Colon Cancer Disease Diagnosis Based on Convolutional Neural Network and Fishier Mantis Optimizer

Ci a ion: Mohamed, A.A.A.;
Hançe lio˘gulla i, A.; Rahebi, J.;
Rezaeizadeh, R.; Lopez-Guede, J.M.
Colon Cance Disease Diagnosis
Based on Con olu ional Neu al
Ne wo k and Fishie Man is
Op imize . Diagnos ics 2024,14, 1417.
h ps://doi.o g/10.3390/
diagnos ics14131417
Recei ed: 28 May 2024
Re ised: 25 June 2024
Accep ed: 27 June 2024
Published: 2 July 2024
Copy igh : © 2024 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license (h ps://
c ea i ecommons.o g/licenses/by/
4.0/).
diagnos ics
A icle
Colon Cance Disease Diagnosis Based on Con olu ional Neu al
Ne wo k and Fishie Man is Op imize
Amna Ali A. Mohamed 1, Aybaba Hançe lio˘gulla i 2, Ja ad Rahebi 3,* , Rez an Rezaeizadeh 4
and Jose Manuel Lopez-Guede 5,*
1Depa men o Ma e ial Science and Enginee ing, Uni e si y o Kas amonu, Kas amonu 37150, Tu key;
[email p o ec ed]
2Depa men o Physics, Uni e si y o Kas amonu, Kas amonu 37150, Tu key; [email p o ec ed]
3Depa men o So wa e Enginee ing, Is anbul Topkapi Uni e si y, Is anbul 34087, Tu key
4Depa men o Physics, Facul y o Science, Uni e si y o Guilan, Rash P.O. Box 41335-1914, Guilan, I an;
[email p o ec ed]
5Depa men o Sys ems and Au oma ic Con ol, Facul y o Enginee ing o Vi o ia-Gas eiz, Uni e si y o he
Basque Coun y (UPV/EHU), C/Nie es Cano 12, 01006 Vi o ia-Gas eiz, Spain
*Co espondence: [email p o ec ed] (J.R.); [email p o ec ed] (J.M.L.-G.)
Abs ac : Colon cance is a p e alen and po en ially a al disease ha demands ea ly and accu a e
diagnosis o e ec i e ea men . T adi ional diagnos ic app oaches o colon cance o en ace
limi a ions in accu acy and e iciency, leading o challenges in ea ly de ec ion and ea men . In
esponse o hese challenges, his pape in oduces an inno a i e me hod ha le e ages a i icial
in elligence, speci ically con olu ional neu al ne wo k (CNN) and Fishie Man is Op imize , o he
au oma ed de ec ion o colon cance . The u iliza ion o deep lea ning echniques, speci ically CNN,
enables he ex ac ion o in ica e ea u es om medical imaging da a, p o iding a obus and e icien
diagnos ic model. Addi ionally, he Fishie Man is Op imize , a bio-inspi ed op imiza ion algo i hm
inspi ed by he hun ing beha io o he man is sh imp, is employed o ine- une he pa ame e s o
he CNN, enhancing i s con e gence speed and pe o mance. This hyb id app oach aims o add ess
he limi a ions o adi ional diagnos ic me hods by le e aging he s eng hs o bo h deep lea ning
and na u e-inspi ed op imiza ion o enhance he accu acy and e ec i eness o colon cance diagnosis.
The p oposed me hod was e alua ed on a comp ehensi e da ase comp ising colon cance images,
and he esul s demons a e i s supe io i y o e adi ional diagnos ic app oaches. The CNN–Fishie
Man is Op imize model exhibi ed high sensi i i y, speci ici y, and o e all accu acy in dis inguishing
be ween cance and non-cance colon issues. The in eg a ion o bio-inspi ed op imiza ion algo i hms
wi h deep lea ning echniques no only con ibu es o he ad ancemen o compu e -aided diagnos ic
ools o colon cance bu also holds p omise o enhancing he ea ly de ec ion and diagnosis o
his disease, he eby acili a ing imely in e en ion and imp o ed pa ien p ognosis. Va ious CNN
designs, such as GoogLeNe and ResNe -50, we e employed o cap u e ea u es associa ed wi h colon
diseases. Howe e , inaccu acies we e in oduced in bo h ea u e ex ac ion and da a classi ica ion due
o he abundance o ea u es. To add ess his issue, ea u e educ ion echniques we e implemen ed
using Fishie Man is Op imize algo i hms, ou pe o ming al e na i e me hods such as Gene ic
Algo i hms and simula ed annealing. Encou aging esul s we e ob ained in he e alua ion o di e se
me ics, including sensi i i y, speci ici y, accu acy, and F1-Sco e, which we e ound o be 94.87%,
96.19%, 97.65%, and 96.76%, espec i ely.
Keywo ds: con olu ional neu al ne wo k; me aheu is ic me hods; FMO; Fishie Man is Op imize ;
colon cance
1. In oduc ion
In ecen yea s, colon cance has eme ged as a signi ican cause o mo ali y, a ec ing
millions o indi iduals wo ldwide [
1
–
3
]. Li es yle ac o s, aging, and gene ics a e known
Diagnos ics 2024,14, 1417. h ps://doi.o g/10.3390/diagnos ics14131417 h ps://www.mdpi.com/jou nal/diagnos ics
Diagnos ics 2024,14, 1417 2 o 16
o con ibu e o he de elopmen o colon cance , wi h esea ch es ablishing a clea link
be ween he consump ion o p ocessed mea s and alcohol and an inc eased isk o de el-
oping he disease [
4
,
5
]. Mo eo e , s udies indica e a highe p e alence o his disease in
de eloped coun ies, wi h app oxima ely 65% o cases diagnosed in hese egions [6].
Howe e , adi ional app oaches o colon cance diagnosis p esen signi ican chal-
lenges in e ms o ea ly iden i ica ion and accu a e diagnosis, making i a p e alen and
li e- h ea ening disease [
5
]. Howe e , adi ional app oaches o colon cance diagnosis
p esen signi ican challenges in e ms o ea ly iden i ica ion and accu a e diagnosis, mak-
ing i a p e alen and li e- h ea ening disease [
1
–
3
]. To add ess hese challenges and
enhance he accu acy and e ec i eness o colon cance diagnosis, ad anced echnologies
such as machine lea ning and deep lea ning ha e been explo ed in medical image analy-
sis [
7
–
9
]. A i icial in elligence o e s solu ions o he limi a ions o adi ional app oaches
by u ilizing echniques such as a i icial neu al ne wo ks (ANNs), suppo ec o machine
(SVM), uzzy me hods, expe sys ems, and me aheu is ic me hods o disease diagnosis
using medical images [10–17].
