a ions.
The second impo an pa ame e is he size o he
image ma ix linked o he image ea u es. We gene -
ally suppose ha when p ocessed e inal images ha e
a lowe esolu ion, he objec s o in e es lose hei con-
as , and he e ec o segmen a ion is he e o e wo se.
This si ua ion can be modeled by educing he image
ma ix, as shown in Fig. 13. When p ocessing educed
images, objec de ail is educed, and smoo hness o he
e inal lesion model is comp omised.
Fig. 13: Compa ison o a segmen a ion model wi h 300 i e -
a ions: na i e e inal image da a (le ) and educed
image ma ix, 50 % o o iginal size ( igh ).
As he second pa o he es s, a quan i a i e com-
pa ison was pe o med. The accu acy o he p o-
posed e inal lesion segmen a ion model was e alua ed
agains con en ional segmen a ion me hods (Tab. 1).
A sample o 50 e inal image eco ds was es ed in
which he RoIs o he e inal lesions we e ex ac ed.
The ollowing pa ame e s we e conside ed in he quan-
i a i e compa ison:
•Va ia ion o In o ma ion (VI ): his pa ame e
measu es he dis ance be ween wo segmen a ion
classes C1and C2in he sense o he a e age con-
di ional en opy, which is gi en by:
V I(C1, C2) = H(C1) + H(C2)−2I(C1, C2).(13)
•Rand Index (RI ): his pa ame e measu es he
le el o simila i y be ween he wo egions. RI
compa es he compa ibili y o assignmen be ween
wo pai s o elemen s in wo clus e s. The RI is
gi en by:
RI(C1, C2) = 2(n11 +n00)
N(N−1) ,(14)
whe e Ndeno es he o al numbe o pixels, n11 is
he numbe o pai s in he same clus e C1and C2
and n00 is he numbe o pai s assigned o di e en
clus e s. RI has alues o [0; 1], whe e 0 indica es
ha he clus e s a e comple ely dissimila , while
1 indica es ha he wo clus e s a e iden ical.
•Mean Squa ed E o (MSE): his is an es ima-
o measu ing he a e age o e o squa es be ween
wo segmen a ion esul s. MSE ep esen s a isk
unc ion ha co esponds o he expec ed alue o
a squa ed o quad a ic e o loss.
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 198

BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
•Dice Coe icien (DSC ): his pa ame e com-
pa es he simila i y be ween wo bina y egions.
When conside ing he bina y egions Xand Y,
whe e Xis he gold s anda d and Yis e inal le-
sion segmen a ion, he Dice Coe icien is gi en by:
DSC(X, Y ) = 2|X∩Y|
|X|+|Y|.(15)
The DSC coe icien is no malized in he ange
[0; 1], whe e 0 indica es no simila i y, and 1 indi-
ca es a ull ag eemen be ween he egions Xand
Y.
Tab. 1: A quan i a i e compa ison o he p oposed model wi h
al e na i e segmen a ion me hods.
P o-
posed
model
O su
Fuzzy
Th eshold-
ing
K-means
clus e -
ing
FCM
clus e -
ing
VI 3.125 3.988 4.887 5.112 3.668
RI 0.987 0.811 0.945 0.654 0.739
MSE 33.254 45.568 44.221 40.257 34.558
DSC 0.911 0.745 0.851 0.574 0.871
Lowe alues indica e a be e esul in VI and MSE.
Highe alues in he RI indica e a g ea e simila i y o
he gold s anda d, which means a be e esul . All
he compa isons we e made agains he gold s anda d,
which is a manual segmen a ion pe o med by a clini-
cal expe . Compa ed o he al e na i e me hods, he
p oposed i e a i ely de o mable segmen a ion model
achie ed be e esul s. This is also due o he cha ac-
e o segmen a ion. Al e na i e segmen a ion gene -
a ed mul i egional segmen a ion in which he in ensi y
spec a o e inal lesions pa ially o e lapped e inal
lesions. The e inal lesions we e hus o e -segmen ed
when e inal blood essels we e p esen .
