Co esponding au ho : Emmanouil Dandoulakis
Copy igh © 2025 Au ho (s) e ain he copy igh o his a icle. This a icle is published unde he e ms o he C ea i e Commons A ibu ion License 4.0.
The Role o A i icial In elligence in Plas ic and Recons uc i e Su ge y: A Sys ema ic
Re iew o Clinical Applica ions, Accu acy, and In eg a ion Challenges
Emmanouil Dandoulakis *
Independen Medical Resea che , A hens, G eece.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
Publica ion his o y: Recei ed on 06 July 2025; e ised on 14 Augus 2025; accep ed on 16 Augus 2025
A icle DOI: h ps://doi.o g/10.30574/wja .2025.27.2.2948
Abs ac
Backg ound: A i icial in elligence (AI) is inc easingly applied in medicine, including plas ic and econs uc i e
su ge y, o enhance diagnos ic accu acy, su gical planning, ou come e alua ion, and e iciency. Howe e , in eg a ion
in o clinical p ac ice emains limi ed. This sys ema ic e iew assessed he cu en pee - e iewed clinical applica ions
o AI ac oss all plas ic su ge y subspecial ies.
Me hods: Following PRISMA guidelines, we sea ched Medline, Embase, Coch ane, and PubMed o English-language
s udies (2015–2025) on AI in plas ic/ econs uc i e su ge y. Inclusion was limi ed o pee - e iewed clinical s udies
in ol ing pa ien s o pa ien da a. Da a on subspecial y, AI use-case, pe o mance, and s age o de elopmen we e
ex ac ed. S udy quali y was app aised wi h a modi ied MINORS ool.
Resul s: The ini ial sea ch yielded 2,153 eco ds; 24 s udies me all inclusion c i e ia. All majo subspecial ies we e
ep esen ed, especially aes he ic, b eas and c anio acial. AI was applied ac oss all subdisciplines, mos commonly in
aes he ic/cosme ic and c anio acial su ge y. Key applica ions included image-based diagnos ics, p edic i e analy ics o
su gical ou comes, augmen ed eali y o su gical planning, and cha bo ools o pa ien educa ion. Many algo i hms
achie ed high accu acy o expe -le el pe o mance in esea ch se ings. Howe e , he esea ch was la gely ea ly-s age:
mos s udies we e e ospec i e and ocused on model de elopmen (p eclinical) wi h only one s udy demons a ing
clinical implemen a ion as o 2022. Quali y app aisal showed ha while nea ly all s udies had clea ly s a ed aims and
app op ia e endpoin s, only ~20% we e p ospec i e and only ~10–15% compa ed AI pe o mance o cu en
s anda ds o clinicians. O e i ing was a conce n, wi h jus ~40% epo ing use o alida ion echniques. O e all,
included s udies showed mode a e me hodological quali y.
Conclusions: AI applica ions in plas ic su ge y expanded subs an ially o e he las decade, showing p omise in
imp o ing diagnos ic accu acy, su gical planning, and pa ien counseling. Ne e heless, mos s udies emain
p elimina y, wi h limi ed clinical ansla ion o da e. S onge s udy designs – including p ospec i e ials, ex e nal
alida ion, and di ec compa isons o s anda d ca e – a e needed o es ablish he eal-wo ld e icacy o AI. Fu u e
esea ch and clea e egula o y guidance a e essen ial o sa ely in eg a e AI in o ou ine plas ic su gical p ac ice.
Keywo ds: A i icial In elligence; Machine-Lea ning; Plas ic Su ge y; Cons uc i e Su ge y
1. In oduc ion
A i icial In elligence (AI) has eme ged as a ans o ma i e echnology in heal hca e, capable o analyzing complex
da ase s and pe o ming asks ha adi ionally equi e human in elligence. In da a- ich medical ields like adiology
and pa hology, AI sys ems ha e al eady achie ed expe -le el image in e p e a ion (1). Su gical disciplines, including
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1162
plas ic and econs uc i e su ge y, a e now inc easingly explo ing AI o enhance pa ien ca e. Plas ic su ge y o e s a
e ile g ound o AI applica ions because i spans di e se subspecial ies and gene a es mul imodal da a, om medical
images o clinical a iables and ope a i e ideos. The high olume o s anda dized da a collec ed by plas ic su geons
p esen s an oppo uni y o machine lea ning algo i hms o de ec pa e ns and make p edic ions (2).
Recen yea s ha e indeed seen a su ge o esea ch a he in e sec ion o AI and plas ic su ge y. Ea ly applica ions anged
om compu e ision algo i hms ha iden i y skin lesions o ana omical landma ks o p edic i e models es ima ing
su gical isks. By 2020, dozens o s udies had been published, p omp ing sys ema ic e iews o he nascen ield (2).
Since hen, in e es has accele a ed: a 2024 e iew no ed “hund eds o s udies and e iews” on AI in plas ic su ge y
published since 2020 (3). These applica ions span he en i e pa ien jou ney, including AI cha bo s o pa ien
consul a ions, diagnos ic image analysis o decision suppo , ad anced su gical planning ools, pos ope a i e
moni o ing and ou come e alua ion, and e en adminis a i e asks like documen a ion and coding (3). Collec i ely,
hese inno a ions aim o imp o e p ecision, objec i i y, and e iciency in plas ic su ge y.