Addi ionally, inno a i e me hods o diagnosing and ca ego izing colon cance his opa ho-
logical images a e essen ial o imp o e he p ecision o diagnosis and ul ima ely imp o e pa ien
ou comes. This e ised in oduc ion p o ides a clea e and mo e s uc u ed o e iew o he
challenges in colon cance diagnosis and he ole o a i icial in elligence in add essing hese
challenges. I elimina es epe i ions and o ganizes he in o ma ion in a mo e ocused and
cohe en manne .
In gene al, p og ess in medical image analysis shows po en ial o ans o ming
ea ly de ec ion and diagnosis o colon cance , ul ima ely leading o imp o ed ou comes o
pa ien s. Resea che s a e s i ing o op imize he diagnos ic p ocess and enhance he o e all
managemen o his li e- h ea ening illness by le e aging s a e-o - he-a echnologies and
inno a i e algo i hms. Gi en he signi ican global bu den o colon cance , inno a i e
app oaches a e con inuously sough o enhance diagnos ic accu acy and ea men e icacy.
This esea ch add esses his need by ocusing on he ollowing key a eas:
•
in oduc ion o an inno a i e me hod o ca ego izing colon cance his opa hological
images wi hou ea u e selec ion using PCA;
•
u iliza ion o an in elligen ea u e selec ion me hod wi h FMO algo i hm o enhance
he p ecision o colon cance diagnosis;
•
in eg a ion o AI, deep lea ning, and bio-inspi ed op imiza ion algo i hms o im-
p o ed ea ly de ec ion and diagnosis o colon cance ;
•
ocus on s eamlining he de ec ion p ocess, imp o ing diagnos ic accu acy, and
ul ima ely enhancing pa ien ou comes;
•
po en ial e olu ioniza ion o cance diagnosis and ea men h ough cu ing-edge
echnologies in medical imaging analysis.
The s udy o ganiza ion p o ides a b ie ou line o he con en in each sec ion. Sec ion 2
e iews he li e a u e, Sec ion 3desc ibes he ma e ials and me hods, Sec ions 4and 5
p esen s he esul s and discussion, and Sec ion 6p o ides he conclusion. This o e iew
helps eade s unde s and he s uc u e o he s udy and an icipa e he con en o each
sec ion.
2. Li e a u e Re iew
In ecen yea s, colon cance has claimed nume ous li es, making i a signi ican
conce n wo ldwide [
18
]. P e en i e measu es, including a heal hy li es yle and egula
sc eening, a e essen ial o educing he isk o colon cance [
19
]. Diagnos ic imaging
plays a c ucial ole in iden i ying a ious diseases, including Alzheime ’s disease, Mul iple
Scle osis, and colo ec al ca cinoma [
20
]. Wi h i s li e- h ea ening na u e, colon cance
equi es ea ly de ec ion and accu a e diagnosis o e ec i e ea men [21].
Medical imaging modali ies such as CT scans and MRI echniques aid in diagnosing
colon cance by de ec ing abno mal cell g ow h in he colon [
22
]. Li es yle ac o s, aging,
and gene ics con ibu e o he de elopmen o colon cance , wi h p ocessed mea s and
Diagnos ics 2024,14, 1417 3 o 16
alcohol consump ion being associa ed wi h inc eased isk [
23
]. Sc eening me hods like
colonoscopy and his opa hological sc eenings a e i al o ea ly de ec ion and p e en-
ion [
24
]. Medical imaging, coupled wi h ad ancemen s in machine lea ning and deep
lea ning, shows p omise in imp o ing he accu acy o colon cance diagnosis [25].
Table 1summa izes a ious esea ch s udies aimed a imp o ing he diagnosis and
p ognosis o colo ec al cance h ough inno a i e me hodologies such as au oma ed al-
go i hms, con olu ional neu al ne wo ks (CNNs), and in eg a ion o medical da a wi h
a i icial in elligence (AI) echniques. Each s udy ou lines i s aims, ad an ages, disad an-
ages, and esul s, showcasing di e se app oaches o add essing he challenges in colo ec al
cance diagnosis and ea men .
Table 1. Ad ancemen s in colo ec al cance diagnosis and p ognosis: a compa a i e s udy o
inno a i e app oaches.
Aims Ad an ages Disad an ages Resul s
Re
1. De elop an au oma ed algo i hm o
de ec ing and ca ego izing hype plas ic and
adenoma ous colo ec al polyps du ing
colonoscopy.
2. U ilize ans e lea ning om la ge
nonmedical da ase s o enhance he
p ecision o polyp de ec ion and
classi ica ion.
3. In es iga e he capabili y o he p oposed
me hod o aid endoscopis s in p omp ly
esec ing adenoma ous polyps.
1. The algo i hm s eamlines he p ocess o
iden i ying and ca ego izing colo ec al
polyps, he eby educing he ime and
expenses associa ed wi h manual
examina ion.
2. Demons a ing high p ecision, ecall a e,
and accu acy compa ed o isual inspec ion
by endoscopis s, he algo i hm po en ially
mi iga es he isk o o e looking
adenoma ous polyps.
3.
D awing insigh s om nonmedical da ase s
en iches he algo i hm’s pe o mance,
unde sco ing he e icacy o ans e
lea ning in analyzing medical images.
4. By p ecisely iden i ying adenoma ous
polyps, he algo i hm acili a es p omp
esec ion be o e hey p og ess in o in asi e
cance , which could enhance
pa ien ou comes.
1. Despi e he e ec i eness o ans e
lea ning, he e may be a disconnec
be ween nonmedical sou ce asks and
medical a ge asks, necessi a ing u he
explo a ion o op imize ea u e selec ion.
2. Ga he ing and anno a ing a subs an ial
olume o medical da a o ine- une CNN
a chi ec u e may pose challenges, gi en he
limi ed a ailabili y and complexi y o
medical da ase s.
3. The p oposed me hod may demand
sophis ica ed compu a ional esou ces and
expe ise in deep lea ning, po en ially
limi ing i s applicabili y ac oss all
heal hca e se ings.
The p oposed algo i hm demons a ed compa able
p ecision bu supe io ecall a e and accu acy
compa ed o isual inspec ion by endoscopis s.