Finally, complica ions ela ed o ime in he p o-
posed segmen a ion model we e examined. The seg-
men a ion algo i hm was as e ec i e as i was ime-
consuming . This complica ion was linked o se e al
phenomena. Fi s , he numbe o i e a ions was a sig-
ni ican issue, and gene ally we a emp ed o educe
he numbe o i e a ions in o de o educe compu -
ing ime. Howe e , ewe i e a ions could lead o im-
p ope segmen a ion esul s. Second, image esolu ion
a ec ed compu ing ime. The segmen a ion p ocedu e
wo ked mo e quickly on images wi h a lowe esolu ion.
The es ing p ocedu e was done o a selec ed numbe
o i e a ions: 50, 150, and 300. The image ma ix was
concu en ly educed o 75 %, 50 %, and 40 % o he
o iginal size. Table 2 and Tab. 3 show he a e age e-
sul s o 40 e inal images in which he mos signi ican
lesions we e selec ed o segmen a ion.
Tab. 2: Time equi ed (in seconds) o 40 e inal eco ds wi h
di e en segmen a ion se ings. Resul s a e a e aged.
Image esizing
Numbe o
i e a ions 100 % 75 % 50 % 40 %
50 4.32 s 3.55 s 3.12 s 2.49 s
150 12.25 s 9.44 s 8.44 s 7.63 s
300 15.95 s 11.52 s 9.25 s 8.74 s
Tab. 3: Time equi ed (in seconds) o 40 e inal eco ds wi h
di e en segmen a ion se ings. Resul s a e median.
Image esizing
Numbe o
i e a ions 100 % 75 % 50 % 40 %
50 4.12 s 3.41 s 3.74 s 2.66 s
150 12.65 s 9.84 s 9.21 s 8.44 s
300 15.41 s 12.12 s 9.47 s 8.98 s
7. Conclusion
Analysis o e inal lesions was p esen ed in his pape .
Modeling o e inal lesions has di ec implica ions in
he clinical p ac ice o oph halmology. E alua ing he
geome ical pa ame e s o e inal lesions in e inal im-
ages is a c i ical s ep in clinical diagnosis.
A complex segmen a ion me hod o modeling e i-
nal lesions was p oposed. Fi s , an image p ep ocess-
ing algo i hm was in oduced. The algo i hm enhanced
ea u es in a e inal image by boos ing he con as be-
ween indi idual e inal lesions and he e inal image
backg ound.
The main ea u e o he segmen a ion algo i hm was
he de o mable segmen a ion p ocedu e ha allowed
geome ical ea u es o be modi ied o e ime wi h
a p ede ined numbe o i e a ions. Expe imen ally, 300
i e a ions we e ound a good comp omise in e ms o
he algo i hm’s pe o mance and ime equi ed o com-
pu ing. In he inal s ep o he segmen a ion p oce-
du e, a bina y model o a e inal lesion was made. This
model was based on he classi ica ion o he ac i e con-
ou ’s ene gy map.
A limi a ion o he model was simul aneous segmen-
a ion o mul iple e inal lesions. All o hese lesions
would be classi ied in o a bina y model o he same
class. I was he e o e no possible o di e en ia e
be ween indi idual lesions in he segmen a ion model.
Fu u e wo k will in ol e de eloping an imp o ed model
o e ing a classi ica ion p ocedu e o indi idual e inal
lesions. This me hod would p o ide acking o dis-
c e e lesions. Using his p ocedu e, he clinical de el-
opmen o indi idual e inal lesions could be acked.
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 199
BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
Acknowledgmen
This wo k and con ibu ions we e suppo ed by he
p ojec SV4508811/2101 Biomedical Enginee ing Sys-
ems XIV. This s udy was also suppo ed by he
Czech Science Founda ion (GACR) esea ch p ojec
2017 No. 17-03037S In es men e alua ion o medical
de ice de elopmen un a he Facul y o In o ma ics
and Managemen , Uni e si y o H adec K alo e, Czech
Republic. This s udy was suppo ed by he Czech Sci-
ence Founda ion (GACR) esea ch p ojec ETA No.