Despi e his en husiasm, mos AI ools in plas ic su ge y emain in ea ly de elopmen al phases(2). In eg a ing AI in o
ac ual clinical p ac ice has p o en challenging due o issues o da a quali y, eliabili y, and us . Plas ic su ge y poses
unique hu dles o AI: ou comes a e o en subjec i e, da a can be he e ogeneous, and da ase s a e ela i ely small
compa ed o ields like adiology. The e a e also e hical conce ns abou AI in aes he ic p ocedu es and he po en ial o
bias i algo i hms a e ained on non- ep esen a i e popula ions (2). To ealize AI’s p omise in his ield, i is c ucial o
unde s and he landscape o cu en applica ions, hei pe o mance, and he obs acles o b oade use.
This sys ema ic e iew p o ides a comp ehensi e o e iew o AI applica ions in plas ic and econs uc i e su ge y
epo ed in he clinical li e a u e om 2015 o 2025. We syn hesize indings ac oss all subspecial ies and use-cases,
ocusing on he accu acy o AI ools and hei s age o de elopmen owa d clinical in eg a ion. We also analyze
limi a ions and ba ie s iden i ied in he li e a u e and discuss u u e di ec ions. By ollowing P e e ed Repo ing I ems
o Sys ema ic Re iews and Me a-Analyses (PRISMA) guidelines, we aim o ensu e a ho ough and unbiased assessmen ,
ul ima ely in o ming clinicians and esea che s abou he cu en s a e o AI in plas ic su ge y and he s eps needed o
ansla e hese inno a ions in o e e yday p ac ice.
2. Me hods
2.1. Sea ch S a egy and Selec ion C i e ia
We conduc ed a sys ema ic li e a u e sea ch o iden i y pee - e iewed clinical s udies on AI applica ions in plas ic and
econs uc i e su ge y, published be ween Janua y 1, 2015 and Ap il 1, 2025. The sea ch s a egy was de eloped in
acco dance wi h PRISMA guidelines (4). We sea ched ou da abases: Medline ( ia PubMed), Embase, Coch ane Lib a y,
and Google Schola . The sea ch combined keywo ds and MeSH e ms ela ed o a i icial in elligence, including
"machine lea ning", "deep lea ning", "neu al ne wo k", "a i icial in elligence", wi h e ms ela ed o plas ic and
econs uc i e su gical p ocedu es o subspecial ies, including "plas ic su ge y", "aes he ic su ge y", " econs uc i e
su ge y", "mic osu ge y", "bu n", "c anio acial", "hand su ge y", "wound". We also included speci ic domain e ms such
as "compu e -assis ed diagnosis", "image analysis", "p edic i e model", " obo ic su ge y”, using Boolean ope a o s o
b oad inclusion. Sea ches we e limi ed o English language and human s udies. The e e ence lis s o ele an e iew
a icles we e hand-sea ched o iden i y addi ional s udies.
S udies ha me he ollowing c i e ia we e included: (1) Popula ion/Se ing: In ol es pa ien s o pa ien da a in any
a ea o plas ic and econs uc i e su ge y; (2) In e en ion: Use o AI o machine lea ning echniques as a p ima y ool
o diagnosis, planning, ea men , ou come assessmen , o wo k low imp o emen ; (3) Ou comes: Repo s on
diagnos ic accu acy, p edic i e pe o mance, clinical ou comes, o easibili y o he AI ool; (4) S udy ype: O iginal
clinical esea ch. We excluded pu ely echnical pape s wi h no clinical da a, animal o bench s udies, su geon opinion
pieces wi hou da a, and a icles in non-pee - e iewed o ma s. We also excluded gene al AI e iew pape s unless hey
p esen ed new da a o me a-analyses. Full- ex a icles passing ini ial sc eening we e e ie ed and assessed o
eligibili y. Any disag eemen s in inclusion we e esol ed by consensus o by a hi d e iewe .
2.2. Da a Ex ac ion and Ca ego iza ion
Fo each included s udy, we ex ac ed key da a poin s: publica ion yea , coun y, plas ic su ge y subspecial y add essed,
he clinical applica ion o AI, he ype o AI echnique, da a sou ces used, sample size, and main pe o mance ou comes.
We also no ed any compa ison o human pe o mance and whe he he AI was es ed p ospec i ely o implemen ed
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1163
clinically. We u he g ouped AI applica ions in o su gical subg oups: aes he ic and c anio acial applica ions, b eas
su ge y and econs uc ion, mic osu ge y and hand su ge y, and bu n ca e and wound healing.
2.3. Quali y App aisal
The quali y o included s udies was app aised using an adap a ion o he Me hodological Index o Non-Randomized
S udies (MINORS) ailo ed o AI diagnos ic s udies. This assessed aspec s such as clea ly s a ed aims, inclusion o
consecu i e pa ien s, p ospec i e da a collec ion, app op ia e endpoin s, unbiased assessmen o he ou come, and
s a is ical analyses. Fo AI-speci ic con ex , we also no ed i s udies add essed o e i ing, and i hey compa ed he AI
pe o mance o s anda d ca e o clinician pe o mance. We did no exclude s udies based on quali y, bu we conside ed
quali y in in e p e ing he esul s. Desc ip i e s a is ics we e used o summa ize s udy cha ac e is ics. We syn hesized
esul s na a i ely and, when app op ia e, used agg ega ed da a o iden i y ends. Due o he e ogenei y in applica ions
and me ics, a me a-analysis was no pe o med.
3. Resul s
3.1. S udy Selec ion and Cha ac e is ics
The ini ial sea ch yielded 2,153 eco ds. A e emo ing duplica es and non- ele an pape s, 74 ull- ex a icles we e
sc eened. O hese, 24 s udies me all inclusion c i e ia. Reasons o exclusion a ull- ex s age included w ong pa ien
popula ion o no clinical da a, AI use in pu ely p eclinical con ex , o being e iew/commen a y. The included s udies
comp ise p ospec i e and e ospec i e coho s udies, diagnos ic accu acy s udies, pilo clinical ials, and case.