Ou pe o ming p io s a e-o - he-a me hods wi h
minimal p ep ocessing, he algo i hm p o ed
e ec i e in assis ing endoscopis s in iden i ying
o e looked adenoma ous polyps. These
encou aging ou comes sugges ha he p oposed
me hod holds p omise o enhancing he ea ly
de ec ion and diagnosis o colo ec al cance ,
ul ima ely leading o imp o ed pa ien ou comes.
[
26
]
1.
De elop a Dual-Pa h Con olu ional Neu al
Ne wo k (DP-CNN) o au oma ically de ec
in es inal polyps om colonoscopy images.
2. Valida e he e ec i eness o he p oposed
DP-CNN model in de ec ing polyps
h ough expe imen al esul s.
3. E alua e he pe o mance o he p oposed
me hod in e ms o ecall, p ecision,
F1-Sco e, F2-Sco e, and o e all accu acy
ac oss di e en da abases.
1.
The p oposed DP-CNN model exhibi s high
ecall, p ecision, F1-Sco e, and F2-Sco e in
iden i ying polyps in bo h he CVC
ColonDB and ETIS-La ib da abases.
2. The p oposed me hod o e s lowe
complexi y and ewe lea nable pa ame e s
compa ed o exis ing deep lea ning models,
making i sui able o eal- ime applica ions
and scena ios wi h limi ed compu ing
esou ces.
3. The DP-CNN a chi ec u e, coupled wi h a
sigmoid classi ie , e ec i ely iden i ies
polyps om colonoscopy images,
acili a ing ea ly diagnosis
and in e en ion.
1. The s udy ocuses speci ically on de ec ing
polyps om colonoscopy images and may
no be di ec ly ans e able o o he medical
image analysis asks.
2.
The pe o mance o he p oposed me hod is
con ingen on he quali y and di e si y o
he aining da ase s, po en ially es ic ing
i s applicabili y o di e en pa ien
demog aphics o imaging condi ions.
3. Fu he op imiza ion and ine- uning o
hype pa ame e s may be necessa y o
enhance he obus ness and gene alizabili y
o he p oposed me hod ac oss a ious
da ase s and clinical en i onmen s.
The DP-CNN model achie es high accu acy in
de ec ing polyps, wi h ecall a es o 99.20% and
92.85%, p ecision a es o 100% and 89.81%,
F1-Sco es o 99.60% and 91.00%, and F2-Sco es o
99.83% and 89.91% on he CVC ColonDB and
ETIS-La ib da abases, espec i ely. Compa a i e
analysis e eals supe io pe o mance compa ed o
exis ing me hods, demons a ing i s po en ial o
au oma ing polyp de ec ion and enhancing ea ly
colo ec al cance diagnosis.
[
27
]
1. To inno a e ensembles in eg a ing
con olu ional neu al ne wo ks (CNNs) and
ans o me s o seman ic segmen a ion.
2. To alida e he e icacy o me ging di e se
models, aining me hodologies, and
op imiza ion echniques o o ge mo e
po en ensembles.
3. To p opose a no el app oach o acqui ing
segmen a ion masks ia in e media e
p edic ion a e aging.
4. To ex end indings o di e se applica ion
domains and in es iga e s a egies o
adap ing he model o esou ce-cons ained
ha dwa e.
1. Ensembles combining CNNs and
ans o me s exhibi supe io polyp
segmen a ion compa ed o al e na i e
me hods.
2. In e media e p edic ion a e aging
diminishes o e i ing, bols e ing he
model’s esilience.
3. The p oposed me hodology displays
po en ial ac oss mul iple domains beyond
polyp segmen a ion.
4. Ongoing esea ch will explo e dis illa ion
me hods and p uning echniques o ailo
he model o low-cos ha dwa e.
1. Ensembles may necessi a e subs an ial
compu a ional esou ces o bo h aining
and in e ence.
2. Pe o mance could luc ua e based on
da ase cha ac e is ics used o aining and
e alua ion.
3. Ensembles pose challenges in disce ning
he con ibu ions o indi idual models o
he inal segmen a ion ou come.
The de ised ensembles excel ac oss i e majo polyp
segmen a ion da ase s, no ably ou pe o ming
leading me hods on wo da ase s wi hou speci ic
ine- uning. A no el s a egy o a e aging
in e media e p edic ions signi ican ly con ibu es o
mi iga ing o e i ing and e ining model
con ibu ions, unde sco ing i s pi o al ole in he
ensembles’ success
[
28
]
1. C ea e MEDomics, a dynamic
in as uc u e o o ganizing di e se heal h
da a and ensu ing da a quali y.
2. U ilize a i icial in elligence o p edic
indi idual p ognosis in oncology using
MEDomics.
3. Valida e he e ec i eness o he MEDomics
amewo k in oncology by iden i ying
co ela ions be ween clinical ac o s and
mo ali y.
4. U ilize na u al language p ocessing (NLP)
o con inuously upda e p ognosis es ima es
as disease condi ions e ol e.
1. MEDomics sys ema ically o ganizes heal h
da a and con inuously e alua es da a
quali y.
2. The amewo k iden i ies clinically
signi ican associa ions, such as he s ong
link be ween he F amingham isk sco e
and cance mo ali y.
3. Disco e ies like he F amingham isk
sco e’s impac on mo ali y can guide
clinical decisions, po en ially enhancing
pa ien ou comes.
4. NLP enables ongoing adjus men s o
p ognosis es ima es, enabling pe sonalized
and imely in e en ions.
1. Many hospi als may lack eadiness o
in eg a e da a science in o clinical
wo k lows, hinde ing he widesp ead
adop ion o sys ems like MEDomics.
2. Ensu ing da a accu acy and eliabili y
wi hin MEDomics equi es ongoing
a en ion and esou ce alloca ion.
3. The complexi y o AI algo i hms and NLP
echniques may impede unde s anding
among clinicians and heal hca e p o ide s.
4. De eloping and sus aining dynamic
in as uc u es like MEDomics en ails
subs an ial in es men s in pe sonnel,
echnology, and in as uc u e.
MEDomics p o es i s e icacy in oncology by
e ealing he s ong associa ion be ween he
F amingham isk sco e and cance mo ali y ac oss
di e en s ages. In eg a ion o NLP acili a es
con inual p ognosis upda es, adap ing o e ol ing
disease condi ions. This amewo k o e s a
p omising a enue o le e aging AI and di e se
heal h da a o enhance indi idual p ognosis and
guide clinical decision-making in oncology.