TL01000302 Medical de ices de elopmen as an e i-
cien in es men o public and p i a e en i ies.
Re e ences
[1] KUBICEK, J., J. TIMKOVIC, M. PENHAKER,
M. AUGUSTYNEK, I. BRYJOVA and V. KASIK.
Ex ac ion o Op ical Disc Geome ical Pa am-
e e s wi h Using o Ac i e Snake Model wi h
G adien Di ec ional In o ma ion. In: In elli-
gen In o ma ion and Da abase Sys ems. Cham:
Sp inge , 2017, pp. 445–454. ISBN 978-3-319-
54429-8. DOI: 10.1007/978-3-319-54430-4_43.
[2] MORRISON, D., E. D. BOTHUN, G. YING,
E. DANIEL, A. BAUMRITTER and G. QUINN.
Impac o numbe and quali y o e inal images in
a elemedicine sc eening p og am o ROP: esul s
om he e-ROP s udy. Jou nal o Ame ican As-
socia ion o Pedia ic Oph halmology and S abis-
mus. 2016, ol. 20, iss. 6, pp. 481–485. ISSN 1091-
8531. DOI: 10.1016/j.jaapos.2016.08.004.
[3] BOWL, W., B. LORENZ, K. STIEGER,
S. SCHWEINFURTH, K. HOLVE, C. FRIED-
BURG and M. ANDRASSI-DARIDA. Co ela ion
o cen al isual unc ion and ROP isk ac o s in
p ema u es wi h and wi hou acu e ROP a he
age o 6-13 yea s: he Giessen long- e m ROP
s udy. B i ish Jou nal o Oph halmology. 2016,
ol. 100, iss. 9, pp. 1238–1244. ISSN 0007-1161.
DOI: 10.1136/bjoph halmol-2015-307855.
[4] KARP, K. A., A. BAUMRITTER and D. J.
PEARSON. T aining e inal image s o
e inopa hy o p ema u i y (ROP) sc eening.
Jou nal o Ame ican Associa ion o Pedi-
a ic Oph halmology and S abismus. 2016,
ol. 20, iss. 3, pp. 214–219. ISSN 1091-8531.
DOI: 10.1016/j.jaapos.2016.01.016.
[5] RANI, P. and E. R. RAJKUMAR. Classi ica-
ion o e inopa hy o p ema u i y using back
p opaga ion neu al ne wo k. In e na ional Jou -
nal o Biomedical Enginee ing and Technology.
2016, ol. 22, iss. 4, pp. 338–348. ISSN 1752-6418.
DOI: 10.1504/IJBET.2016.081221.
[6] SWEENEY, A. R., Q. ZHANG, R. K. WANG
and K. A. REZAEI. Op ical Cohe ence To-
mog aphy Mic oangiog aphy Imaging o Ci -
cumsc ibed Cho oidal Hemangioma. Oph halmic
Su ge y, Lase s and Imaging Re ina. 2018,
ol. 49, iss. 2, pp. 134–137. ISSN 2325-8160.
DOI: 10.3928/23258160-20180129-09.
[7] KAUR, J. and D. MITTAL. A gene alized me hod
o he segmen a ion o exuda es om pa hological
e inal undus images. Biocybe ne ics and Biomed-
ical Enginee ing. 2018, ol. 38, iss. 1, pp. 27–53.
ISSN 0208-5216. DOI: 10.1016/j.bbe.2017.10.003.
[8] TAN, J. H., H. FUJITA, S. SIVAPRASAD, S. V.
BHANDARY, A. K. RAO, K. C. CHUA and
U. R. ACHARYA. Au oma ed segmen a ion o ex-
uda es, haemo hages, mic oaneu ysms using sin-
gle con olu ional neu al ne wo k. In o ma ion Sci-
ences. 2017, ol. 420, iss. 1, pp. 66–76. ISSN 0020-
0255. DOI: 10.1016/j.ins.2017.08.050.
[9] SAMBATURU, B., B. SRINIVASAN,
S. M. PRABHU, K. T. RAJAMANI,
T. PALANISAMY, G. HARITZ and D. SINGH.
A no el deep lea ning based me hod o e inal
lesion de ec ion. In: In e na ional Con e -
ence on Ad ances in Compu ing, Commu-
nica ions and In o ma ics. Udupi: IEEE,
2017, pp. 33–37. ISBN 978-1-5090-6367-3.