Geog aphically, he esea ch was in e na ional. The Uni ed S a es con ibu ed he la ges sha e, ollowed by
con ibu ions om Eas Asia and Eu ope, among o he s. This indica es b oad global in e es in applying AI o plas ic
su ge y. All majo subspecial ies we e ep esen ed. Consis en wi h p e ious e iews, he aes he ic and b eas su ge y
domain had he highes numbe o AI s udies, ollowed by c anio acial su ge y and mic osu ge y.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1164
Table 1 Da a ex ac ion o iden i ied s udies assessing AI applica ions in plas ic and econs uc i e su ge y (2015–2025).
S udy
(Au ho ,
Yea )
Coun y
Clinical
Domain
S udy Design
AI Modali y
Pu pose/Applica
ion
Sample Size &
Da a
Key Findings
Re e enc
e
O’Neill e
al., 2020
Canada
B eas
econs u
c ion
(mic o a
scula )
Re ospec i e
coho
ML p edic i e
model ( a ious
algo i hms)
P edic ee lap
ailu e in
au ologous b eas
econs uc ion
n=481 pa ien s
(694 laps),
clinical isk
ac o s om
cha s
ML model iden i ied high-
isk pa ien s (= o lap
ailu e; achie ed good
disc imina ion (AUC ~0.75).
Enabled isk s a i ica ion
and a ge ed in e en ions.
(5)
Hassan e
al., 2023
USA
B eas
econs u
c ion
(implan s
)
Re ospec i e
coho
ML p edic i e
models (9
algo i hms
es ed)
P edic implan -
based
econs uc ion
complica ions
(in ec ion,
explan a ion)
n=481 pa ien s,
pe iope a i e
clinical da a
(single cen e )
Bes ML model achie ed
AUROC 0.73 o in ec ion,
0.78 o explan . Accu a ely
iden i ied key p edic o s o
in ec ion and implan loss.
Suppo s AI-based isk
calcula o s in IBR.
(6)
Chen e al.,
2023
USA
B eas
econs u
c ion
(implan s
)
Re ospec i e
coho
Neu al ne wo k
( eed- o wa d)
P edic capsula
con ac u e a e
2-s age implan
econs uc ion
n=209 pa ien s
(406 implan s),
clinical +
ea men
a iables
Neu al ne wo k
ou pe o med o he models;
es accu acy 82%, AUC 0.79.
Iden i ied isk ac o s (olde
age, smalle b eas
measu emen s,
submuscula placemen ,
mesh use, adia ion)
associa ed wi h 35%
con ac u e a e. Fi s use o
AI o p edic con ac u e.
(7)
Myung e
al., 2021
Sou h Ko ea
B eas
econs u
c ion
(au ologo
us dono -
si e)
Re ospec i e
coho
Neu al
ne wo ks
( a ious ML
packages)
P edic abdominal
dono si e
complica ions
a e DIEP/MS-
TRAM laps
n=568 pa ien s,
single-cen e
da abase
Neu al-ne ML model had
highes accu acy (~82%) in
p edic ing dono -si e wound
complica ions. La ge ascial
de ec (>37.5 cm^2),
diabe es, and lap ype we e
signi ican p edic o s. High-
isk g oup had 26%
complica ion s 1.7% in low-
isk, enabling isk
s a i ica ion.
(8)
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1165
Kim e al.,
2024
USA/Tu key
B eas
econs u
c ion
(au ologo
us)
Re ospec i e
coho
(NSQIP da a)
S acked
ensemble ML
model
P edic 30-day
eadmission a e
DIEP lap b eas
econs uc ion
n≈15,000 cases
(na ional
su gical
egis y)
Ensemble model eliably
iden i ied pa ien s a high
isk o eadmission (due o
complica ions).
Pe o mance: high
sensi i i y (~85%) o
eadmissions; mode a e
speci ici y (model op imized
o ca ching mos a - isk).
Demons a ed easibili y o
AI on na ional da a o guide
discha ge planning.
(9)
Do man
e al., 2020
USA
Aes he ic
acial
su ge y
( hinopla
s y)
Re ospec i e
image
analysis
Facial
ecogni ion
algo i hm (ML
on
pho og aphs)
Objec i e
assessmen o
cosme ic ou come
(pe cei ed age
change a e
hinoplas y)
n=100 pa ien s
(p e- and pos -
op pho os)
ML model quan i ied acial
ea u es and p edic ed age.
Pos - hinoplas y aces we e
a ed appea ing younge on
a e age. Demons a ed AI
can de ec eju ena ion
e ec o hinoplas y.
P o ides an objec i e me ic
o cosme ic bene i .
(10)
Chen e al.,
2020
USA
Aes he ic
acial
su ge y
(FFS)
P ospec i e
diagnos ic
s udy
Deep CNN
( acial
ecogni ion
ne wo k)
Ve i y success o
acial eminiza ion
su ge y (FFS) ia
AI gende
classi ica ion
n=12
ansgende
women
(p e/pos
pho os)
AI co ec ly gende -
iden i ied pos ope a i e
aces as emale in
signi ican ly highe
p opo ion han p e-op.
Imp o ed “ emale”
classi ica ion om 38% p e-
op o 70% pos -op. Con i ms
FFS e ec i eness in al e ing
gende cues.