[
29
]
1. Demons a e he iabili y o in eg a ing
se um Raman spec oscopy wi h a
con olu ional neu al ne wo k (CNN)
model o he diagnosis o ou cance
ypes: gas ic, colon, ec al, and lung cance .
2. Assess he accu acy o his in eg a ed
app oach and isualize he CNN-ex ac ed
ea u es speci ically o ec al cance
diagnosis.
3.
Explo e he po en ial o using se um Raman
spec oscopy and CNN o di e en ia e
be ween cance and heal hy indi iduals,
wi h a pa icula ocus on ec al cance .
1. The amalgama ion o se um Raman
spec oscopy and CNN achie ed a no able
classi ica ion accu acy o 94.5%, showcasing
i s e ec i eness in diagnosing di e se
cance ypes.
2. The isualiza ion o CNN-ex ac ed
ea u es aids in deciphe ing chemical
composi ions, o e ing po en ial insigh s
in o cance de elopmen mechanisms.
3. Se um Raman spec oscopy p esen s a
cos -e ec i e, apid, and non-des uc i e
me hod o cance sc eening, po en ially
acili a ing p omp de ec ion and
in e en ion.
1. The opaque na u e o CNN models
impedes anspa ency in unde s anding he
lea ning p ocess, po en ially limi ing
in e p e abili y in he diagnos ic p ocess.
2. Despi e he high accu acy obse ed, he
p ecise mechanisms unde lying
biochemical subs ances in di e en cance
ypes emain incomple ely unde s ood,
necessi a ing u he esea ch
o cla i ica ion.
3.
While he s udy ocuses on diagnosing ou
speci ic cance ypes, he applicabili y o he
app oach o o he cance ypes may equi e
addi ional alida ion and
op imiza ion e o s.
The in eg a ion o se um Raman spec oscopy wi h a
CNN model achie ed a no able 94.5% accu acy in
diagnosing mul iple cance ypes. Visualiza ion o
CNN ea u es highligh ed signi ican di e ences
be ween cance and heal hy samples, indica ing
po en ial o non-in asi e cance sc eening and
wa an ing u he esea ch in o i s mechanisms and
applicabili y.
[
30
]
Diagnos ics 2024,14, 1417 4 o 16
Table 1. Con .
Aims Ad an ages Disad an ages Resul s
Re
1. De elop an au oma ed sys em employing
con olu ional neu al ne wo k (CNN) and
Ranking algo i hm o colo ec al cance
de ec ion, aiming o alle ia e pa hologis s’
wo kload and enhance diagnos ic accu acy.
2. Assess he easibili y and e icacy o
u ilizing deep lea ning echniques in
issue-based diagnos ics, u ilizing openly
a ailable digi al pa hology da ase s.
3. In es iga e he po en ial o in eg a ing
CNN and Long Sho -Te m Memo y
(LSTM) o op imize pe o mance and
e icacy in colo ec al cance diagnosis and
po en ially ex end applicabili y o di e se
cance ypes.
1. The p oposed model exhibi s supe io
p edic ion accu acy compa ed o exis ing
me hods, po en ially acili a ing ea ly
de ec ion and p e en ion o colon cance .
2. Se ing as a sc eening ool, he au oma ed
sys em has he po en ial o educe
pa hologis s’ wo kload and minimize
diagnos ic subjec i i y.
3. Inco po a ing CNN and LSTM enhances
pe o mance and expedi es diagnosis,
he eby imp o ing sys em e iciency.
1.
Manual ea u e selec ion om da ase s may
be labo ious and could po en ially impede
sys em e iciency, necessi a ing me iculous
conside a ion and po en ial op imiza ion.
2. The model’s ocus on de ec ing colo ec al
cance may cons ain i s u ili y o o he
cance ypes, wa an ing u he esea ch o
b oaden i s scope.
3. The model’s pe o mance may be
in luenced by he quali y and di e si y o
inpu da ase s, po en ially limi ing i s
gene alizabili y ac oss a ious popula ions
o imaging condi ions.
The p oposed model, employing CNN and Ranking
algo i hm, demons a es supe io pe o mance in
colo ec al cance diagnosis compa ed o exis ing
me hods, as indica ed by highe Recall, P ecision,
and Accu acy me ics. In eg a ion o CNN and
LSTM enhances he model’s e iciency and opens
a enues o po en ial expansion o iden i y a ious
cance ypes, p omising ad ancemen s in medical
image diagnosis amewo ks.
[
31
]
1. Explo e he spa ial dis ibu ion o T cell
subse s in he umo mic oen i onmen
among colon cance pa ien s.
2.
Es ablish connec ions be ween spa ial T cell
dis ibu ion and p e iously analyzed
genomic da a in he AC-ICAM colon cance
pa ien g oup.
3. In es iga e he po en ial p ognos ic
signi icance o T cell spa ial dis ibu ion
conce ning pa ien su i al and
Immunologic Cons an o Rejec ion (ICR)
ansc ip ome co ela ion.
1.
P o ides aluable insigh s in o he in ica e
in e play be ween immune cells and he
de elopmen o colon cance .
2. In eg a es spa ial T cell dis ibu ion da a
wi h genomic insigh s, enhancing
comp ehension o umo -immune
dynamics.
3. O e s po en ial p ognos ic implica ions by
associa ing T cell spa ial dis ibu ion wi h
pa ien su i al and ICR
ansc ip ome co ela ion.
1. Res ic ed o a speci ic coho o colon
cance pa ien s (n = 90), which could limi
applicabili y o b oade pa ien
popula ions.
2. Relies on a specialized mul iplex
immuno luo escence assay, po en ially
in oducing echnical limi a ions and
a iabili y.
3. Requi es addi ional alida ion and
eplica ion in la ge coho s o con i m he
p ognos ic ele ance o T cell spa ial
dis ibu ion in colon cance
Tumo -in il a ing T cell sub ypes showed
compa able densi ies, wi h p oli e a i e and
G anzyme B-exp essing T cells loca ed mainly
wi hin he umo epi helium. Immune-ac i e
sub ypes exhibi ed inc eased immune cell densi y
and educed dis ances be ween ce ain T cell
sub ypes and umo cells, co ela ing wi h imp o ed
su i al ou comes.