DOI: 10.1109/ICACCI.2017.8125812.
[10] QU, M., C. NI, M. CHEN, L. ZHENG,
L. DAI, B. SHENG, P. LI and Q. WU. Au o-
ma ic diabe ic e inopa hy diagnosis using ad-
jus able oph halmoscope and mul i-scale line op-
e a o . Pe asi e and Mobile Compu ing. 2017,
ol. 41, iss. 1, pp. 490–503. ISSN 1574-1192.
DOI: 10.1016/j.pmcj.2017.04.003.
[11] KUBICEK, J., J. TIMKOVIC,
A. KRESTANOVA, M. AUGUSTYNEK,
M. PENHAKER and I. BRYJOVA. Mo pho-
logical segmen a ion o e inal blood essels
and consequen o uosi y ex ac ion. Jou nal
o Telecommunica ion, Elec onic and Compu e
Enginee ing. 2018, ol. 10, iss. 1, pp. 73–77.
ISSN 2180-1843.
[12] KUBICEK, J., J. TIMKOVIC, J. SLONKA,
M. PENHAKER, M. AUGUSTYNEK and
I. BRYJOVA. Op ical ne e segmen a ion using
he ac i e shape me hod. Leka a Technika. 2016,
ol. 46, iss. 1, pp. 13-20. ISSN 0301-5491.
[13] KUBICEK, J., J. TIMKOVIC, M. AU-
GUSTYNEK, M. PENHAKER and
M. POKRYVKOVA. Op ical Ne e Disc Seg-
men a ion Using Ci cual In eg o Di e encial
Ope a o . In: Ad anced Compu e and Com-
munica ion Enginee ing Technology. Cham:
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 200
BIOMEDICAL ENGINEERING VOLUME: 17 |NUMBER: 2 |2019 |JUNE
Sp inge , 2016, pp. 387–396. ISBN 978-3-319-
24582-9. DOI: 10.1007/978-3-319-24584-3_32.
[14] TIMKOVIC, J., J. NEMCANSKY, D. CHOLE-
VIK, P. MASEK, R. AUTRATA and I. KREJ-
CIROVA. A new modi ied echnique o he ea -
men o high- isk p e h eshold ROP unde he di-
ec isual con ol o Re Cam 3. Biomedical Pa-
pe s. 2015, ol. 159, iss. 3, pp. 413–416. ISSN 1213-
8118. DOI: 10.5507/bp.2015.027.
[15] WANG, L., L. HE, A. MISHRA and C. LI.
Ac i e con ou s d i en by local Gaussian dis-
ibu ion i ing ene gy. Signal P ocessing. 2009,
ol. 89, iss. 12, pp. 2435–2447. ISSN 0165-1684.
DOI: 10.1016/j.sigp o.2009.03.014.
[16] MARTINEK, R., R. KAHANKOVA, J. NE-
DOMA, M. FAJKUS and K. CHOLEVOVA. Fe-
al ECG P ep ocessing Using Wa ele T ans o m.
In: P oceedings o he 10 h In e na ional Con e -
ence on Compu e Modeling and Simula ion. New
Yo k: ACM P ess, 2018 pp. 39–43. ISBN 978-1-
450-36339-6. DOI: 10.1145/3177457.3177503.
[17] MARTINEK, R., G. RAZERA, R. KAHANKOVA
and J. ZIDEK. Op imiza ion o he T aining Sym-
bols o Minimum Mean Squa e E o Equal-
ize . In: Ad ances in In elligen Sys ems and
Compu ing. Cham: Sp inge , 2018, pp. 272–287.
ISBN 978-3-319-60833-4. DOI: 10.1007/978-3-319-
60834-1_28.