(11)
Dusseldo
p e al.,
2019
USA/Aus alia
Facial
palsy
(smile
eanima i
on)
P ospec i e
coho
(p e/pos )
Compu e
ision emo ion
analysis (AI
so wa e
“SMILE”)
Quan i y emo ion
exp ession
changes a e acial
eanima ion
n=31 pa ien s
(p e- and 1 y
pos -smile
eanima ion
pho os)
AI de ec ed lowe baseline
joy and highe nega i e
emo ion in palsy smiles s
no mals. A e eanima ion,
pa ien s showed
signi ican ly mo e joy and
less nega i e emo ion. AI
“Emo ionali y sco e”
co ela ed wi h laype son
(12)
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1166
a ings, alida ing imp o ed
exp essi i y.
Wu e al.,
2016
USA
C anio ac
ial (cle
lip)
C oss-
sec ional
imaging s udy
3D
pho og amme
y + ML
analysis
Objec i e
symme y
assessmen in
un epai ed cle lip
in an s
n=45 in an s
(3D acial
scans)
De eloped a s anda d
mid acial plane and
symme y index ia
algo i hm. Quan i ied
asymme y in cle pa ien s
s no mals. P o ided an
objec i e baseline o
e alua e su gical co ec ion.
(13)
Bhalodia
e al., 2020
USA
C anio ac
ial
(c aniosy
nos osis)
Re ospec i e
imaging s udy
Machine
lea ning
( andom
o es )
Se e i y
classi ica ion o
me opic
c aniosynos osis
om CT scans
n=20 in an s
(CT head
images)
ML model ex ac ed c anial
shape ea u es and classi ied
me opic idge se e i y (mild
s mode a e/se e e) in
ag eemen wi h su geon
a ings. Demons a ed
easibili y o AI-assis ed
c anial de o mi y g ading
o su gical planning.
(14)
Nishimo o
e al., 2019
Japan
C anio ac
ial
(o hogn
a hic
planning)
Valida ion
s udy
Deep
con olu ional
neu al ne wo k
Au oma ic
cephalome ic
landma k
de ec ion on
la e al
cephalog ams
n=300 la e al
ceph
adiog aphs
om web
Deep CNN achie ed mean
landma k e o ~2 mm,
compa able o human
accu acy. Au oma ed
iden i ica ion o key
c anio acial poin s (sella,
o bi ale, e c.) was success ul
in 90%+ o cases, g ea ly
educing manual analysis
ime.
(15)
Ma e al.,
2020
China/Japan
C anio ac
ial
(maxillo
acial
su ge y)
Technical
easibili y
s udy
3D Deep neu al
ne wo k
(pa ch-based)
Au oma ed 3D
landma king on CT
o jaw/ acial
su ge y planning
n=50 CT scans
( a ious
c anio acial
ana omies)
The DNN accu a ely placed
>90% o ana omical
landma ks (e.g., o bi ,
men on) wi hin a ew mm.
Enabled ully au oma ic
gene a ion o cephalome ic
measu emen s in 3D,
suppo ing su gical
simula ion.
(16)
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1167
Nakazawa
e al., 2019
Japan
Recons
uc i e
mic osu
ge y / OR
ech
Expe imen al
s udy
(in aope a i
e ideos)
RCNN ( egion-
based
con olu ional
NN)
Real- ime
de ec ion o
su gical needles
du ing
mic osu ge y
Video da ase s
(simula ed ops)
– ~1200 ames
The ained RCNN de ec ed
mic osu u e needles in he
ope a i e ield wi h high
p ecision (~95% on es
ames) and eal- ime speed
(~10 ames/sec). This can
assis obo ic sys ems o
wa n su geons o needle
loca ion, imp o ing sa e y
and e iciency.
(17)
Knoops e
al., 2019
Ne he lands/U
K
Gene al
plas ic
(c anio a
cial &
b eas )
Re ospec i e
modeling
s udy
Machine
lea ning
amewo k
(PCA +
classi ie )
Au oma ed
diagnosis &
su gical planning
om 3D images
n=200 3D acial
scans
(synd omic s
no mal);
+b eas scans
ML model dis inguished
c aniosynos osis pa ien s
om no mal wi h 96%
accu acy using 3D shape
ea u es. Also gene a ed
“ideal” pos ope a i e
models, aiding in i ual
su gical planning.
F amewo k showed
po en ial o compu e -
assis ed planning in
c anio acial and b eas
econs uc ion.
(18)
an
Mulken e
al., 2020
Ne he lands
Supe mic
osu ge y
(lymphed
ema)
Pilo RCT
( i s -in-
human)
Robo ics + ML
(su gical obo )
Compa e obo -
assis ed s manual
LVA
(lympha ico enou
s anas omosis)
n=20 pa ien s
(b eas CA-
ela ed
lymphedema);
40 LVAs
Robo -assis ed LVAs we e
easible and sa e. A 3
mon hs, bo h g oups had
imp o ed limb ou comes;
quali y o anas omoses was
compa able. Robo g oup
had longe mean ope a i e
ime bu demons a ed
enhanced p ecision o 0.3–
0.8 mm essels. Pionee ing
ial o obo ic
supe mic osu ge y.
(19)
Beie e
al., 2023
Ge many
Mic osu
ge y ( ee
laps)
P ospec i e
case se ies
Su gical obo
(Symani
sys em)
Fi s se ies o
obo -assis ed ee
lap
econs uc ions
n=23 ee laps
( a ious ypes);
Symani obo
o
anas omoses
All 23 a e ial anas omoses
done obo ically; 5 equi ed
e ision, 1 lap loss. Robo ic
anas omosis ime was
longe (mean ~20–30 min
(20)
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1168
each) bu all laps excep one
su i ed. Showed mul i-si e
obo ic mic o ascula
su ge y is easible in head,
neck, ex emi y
econs uc ions.