[
32
]
3. Ma e ial and Me hods
The p oposed me hod o diagnosing bo h cance and non-cance pa ien s is designed
o in eg a e se e al s eps o he analysis o his opa hological images. Ini ially, he me hod
in ol es ga he ing and p e-p ocessing sample images om a da ase o colon diseases. The
p e-p ocessing s ep includes noise elimina ion, adjus men , and image quali y enhancemen
echniques such as his og am balancing. The esea ch employs colo images wi h a ligh
in ensi y ange o 0 o 255 o each channel. The p oposed me hodology combines he
Fishie Man is Op imize (FMO) algo i hm wi h con olu ional neu al ne wo ks (CNNs)
o enhance he accu acy and eliabili y o colon cance diagnosis. This in eg a ion aims o
imp o e model pe o mance and in e p e abili y by op imizing ea u e selec ion du ing
CNN aining. The me hod u ilizes con olu ional neu al ne wo ks based on GoogleNe
and ResNe -50 o ea u e ex ac ion om his opa hological images. Tex u al ea u es o
he images a e ex ac ed using CNN me hods, and essen ial ea u es o machine lea ning
a e calcula ed. Fea u e selec ion is ep esen ed as a bina y challenge, wi h a ea u e
ec o o n dimensions indica ing he p esence o n ea u es, whe e each elemen is ei he
ze o o one. The FMO algo i hm is employed o ea u e selec ion due o i s abili y o
simula e lea ning beha io in con olu ional algo i hms, i s supe io accu acy compa ed
o o he me a-heu is ic op imiza ion me hods, and i s capabili y o pe o m global and
local sea ches op imally. The subsequen s age in ol es aining he machine lea ning
algo i hm using he op imal ea u e ec o o he classi ica ion o cance and non-cance
his opa hological images. The p oposed me hod also includes c ea ing an op imal ea u e
ec o and employing machine lea ning o dimensionali y educ ion and classi ica ion.
The inal s age in ol es e alua ing he p oposed me hod using es ing da a. The phases
o he p oposed me hod o dis inguishing be ween cance and non-cance colon images
include collec ing his opa hological samples ela ed o colon diseases.
■The samples a e p e-p ocessed o elimina e noise and enhance image quali y.
■Fea u e ex ac ion is pe o med using CNNs based on GoogleNe and ResNe -50.
■Tex u al ea u es o he images a e ex ac ed using CNN me hods.
■Essen ial ea u es o machine lea ning a e calcula ed om he images.
■
Fea u e selec ion is ep esen ed as a bina y challenge wi h a ea u e ec o o n dimensions.
■
The FMO algo i hm is u ilized o ea u e selec ion due o i s supe io accu acy and
op imiza ion capabili ies.
■
The machine lea ning algo i hm is ained using he op imal ea u e ec o o classi ica ion.
■
Machine lea ning is employed o dimensionali y educ ion and classi ica ion o he images.
■
The p oposed me hod is e alua ed using es ing da a o assess i s pe o mance in
dis inguishing be ween cance and non-cance colon images. O e all, he p oposed
me hod in eg a es adi ional da a p ep ocessing echniques wi h inno a i e ea u e
Diagnos ics 2024,14, 1417 5 o 16
selec ion using he FMO algo i hm and CNN aining o enhance he accu acy and
eliabili y o colon cance diagnosis. The p oposed me hodology’s amewo k o he
diagnosis o pa ien s wi h colon cance is shown in Figu e 1. The isual ep esen a ion
o he model concep ual amewo k in Figu e 2illus a es he seamless in eg a ion o
hese componen s, highligh ing he no el app oach aken in his esea ch.
Diagnos ics 2024, 14, x FOR PEER REVIEW 7 o 18
Figu e 1. The p oposed me hodology’s amewo k o he diagnosis o pa ien s wi h colon cance .
P e-p ocessing
his opa hological images
om noise cancella ion
wi h il e ing me hods
Imp o e his opa hological da ase
images quali y by balancing he
his og am and smoo hing he image
size
A ea u e ec o is used o
ain he machine lea ning
Reduce sample size and machine
lea ning inpu
Fea u e ex ac ion by
CNN based on GoogleNe and
ResNe
-50 models
Mul iple ea u e ec o s a e c ea ed as
membe s wi hin he andom FMO
algo i hm, and hese ea u e ec o s a e
using o de e mine machine lea ning
inpu .
Fea u e ec o s a e e alua ed
wi h classi ica ion accu acy
and numbe o selec ed
ea u es
Selec ion o op imal ea u e
ec o s om FMO
Upda e ea u e ec o s wi h
aining and lea ning phases
Bina ize ea u e ec o s wi h
FMO pa ame e
A uni is adding o he coun e and he
pa ame e s a e ini ializing.
Upda e?
Yes
No
The op imal ea u e ec o is using o
machine lea ning and classi ica ion o
cance and non-cance colon pa ien s.
Cance
Non-cance
E alua ion o he p oposed
app oach h ough accu acy and
classi ica ion e o me ics o
bo h cance and non-cance
indi iduals
E e y ea u e ec o is conside ed as a membe o FMO
algo i hm
load he his opa hological
images om he da ase
Tes samples
Figu e 1. The p oposed me hodology’s amewo k o he diagnosis o pa ien s wi h colon cance .

Diagnos ics 2024,14, 1417 6 o 16
Diagnos ics 2024, 14, x FOR PEER REVIEW 8 o 18
Inpu Laye accep s his opa hological images as
pixel in ensi ies in ma ix o m
1
Con olu ional Laye s ex ac ea u es using
specialized il e s and ac i a ion unc ions like
ReLU.
Pooling Laye s downsize ea u e maps o educe
complexi y and o e i ing h ough max o a e age
pooling.
Fully Connec ed Laye s classi y lea ned ea u es
a e la ening, wi h a iable neu on numbe s.
Ou pu Laye in e p e s ea u es o bina y (cance
s. non-cance ) o mul i-class classi ica ion using
ac i a ion unc ions like sigmoid o so max.
The CNN specializes in analyzing his opa hological
images o malignancy ea u es, in eg a ed wi h he
Fishie Man is Op imize (FMO) o enhanced
e icacy.
The syne gy o deep lea ning in CNNs and FMO's
op imiza ion enhances accu acy and e iciency in
colon cance diagnosis, cap u ing ele an ea u es
and educing compu a ional complexi y.
2
3
4
5
6
7
Figu e 2. Con olu ional neu al ne wo k o colon cance diagnosis.
Fishie Man is Op imize Algo i hm
The ishe man is exhibi s in elligen hun ing beha io s, conside ing a ious scena -
ios and adjus ing i s posi ion acco dingly. I seeks he op imal loca ion o p ey o ish.