[18] MARTINEK, R., R. KAHANKOVA, J. NE-
DOMA, M. FAJKUS, H. NAZERAN and
J. NOWAKOVA. Adap i e Signal P ocessing o
Fe al PCG Reco ded by In e e ome ic Senso .
In: Ad ances in In elligen Sys ems and Com-
pu ing. Cham: Sp inge , 2018, pp. 235–243.
ISBN 978-3-319-68526-7. DOI: 10.1007/978-3-319-
68527-4_26.
[19] SCHMID, M., F. RIGANTI-FULGINEI,
I. BERNABUCCI, A. LAUDANI, D. BIBBO,
R. MUSCILLO, A. SALVINI and S. CON-
FORTO. SVM e sus MAP on Accele ome e
Da a o Dis inguish among Locomo o Ac i i ies
Execu ed a Di e en Speeds. Compu a ional
and Ma hema ical Me hods in Medicine. 2013,
ol. 2013, no. 343084, pp. 1–7. ISSN 1748-670X.
DOI: 10.1155/2013/343084.
Abou Au ho s
Jan KUBICEK was bo n in Os a a, Czech Repub-
lic. He ecei ed an M.Sc. in Biomedical Enginee ing
in 2012 and a Ph.D. in Technical Cybe ne ics special-
izing in Biomedical Enginee ing in 2018. His esea ch
in e es s include applied image and signal p ocessing
in medicine and s a is ical analysis o biomedical da a.
Ju aj TIMKOVIC M.D., Ph.D. is he head o
he Cen e o Child en wi h Visual Impai men ,
Uni e si y Hospi al o Os a a, Czech Republic and
a membe o he commi ee o he Czech Socie y o
Pedia ic Oph halmology and S abismus. He holds
a Ph.D. in Pedia ic Oph halmology om he Masa yk
Uni e si y in B no, Czech Republic. His esea ch and
publica ion in e es s include pedia ic oph halmology
and neu o-oph halmology.
Ma ek PENHAKER ecei ed his M.Sc. in
Measu emen and Con ol in 1996. He comple ed
a Ph.D. in 2000 a VSB–Technical Uni e si y o
Os a a in Technical Cybe ne ics. He has been
an Associa e P o esso in Biomedical Enginee ing
since 2016. His esea ch in e es s include biomedical
enginee ing, especially medical de ices and home
eleme y and signal p ocessing.
Da id OCZKA was bo n in Os a a, he Czech
Republic. He ecei ed an M.Sc. in Biomedical
Enginee ing in 2017. His esea ch ocuses on image
and signal p ocessing in medicine.
Ve onika KOVAROVA was bo n in T inec,
Czech Republic. She ecei ed a mas e ’s deg ee in
2018. He esea ch in e es s include applied image
p ocessing o e inal da a.
Alice KRESTANOVA was bo n in Opa a, Czech
Republic. She ecei ed a mas e ’s deg ee in 2018. He
esea ch in e es s include applied image p ocessing o
e inal da a.
Ma in AUGUSTYNEK was bo n in Os a a,
Czech Republic. He ecei ed habili a ion quali ica ions
in Technical Cybe ne ics in 2017. His esea ch ocuses
on special medical de ices and diagnos ic me hods.
Ma in CERNY ecei ed an M.Sc. specializ-
ing in Measu emen and Con ol in Biomedical
Enginee ing in 2005 and a Ph.D. in Technical Cybe -
ne ics specializing in Biomedical Enginee ing in 2012
om VSB–Technical Uni e si y o Os a a, Facul y o
Elec ical Enginee ing and Compu e Science, Czech
Republic. Today, he is an Associa e P o esso a
VSB–Technical Uni e si y o Os a a a he Depa -
men o Cybe ne ics and Biomedical Enginee ing. He
specialized in biomedical enginee ing, elemedicine,
ine ial senso s, and emo e home ca e sys ems. He
is esponsible o li ing labs ha ocus on mode n
echnologies o elde ly and disabled pe sons, which
is a p ojec d i en by VSB–Technical Uni e si y o
Os a a.
c
2019 ADVANCES IN ELECTRICAL AND ELECTRONIC ENGINEERING 201