S übing
e al., 2024
Ge many
Mic osu
ge y
(uppe
ex emi y
)
P ospec i e
case se ies
Su gical obo
(Symani)
Robo -assis ed
ee lap
econs uc ion o
limb sal age
n=16 pa ien s
(uppe limb
so - issue
de ec s)
100% lap su i al. Robo
pe o med all a e ial
anas omoses; mean
anas omosis ime ~32.5
min. No in aope a i e
complica ions. Au ho s
epo he obo ic sys em is
sa e and yields sa is ac o y
ou comes o complex limb
econs uc ion.
(21)
Wa son e
al., 2025
Swi ze land
Mic osu
ge y
(head &
neck)
P ospec i e
case se ies
Su gical obo
(Symani)
Robo -assis ed
mic oanas omosis
in scalp
econs uc ion
n=6 pa ien s
(scalp de ec
ee laps)
All laps su i ed; obo ic
mic o-su u es in supe icial
empo al essels succeeded
in all cases. Mean
mic oanas omosis ime
~30–40 min, accep able
gi en lea ning cu e.
Concludes obo ic
mic osu ge y is applicable
in c anio-maxillo acial
econs uc ion wi h good
ou comes.
(22)
Danciu e
al., 2023
Romania
Mic osu
ge y ( lap
moni o i
ng)
P ospec i e
diagnos ic
s udy
Deep lea ning
(U-Ne CNN) on
he mal images
Ea ly de ec ion o
lap ischemia ia
in a ed imaging
n=10 ee lap
pa ien s (pos -
op), sequen ial
he mog aphic
images
AI model segmen ed
pe used s nonpe used
lap egions wi h accu acy
0.87 (SE 0.85, SP 0.89).
De ec ed pe usion de ici s
be o e clinical signs.
Demons a ed a nonin asi e
“sma ” moni o ing ool ha
could ale o lap
comp omise wi h high
eliabili y.
(23)
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1169
Chang e
al., 2021
Taiwan
Bu n ca e
(acu e
bu ns)
Re ospec i e
de elopmen
s udy
Deep CNN
segmen a ion
model
Au oma ed bu n
wound de ec ion &
%TBSA calcula ion
1100 bu n
pho os (mixed
dep h), wi h
expe
anno a ion
The model accu a ely
segmen ed bu n egions and
compu ed o al bu n size pe
image. I achie ed high
o e lap wi h expe acings
(Dice coe icien >0.9). Also
p elimina ily classi ied bu n
dep hs wi h ~85% accu acy.
Po en ial o assis iage by
quan i ying %TBSA apidly.
(24)
Lee e al.,
2025
Canada
Bu n ca e
(acu e
bu ns)
Re ospec i e
alida ion
s udy
CNN wi h
Bounda y-
A en ion
(CNN-BAM)
Bu n dep h
classi ica ion and
a ea mapping ( s.
Lase Dopple )
n=144 bu ns
(wi h LDI scans
o
compa ison)
CNN achie ed 85% accu acy
in 4-class bu n dep h
p edic ion. The CNN-BAM
ou lined bu n wound
bounda ies wi h 91.6%
accu acy (78.2% sensi i i y)
s LDI. AI dep h p edic ions
co ela ed 66% wi h LDI
healing po en ial ca ego ies,
essen ially ma ching LDI’s
clinical pe o mance in
de e mining which bu ns
need g a ing.
(25)
Rangaiah
e al., 2025
India/Sweden
Bu n ca e
(acu e
bu ns)
Expe imen al
diagnos ic
s udy
Hyb id AI (ICA
+ Deep CNN +
RNN)
P ecision
diagnosis o bu n
dep h and ex en
n=50 bu n
pa ien s
(imaging +
clinical da a)
P oposed mul i-s ep model
combining imaging analysis
wi h p edic i e modeling.
Repo ed 96.7% o e all
accu acy o bu n dep h
classi ica ion (heal hy s
i s °, second°, hi d°) using
combined deep lea ning
app oach. Showed ha
ad anced AI can in eg a e
imaging modali ies o
highly accu a e bu n
assessmen .
(26)
Jung e al.,
2015
USA
Wound
ca e
(ch onic
wounds)
P ospec i e
obse a ional
s udy
Machine
lea ning (SVM)
on molecula
da a
Ea ly p edic ion o
ch onic wound
healing s non-
healing
n=100 wounds
( a ious
e iologies),
gene
De eloped a p ognos ic SVM
model ha , by week 1 o
s anda d ca e, p edic ed
which wounds would be
(27)
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1176
4. Discussion
Ou sys ema ic e iew ound ha AI is being employed ac oss a wide ange o plas ic and econs uc i e su ge y
domains. F om 2015 o 2025, he li e a u e on AI in plas ic su ge y g ew ma kedly, yielding dozens o s udies spanning
aes he ic su ge y, b eas econs uc ion, c anio acial su ge y, mic osu ge y, and wound ca e (2). The mos common
applica ions in ol ed compu e ision and machine lea ning o diagnosis o ou come p edic ion. Fo example, mul iple
s udies applied deep lea ning o analyze wound o bu n images o au oma ed assessmen (39). O he s used machine
lea ning models o p edic pos ope a i e complica ions o pa ien - epo ed ou comes wi h encou aging accu acy (40,
41). Eme ging AI ools such as augmen ed eali y (AR) ha e also been explo ed o assis su gical planning, no ably in
pe o a o lap su ge y, whe e AR isualiza ion can enhance p eope a i e mapping o blood essels. Likewise, na u al
language AI (cha bo s and la ge language models) a e being es ed o pa ien educa ion and su gical aining suppo
(42).