Addi ionally, he ishe man is displays uni o m beha io s, including p epa a ions o a -
acking o abandoning he cu en hun ing s a e.
The p oposed me hod makes use o he FMO algo i hm ou lined in e e ence [32].
This algo i hm sys ema ically ad ances h ough i e a ions, g adually b inging he man is
close o i s p ey. Th ough his p ocess, he algo i hm p og essi ely na ows down he
po en ial scena ios, as illus a ed in Equa ion (1). He e, he pa ame e “m” signi ies he
ini ial s a es, while “ ” ep esen s he s a es a he cu en i e a ion s age.
𝑚󰇛𝑡󰇜= 𝑚−𝑚·𝑖𝑡
𝑀𝑎𝑥𝐼𝑡 (1)
The ea u e ec o 𝑋
 will be used in machine lea ning o he classi ica ion o
colon images in o cance ous and non-cance ous ca ego ies.
The da ase u ilized in his s udy, “Lung and Colon Cance His opa hological Im-
ages,” was sou ced om an open-access da ase lib a y a ailable a
“h ps://www.kaggle.com/da ase s/and ewm d/lung-and-colon-cance -his opa hologi-
cal-images”, accessed on 10 Feb ua y 2021. I comp ises 25,000 his opa hology images ca -
ego ized in o i e classes. Each image is sa ed in JPEG o ma wi h dimensions o 768 ×
768 pixels.
Figu e 2. Con olu ional neu al ne wo k o colon cance diagnosis.
Fishie Man is Op imize Algo i hm
The ishe man is exhibi s in elligen hun ing beha io s, conside ing a ious scena ios
and adjus ing i s posi ion acco dingly. I seeks he op imal loca ion o p ey o ish. Addi-
ionally, he ishe man is displays uni o m beha io s, including p epa a ions o a acking
o abandoning he cu en hun ing s a e.
The p oposed me hod makes use o he FMO algo i hm ou lined in e e ence [
32
].
This algo i hm sys ema ically ad ances h ough i e a ions, g adually b inging he man is
close o i s p ey. Th ough his p ocess, he algo i hm p og essi ely na ows down he
po en ial scena ios, as illus a ed in Equa ion (1). He e, he pa ame e “m” signi ies he
ini ial s a es, while “ ” ep esen s he s a es a he cu en i e a ion s age.
m( )=m−m·i
MaxI (1)
The ea u e ec o
Xnew
i
will be used in machine lea ning o he classi ica ion o colon
images in o cance ous and non-cance ous ca ego ies.
The da ase u ilized in his s udy, “Lung and Colon Cance His opa hological Images,”
was sou ced om an open-access da ase lib a y a ailable a “h ps://www.kaggle.com/
da ase s/and ewm d/lung-and-colon-cance -his opa hological-images”, accessed on 10
Feb ua y 2021. I comp ises 25,000 his opa hology images ca ego ized in o i e classes.
Each image is sa ed in JPEG o ma wi h dimensions o 768 ×768 pixels.
Diagnos ics 2024,14, 1417 7 o 16
Augmen a ion P ocedu e:
To augmen he da ase and inc ease i s di e si y, a ious augmen a ion echniques
we e applied o he o iginal images. The augmen a ion p ocess was implemen ed using he
Augmen o package, which p o ides a lexible amewo k o image augmen a ion. The
ollowing augmen a ion echniques we e u ilized:
1. Ro a ion:
■
Range: Images we e o a ed wi hin a ange o angles o simula e a ia ions in
o ien a ion.
■Angle ange: [−15◦, 15◦].
2. T ansla ion:
■
Shi : Images we e shi ed ho izon ally and e ically o simula e ansla ions.
■Shi ange: [−20%, 20%] o image wid h and heigh .
3. Scaling:
■Scale ac o : Images we e scaled o simula e changes in size.
■Scale ange: [0.8, 1.2].
4. C opping:
■
Random c opping: Po ions o he images we e andomly c opped o simula e
a ia ions in composi ion.
■C op size: Images we e c opped o a size o 700 ×700 pixels.
5. Flipping:
■
Ho izon al lipping: Images we e lipped ho izon ally o simula e mi o
e lec ions.
■
Ve ical lipping: Images we e lipped e ically o in oduce addi ional a ia ions.
By applying hese augmen a ion echniques wi h speci ied pa ame e s, he da ase
was augmen ed o a o al o 25,000 images, ensu ing a di e se ep esen a ion o his opa ho-
logical ea u es. This augmen ed da ase was hen used o aining and e alua ing he
p oposed me hod o colon cance de ec ion. The classi ica ion ask o colon images in-
ol ed dis inguishing be ween cance ous and non-cance ous classes. The da ase consis ed
o 25,000 his opa hology images di ided in o i e dis inc ca ego ies, wi h each ca ego y
con aining 5000 images. I e ec i ely con eys in o ma ion abou he classi ica ion ask and
he da ase ela ed o colon images.
Addi ionally, o cla i y and isualiza ion pu poses, six examples o his opa hological
images om he da ase a e p o ided below. Images p e ixed wi h “colon_n_” indica e
heal hy colon issue images, while hose p e ixed wi h “colon_ca_” depic images o colon
cance . Re e o Figu e 3 o a isual ep esen a ion o hese augmen ed images.
His opa hological Fea u es and Classi ica ions:
His opa hology en ails he examina ion o issue samples unde a mic oscope, o en ob-
ained h ough biopsies, whe e minuscule issue agmen s a e ex ac ed and me iculously
analyzed by pa hologis s. This ho ough examina ion is ins umen al in iden i ying bo h
cance ous and p e-cance ous cellula abno mali ies. Apa om colon cance , he colon can
be a ec ed by a ange o o he condi ions, unde sco ing he impo ance o his opa hological
analysis in eaching a de ini i e diagnosis.
Diagnos ics 2024,14, 1417 8 o 16
Diagnos ics 2024, 14, x FOR PEER REVIEW 10 o 18
Figu e 3. Exempla pic u es om he da ase .