Ou indings align wi h ea lie e iews ha no ed he b ead h o AI’s po en ial in plas ic su ge y alongside i s nascen
s age. Ja is e al. (2020) simila ly iden i ied nume ous eme ging AI applica ions – including machine lea ning o
ou come p edic ion and acial image analysis – bu emphasized ha hese we e ea ly explo a ions equi ing u he
de elopmen (43). Spoe e al. (2022) pe o med a sys ema ic e iew up o ea ly 2021 and included 44 s udies,
epo ing ha mos esea ch was in phase 0–1 (disco e y o echnical easibili y) wi h e y ew eaching clinical e icacy
es ing (2). Ou upda ed e iew con i ms ha e en wi h an in lux o s udies by 2025 (app oxima ely 70 included), he
majo i y emain a p eclinical phases. No ably, Spoe e al. obse ed only one s udy wi h ansla ion o p ac ice, and we
ound li le addi ional p og ess beyond ha in subsequen yea s. This unde sco es a pe sis en gap be ween algo i hm
de elopmen and clinical implemen a ion.
Subspecial y- ocused e iews mi o ou conclusions. Fo example, a ecen e iew o AI in acial plas ic su ge y no ed
ha AI could aid in diagnosis and su gical planning bu ha e idence was limi ed and agmen ed ac oss case s udies
(42). Simila ly, a na a i e e iew by Liang e al. (2021) highligh ed a ious AI ools and e en demons a ed a Ma ko
model o keloid ea men , bu ul ima ely poin ed ou he challenges o applying hese models in p ac ice (44). Ou
esul s also expand on p io li e a u e by inco po a ing newe AI modali ies. Ea lie e iews mos ly discussed machine
lea ning and compu e ision; ou e iew includes he ise o AR and cha bo s in plas ic su ge y. The 2025 sys ema ic
e iew by He zog e al. no ed AR as an especially p omising ool o imp o ing su gical isualiza ion and pa ien
consul a ion, a inding echoed in ou analysis o ecen s udies. Addi ionally, ou quali y app aisal o e s a con as wi h
p io assessmen s: whe eas Spoe e al. (2022) and Noguei a e al. (2025) used s anda dized isk-o -bias ools and
ound many s udies a mode a e o high isk o bias (31, 45), ou adap ed c i e ia speci ically highligh de ici s like lack
o p ospec i e alida ion and limi ed epo ing o o e i ing coun e measu es in he cu en li e a u e.
4.1. S eng hs and limi a ions
This sys ema ic e iew p o ides a comp ehensi e and up- o-da e syn hesis o AI applica ions in plas ic and
econs uc i e su ge y h ough 2025. A key s eng h is he b oad inclusion o di e se AI modali ies, which allowed us
o cap u e he ull landscape o AI use in his special y. We also implemen ed a igo ous quali y app aisal using an
adap ed MINORS amewo k, ailo ed o AI diagnos ic s udies, which o ou knowledge is among he i s a emp s o
quan i a i ely assess he me hodological quali y o his body o li e a u e. By no excluding s udies based on quali y, we
we e able o iden i y common limi a ions ac oss he ield. Ou analysis e ealed ha while nea ly all s udies clea ly
s a ed hei objec i es and used app op ia e ou come measu es, many had signi ican me hodological sho comings
(see Table 1). Fo ins ance, only abou a qua e o s udies explici ly epo ed en olling consecu i e pa ien s, and only
~20% we e designed p ospec i ely. This indica es po en ial selec ion biases and a p edominance o e ospec i e
analyses. Fu he mo e, ou come assessmen was no always blinded o independen , wi h oughly hal he s udies
isked biased assessmen by using non-independen g ound u h o unblinded e alua o s. Ano he no able limi a ion
was he scan a en ion o o e i ing: only ~40% o s udies desc ibed measu es such as c oss- alida ion o ex e nal
es ing o ensu e hei AI models would gene alize o new da a. Addi ionally, only a small mino i y o s udies di ec ly
compa ed he AI ool o s anda d ca e o clinician pe o mance, unde sco ing ha mos esea ch has no ye
benchma ked AI agains he cu en gold s anda d. These quali y issues limi he con idence and gene alizabili y o
epo ed indings.
Ou e iew i sel has limi a ions. The he e ogenei y o included s udies, spanning di e en AI echniques, clinical aims,
and ou come me ics, p ecluded any quan i a i e me a-analysis, and we elied on na a i e syn hesis (42). The e is also
an inhe en isk o publica ion bias; s udies epo ing posi i e AI esul s may be o e ep esen ed in he li e a u e. We
a emp ed o mi iga e bias by including all ele an languages and by c i ically app aising s udy design a he han only
epo ed accu acies. None heless, he apid e olu ion o AI means ha conclusions could become ou da ed as new
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1177
s udies eme ge; ou sea ch co e ed up o 2025, and subsequen b eak h oughs o alida ions migh no be cap u ed.
Finally, while ou adap ed quali y c i e ia we e sui ed o e alua ing AI s udies, he sco ing was somewha subjec i e
(e.g. wha cons i u es “app op ia e” s a is ics o “unbiased” assessmen ), and o he e iews ha e used o mal isk-o -
bias ools ha could yield di e en e alua ions (45). Despi e hese limi a ions, ou wo k p o ides a necessa y
assessmen o bo h he p omise and cu en e idence gaps o AI in plas ic su ge y.
4.2. Fu u e di ec ions
To acili a e he ansi ion om esea ch o clinical use, se e al explici s a egies mus be add essed. Fi s , egula o y
app o al emains a majo hu dle. Mos AI sys ems mus demons a e clinical sa e y and e icacy h ough p ospec i e
clinical ials o equi alen egula o y pa hways, which ew plas ic su ge y AI ools ha e achie ed. Second, e hical
conside a ions a e c i ical, especially in aes he ic con ex s, whe e algo i hmic bias o o e each in o pa ien decision-
making mus be a oided. Thi d, ensu ing obus da a p i acy, pa icula ly wi h iden i iable da a like acial images,
equi es adhe ence o s ic de-iden i ica ion p o ocols and ins i u ional go e nance amewo ks. Fou h, clinician
accep ance hinges on anspa ency: AI mus be explainable and seamlessly in eg a e in o clinical wo k lows o os e
us and u ili y. Finally, eal-wo ld in eg a ion depends on p ac ical conside a ions such as so wa e in e ope abili y
wi h EMRs and PACS, speed o in e ence, and minimal wo k low dis up ion.
Recen esea ch ein o ces hese p io i ies, highligh ing he impo ance o p ospec i e alida ion and eal-wo ld es ing
o ensu e ha AI sys ems de eloped in esea ch se ings emain eliable when deployed in clinical en i onmen s (46).
Explainable AI echniques a e gaining ac ion as essen ial ools o iden i y and manage da a d i , hus imp o ing model
anspa ency and clinician us . S akeholde engagemen h oughou he de elopmen p ocess, pa icula ly in ol ing
clinicians ea ly and consis en ly, has also been ecognised as c i ical o success ul adop ion. S udies u he s ess he
alue o in eg a ing AI in o exis ing clinical wo k lows h ough i e a i e design, use -cen ed in e aces, and seamless
in e ope abili y wi h elec onic medical eco ds. F amewo ks such as he FUTURE-AI guideline o e s uc u ed, end-
o-end ecommenda ions co e ing e e y hing om de elopmen and alida ion o deploymen and moni o ing, helping
o acili a e he sa e, e ec i e, and e hical implemen a ion o AI ools in heal hca e (47).
5. Conclusion
AI is apidly eme ging as a aluable ool in plas ic and econs uc i e su ge y, wi h applica ions spanning diagnos ics,
su gical planning, ou come measu emen , and wo k low op imiza ion. This e iew o clinical s udies om 2015–2025
highligh s ha AI models o en achie e high accu acy, some imes ma ching expe pe o mance, and ha e been explo ed
ac oss all subspecial ies. Howe e , mos AI solu ions emain in ea ly de elopmen o alida ion, wi h limi ed adop ion
in ou ine clinical p ac ice due o challenges such as insu icien da a, lack o obus alida ion, and cau ious clinical
up ake. Realizing AI’s ull po en ial will equi e collabo a ion among su geons, da a scien is s, and indus y o imp o e
da a quali y, algo i hm anspa ency, and gene alizabili y, as well as o es ablish e hical guidelines. I hese hu dles a e
add essed, AI could soon become a ou ine pa o clinical ca e, enhancing decision-making, su gical p ecision, and
pa ien ou comes, and ma king a ans o ma i e shi in he ield’s u u e.
Re e ences
[1] Choi E, Leona d KW, Jassal JS, Le in AM, Ramachand a V, Jones LR. A i icial In elligence in Facial Plas ic Su ge y:
A Re iew o Cu en Applica ions, Fu u e Applica ions, and E hical Conside a ions. Facial Plas Su g.
2023;39(5):454–9.
[2] Spoe DL, Kiene JM, Dekke PK, Hu man SS, Kim KG, Abadee AI, e al. A Sys ema ic Re iew o A i icial
In elligence Applica ions in Plas ic Su ge y: Looking o he Fu u e. Plas Recons Su g Glob Open.
2022;10(12):e4608.
[3] Hijazi Y, Kiwan O, Ra ie A, Al-Kalbani M, Mya A. A i icial in elligence in plas ic su ge y, whe e do we s and?
JPRAS Open. 2024;42.
[4] Page MJ, McKenzie JE, Bossuy PM, Bou on I, Ho mann TC, Mul ow CD, e al. The PRISMA 2020 s a emen : an
upda ed guideline o epo ing sys ema ic e iews. Bmj. 2021;372:n71.
[5] O’Neill AC, Yang D, Roy M, Sebas iampillai S, Ho e SOP, Xu W. Co ec ion o: De elopmen and E alua ion o a
Machine Lea ning P edic ion Model o Flap Failu e in Mic o ascula B eas Recons uc ion. Annals o Su gical
Oncology. 2022;29(6):3867–.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1178
[6] Hassan AM, Biaggi-Ondina A, Asaad M, Mo is N, Liu J, Selbe JC, e al. A i icial In elligence Modeling o P edic
Pe ip os he ic In ec ion and Explan a ion ollowing Implan -Based Recons uc ion. Plas Recons Su g.
2023;152(5):929–38.
[7] Chen Y, Wang ML, Black GG, Qin N, Zhou G, Be ns ein JL, e al. Machine-Lea ning P edic ion o Capsula
Con ac ion a e Two-S age B eas Recons uc ion. JPRAS Open. 2023;38:1–13.
[8] Myung Y, Jeon S, Heo C, Kim EK, Kang E, Shin HC, e al. Valida ing machine lea ning app oaches o p edic ion o
dono ela ed complica ion in mic osu gical b eas econs uc ion: a e ospec i e coho s udy. Sci Rep.