4. Resul s and Discussion
4.1. Classi ica ion Using Lea nable Classi ie s o FMO
To de e mine he ideal combina ion o echniques, a ho ough in es iga ion has been
ca ied ou . An au oencode me hod and he FMO algo i hm we e employed collabo a-
i ely on da ase s associa ed wi h colon disease o isola e and choose he mos c i ical a -
ibu es om he inpu aining da ase . The iden ical da ase s used in he i s model we e
ca ego ized using a p e- ained CNN in conjunc ion wi h he FMO me hod. Some im-
po an me ics, like accu acy, F1-Sco e, e c., a e applied o assess he effec i eness o
me hods c ea ed om he con usion ma ix. Fo mul iclass classi ica ion, me ics such as
o al accu acy, ue posi i e a e, and alse posi i e a e we e conside ed. The undamen al
e ms used in his analysis include False Posi i e (FP), T ue Posi i e (TP), T ue Nega i e
(TN), and False Nega i e (FN), which s and o posi i e and nega i e classi ica ions, e-
spec i ely.
These indica o s ha e been used o calcula e accu acy (ACC), sensi i i y (T ue Posi-
i e Ra e (TPR)), speci ici y (T ue Nega i e Ra e (TNR)), p ecision (posi i e p edic i e
alue (PPV)), nega i e p edic i e alue (NPV), and F1-Sco e as ollows:
Accu ac
y
=
TP  TN
TP  TN  FP  FN  10 (2)
S
ensi i i
y

T ue Posi i e Ra e
󰇛
TPR
󰇜
=TP
TP  FN  100 (3)
Speciici
y
󰇛T ue Nega i e Ra e󰇛TNR󰇜󰇜 = TN
TN  FP  100% (4)
P ecision 󰇛posi i e p edic i e alue 󰇛PPV󰇜󰇜 = TP
TP  FP  100% (5)
Nega i e p edic i e alue 󰇛NPV󰇜 = TN
TN  FN  100% (6)
F1 − sco e = 2PPV  TPR
PPV  TPR  100% (7)
Figu e 3. Exempla pic u es om he da ase .
4. Resul s and Discussion
4.1. Classi ica ion Using Lea nable Classi ie s o FMO
To de e mine he ideal combina ion o echniques, a ho ough in es iga ion has been
ca ied ou . An au oencode me hod and he FMO algo i hm we e employed collabo a-
i ely on da ase s associa ed wi h colon disease o isola e and choose he mos c i ical
a ibu es om he inpu aining da ase . The iden ical da ase s used in he i s model
we e ca ego ized using a p e- ained CNN in conjunc ion wi h he FMO me hod. Some
impo an me ics, like accu acy, F1-Sco e, e c., a e applied o assess he e ec i eness o
me hods c ea ed om he con usion ma ix. Fo mul iclass classi ica ion, me ics such as
o al accu acy, ue posi i e a e, and alse posi i e a e we e conside ed. The undamen al
e ms used in his analysis include False Posi i e (FP), T ue Posi i e (TP), T ue Nega-
i e (TN), and False Nega i e (FN), which s and o posi i e and nega i e classi ica ions,
espec i ely.
These indica o s ha e been used o calcula e accu acy (ACC), sensi i i y (T ue Posi i e
Ra e (TPR)), speci ici y (T ue Nega i e Ra e (TNR)), p ecision (posi i e p edic i e alue
(PPV)), nega i e p edic i e alue (NPV), and F1-Sco e as ollows:
Accu acy =TP +TN
TP +TN +FP +FN ×10 (2)
Sensi i i y(T ue Posi i e Ra e(TPR)) =TP
TP +FN ×100 (3)
Speci ici y (T ue Nega i e Ra e (TNR)) = TN
TN +FP ×100% (4)
P ecision (posi i e p edic i e alue (PPV)) = TP
TP +FP ×100% (5)
Nega i e p edic i e alue (NPV) = TN
TN +FN ×100% (6)
F1 −sco e =2PPV ×TPR
PPV +TPR ×100% (7)
4.2. Using Au o-Encode wi h FMO o Colon Disease Da ase
Va ious scena ios we e de eloped and assessed o alida e he e ec i eness o he
p oposed echnique and compa e di e en combina ions. In he ini ial s age, a da ase
ela ed o colon illness was p ocessed by he au oencode , along wi h i e di e en classi ie
Diagnos ics 2024,14, 1417 9 o 16
ypes. The ou comes o he colon illness da ase using he au oencode and he FMO
echnique a e displayed in Table 2.
Table 2. Da ase on colon disease au o-encode using FMO ea u e selec ion algo i hm.
Me hod ACC TPR TNR PPV NPV F1-Scoce
Decision T ee 67.80 67.94 67.94 68.20 68.20 67.93
SVM 68.40 73.83 73.83 79.80 79.80 71.63
KNN 75.35 77.05 77.05 78.50 78.50 76.10
Ensemble 73.00 74.63 74.63 76.30 76.30 73.86
Nai e Bayes 66.60 72.74 72.74 80.10 80.10 70.57
The mos c ucial ac o in assessing his classi ica ion model is accu acy, which is based
on he ue alues o he es ed images ha we e classi ied. The KNN classi ie achie ed a
highe accu acy a e o 75.35%.
4.3. P e-T ained CNN wi h FMO o Colon Disease Da ase
The assessmen o p e- ained CNN wi h FMO was conduc ed using he da ase ela ed
o colon diseases. The simula ion ou come, employing he FMO me hod in conjunc ion
wi h he p e- ained ResNe -50 ne wo k, is p esen ed in Figu e 4.
Diagnos ics 2024, 14, x FOR PEER REVIEW 11 o 18
4.2. Using Au o-Encode wi h FMO o Colon Disease Da ase
Va ious scena ios we e de eloped and assessed o alida e he effec i eness o he
p oposed echnique and compa e diffe en combina ions. In he ini ial s age, a da ase e-
la ed o colon illness was p ocessed by he au oencode , along wi h i e diffe en classi ie
ypes. The ou comes o he colon illness da ase using he au oencode and he FMO ech-
nique a e displayed in Table 2.
Table 2. Da ase on colon disease au o-encode using FMO ea u e selec ion algo i hm.
Me hod ACC TPR TNR PPV NPV F1-Scoce
Decision T ee 67.80 67.94 67.94 68.20 68.20 67.93
SVM 68.40 73.83 73.83 79.80 79.80 71.63
KNN 75.35 77.05 77.05 78.50 78.50 76.10
Ensemble 73.00 74.63 74.63 76.30 76.30 73.86
Nai e Bayes 66.60 72.74 72.74 80.10 80.10 70.57
The mos c ucial ac o in assessing his classi ica ion model is accu acy, which is
based on he ue alues o he es ed images ha we e classi ied. The KNN classi ie
achie ed a highe accu acy a e o 75.35%.
4.3. P e-T ained CNN wi h FMO o Colon Disease Da ase
The assessmen o p e- ained CNN wi h FMO was conduc ed using he da ase e-
la ed o colon diseases. The simula ion ou come, employing he FMO me hod in conjunc-
ion wi h he p e- ained ResNe -50 ne wo k, is p esen ed in Figu e 4.
Figu e 4. Resul o he simula ion based on he ResNe -50 and FMO.
Figu e 4 illus a es ha he accu acy o he DT, SVM, KNN, and ensemble me hods
has been achie ed a 90.82%, 95.01%, 95.04%, and 93.46%, espec i ely. The KNN classi i-
ca ion me hod achie ed 95.04% accu acy, which was he bes esul . KNN is a non-pa a-
me ic supe ised lea ning classi ie ha is mo e accu a e han o he me hods such as
SVM, decision ees, and ensemble me hods. Addi ionally, in his scena io, he highes
accu acy was achie ed using he KNN classi ie wi h ea u es ha we e ob ained using
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
100.00
Accu acy TPR TNR PPV NPV F1Sco e
Decision T ee 90.82 89.35 90.09 83.49 88.09 68.54
SVM 95.01 95.12 94.63 94.21 95.27 94.25
KNN 95.04 97.74 82.95 97.74 96.91 97.74
Ensemble 93.46 95.41 94.54 94.53 95.42 94.97
Figu e 4. Resul o he simula ion based on he ResNe -50 and FMO.
Figu e 4illus a es ha he accu acy o he DT, SVM, KNN, and ensemble me hods has
been achie ed a 90.82%, 95.01%, 95.04%, and 93.46%, espec i ely. The KNN classi ica ion
me hod achie ed 95.04% accu acy, which was he bes esul . KNN is a non-pa ame ic
supe ised lea ning classi ie ha is mo e accu a e han o he me hods such as SVM,
decision ees, and ensemble me hods. Addi ionally, in his scena io, he highes accu acy
was achie ed using he KNN classi ie wi h ea u es ha we e ob ained using he FMO
algo i hm and a p e- ained ResNe -50 ne wo k. Addi ionally, his s udy assessed he
pe o mance o decision ee, SVM, KNN, and ensemble me hods wi h he F1-Sco e me ic,
yielding sco es o 68.54%, 94.25%, 97.74%, and 94.97% o hese algo i hms, espec i ely.
As seen om he F1-Sco e esul s, i can be unde s ood ha he KNN has highe accu acy
han o he me hods.
Diagnos ics 2024,14, 1417 16 o 16
27.
Nisha, J.S.; Gopi, V.P.; Palanisamy, P. Au oma ed Colo ec al Polyp De ec ion Based on Image Enhancemen and Dual-Pa h CNN
A chi ec u e. Biomed. Signal P ocess. Con ol 2022,73, 103465. [C ossRe ]
28.
Nanni, L.; Fan ozzi, C.; Lo eggia, A.; Lumini, A. Ensembles o Con olu ional Neu al Ne wo ks and T ans o me s o Polyp
Segmen a ion. Senso s 2023,23, 4688. [C ossRe ]
29.
Mo in, O.; Valliè es, M.; B auns ein, S.; Gina , J.B.; Upadhaya, T.; Wood u , H.C.; Zwanenbu g, A.; Cha e jee, A.; Villanue a-
Meye , J.E.; Valdes, G. An A i icial In elligence F amewo k In eg a ing Longi udinal Elec onic Heal h Reco ds wi h Real-Wo ld
Da a Enables Con inuous Pan-Cance P ognos ica ion. Na . Cance 2021,2, 709–722. [C ossRe ]
30.
Du, Y.; Hu, L.; Wu, G.; Tang, Y.; Cai, X.; Yin, L. Diagnoses in Mul iple Types o Cance Based on Se um Raman Spec oscopy
Combined wi h a Con olu ional Neu al Ne wo k: Gas ic Cance , Colon Cance , Rec al Cance , Lung Cance . Spec ochim. Ac a
Pa A Mol. Biomol. Spec osc. 2023,298, 122743. [C ossRe ]
31.
Ka hikeyan, A.; Jo hilakshmi, S.; Su hi , S. Colo ec al Cance De ec ion Based on Con olu ional Neu al Ne wo ks (CNN) and
Ranking Algo i hm. Meas. Sens. 2024,31, 100976. [C ossRe ]
32.
Zhu, H.; Roelands, J.; Ahmed, E.I.; S ou en, I.; Hoo n je, R.; an Vlie be ghe, R.L.P.; Ijssels eijn, M.E.; Lei, X.; de Mi anda, N.F.C.C.;
Tollenaa , R.A.E.M. Loca ion Ma e s: Spa ial Dynamics o Tumo -In il a ing T Cell Subse s Is P ognos ic in Colon Cance . F on .
Immunol. 2024,15, 1293618. [C ossRe ]
33.
Shin, Y.; Balasingham, I. Compa ison o Hand-C a Fea u e Based SVM and CNN Based Deep Lea ning F amewo k o Au oma ic
Polyp Classi ica ion. In P oceedings o he 2017 39 h Annual In e na ional Con e ence o he IEEE Enginee ing in Medicine and
Biology Socie y (EMBC), Jeju Island, Republic o Ko ea, 11–15 July 2017; pp. 3277–3280.
34.
S idham, R.W.; Liu, W.; Bishu, S.; Rice, M.D.; Higgins, P.D.R.; Zhu, J.; Nallamo hu, B.K.; Waljee, A.K. Pe o mance o a Deep
Lea ning Model s. Human Re iewe s in G ading Endoscopic Disease Se e i y o Pa ien s wi h Ulce a i e Coli is. JAMA Ne w.
Open 2019,2, e193963. [C ossRe ]
35.
Poudel, S.; Kim, Y.J.; Vo, D.M.; Lee, S.-W. Colo ec al Disease Classi ica ion Using E icien ly Scaled Dila ion in Con olu ional
Neu al Ne wo k. IEEE Access 2020,8, 99227–99238. [C ossRe ]
Disclaime /Publishe ’s No e: The s a emen s, opinions and da a con ained in all publica ions a e solely hose o he indi idual
au ho (s) and con ibu o (s) and no o MDPI and/o he edi o (s). MDPI and/o he edi o (s) disclaim esponsibili y o any inju y o
people o p ope y esul ing om any ideas, me hods, ins uc ions o p oduc s e e ed o in he con en .