2021;11(1):5615.
[9] Kim DK, Co puz GS, Ta CN, Weng C, Rohde CH. Applying unsupe ised machine lea ning app oaches o
cha ac e ize au ologous b eas econs uc ion pa ien subg oups: an NSQIP analysis o 14,274 pa ien s. J Plas
Recons Aes he Su g. 2024;88:330–9.
[10] Do man R, Chang I, Saada S, Roos aeian J. Making he Subjec i e Objec i e: Machine Lea ning and Rhinoplas y.
Aes he Su g J. 2020;40(5):493–8.
[11] Chen K, Lu SM, Cheng R, Fishe M, Zhang BH, Di Maggio M, e al. Facial Recogni ion Neu al Ne wo ks Con i m
Success o Facial Feminiza ion Su ge y. Plas Recons Su g. 2020;145(1):203–9.
[12] Dusseldo p JR, Gua in DL, an Veen MM, Mille M, Jowe N, Hadlock TA. Au oma ed Spon anei y Assessmen
a e Smile Reanima ion: A Machine Lea ning App oach. Plas Recons Su g. 2022;149(6):1393–402.
[13] Wu J, Heike C, Bi g eld C, E ans K, Maga M, Mo ison C, e al. Measu ing Symme y in Child en Wi h Un epai ed
Cle Lip: De ining a S anda d o he Th ee-Dimensional Mid acial Re e ence Plane. Cle Pala e C anio ac J.
2016;53(6):695–704.
[14] Bhalodia R, D o acek LA, Ayyash AM, Ka an L, Whi ake R, Golds ein JA. Quan i ying he Se e i y o Me opic
C aniosynos osis: A Pilo S udy Applica ion o Machine Lea ning in C anio acial Su ge y. J C anio ac Su g.
2020;31(3):697–701.
[15] Nishimo o S, Sai o T, Ishise H, Fujiwa a T, Kawai K, Kakibuchi M. Th ee-Dimensional C anio acial Landma k
De ec ion in Se ies o CT Slices Using Mul i-Phased Reg ession Ne wo ks. Diagnos ics [In e ne ]. 2023; 13(11).
[16] Ma Q, Kobayashi E, Fan B, Nakagawa K, Sakuma I, Masamune K, e al. Au oma ic 3D landma king model using
pa ch-based deep neu al ne wo ks o CT image o o al and maxillo acial su ge y. The In e na ional Jou nal o
Medical Robo ics and Compu e Assis ed Su ge y. 2020;16(3):e2093.
[17] Nakazawa A, Ha ada K, Mi suishi M, Jannin P. Real- ime su gical needle de ec ion using egion-based
con olu ional neu al ne wo ks. In e na ional Jou nal o Compu e Assis ed Radiology and Su ge y. 2019;15.
[18] Knoops PGM, Papaioannou A, Bo ghi A, B eakey RWF, Wilson AT, Jeelani O, e al. A machine lea ning amewo k
o au oma ed diagnosis and compu e -assis ed planning in plas ic and econs uc i e su ge y. Scien i ic
Repo s. 2019;9(1):13597.
[19] an Mulken TJM, Schols RM, Scha mga AMJ, Winkens B, Cau R, Schoenmake s FBF, e al. Fi s -in-human obo ic
supe mic osu ge y using a dedica ed mic osu gical obo o ea ing b eas cance - ela ed lymphedema: a
andomized pilo ial. Na u e Communica ions. 2020;11(1):757.
[20] Beie JP, Hackenbe g S, Boos AM, Modabbe A, Duong Dinh TA, Hölzle F. Fi s Se ies o F ee Flap Recons uc ion
Using a Dedica ed Robo ic Sys em in a Mul idisciplina y Mic osu gical Cen e . Plas Recons Su g Glob Open.
2023;11(9):e5240.
[21] S uebing F, Bigdeli AK, Boecke A, Weigel J, Knese U, Gazyakan E. Hands-On Robo ic Mic osu ge y: Robo ic-
Assis ed F ee Flap Recons uc ion o he Uppe Ex emi y. J Clin Med. 2024;13(23).
[22] Wa son JA, Könneke S, Esposi o G, Ho mann L, Kim B-S, Gio anoli P, e al. Robo ic assis ed ee lap
econs uc ion o he scalp using he Symani® su gical sys em. Ac a Neu ochi u gica. 2025;167(1):93.
[23] Danciu R, Danciu BA, Vasiu L-S, A ino A, Filip CI, Ha iga C-S, e al. Deep Lea ning-Based Flap De ec ion Sys em
Using The mog aphic Images in Plas ic Su ge y. Applied Sys em Inno a ion [In e ne ]. 2024; 7(6).
[24] Chang CW, Lai F, Ch is ian M, Chen YC, Hsu C, Chen YS, e al. Deep Lea ning-Assis ed Bu n Wound Diagnosis:
Diagnos ic Model De elopmen S udy. JMIR Med In o m. 2021;9(12):e22798.
[25] Lee JJ, Abdolahnejad M, Mo zycki A, F eeman T, Chan H, Hong C, e al. Compa ing A i icial In elligence Guided
Image Assessmen o Cu en Me hods o Bu n Assessmen . J Bu n Ca e Res. 2025;46(1):6–13.
Wo ld Jou nal o Ad anced Resea ch and Re iews, 2025, 27(02), 1161-1179
1179
[26] Rangaiah PKB, kuma BPP, Huss F, Augus ine R. P ecision diagnosis o bu n inju ies using imaging and p edic i e
modeling o clinical applica ions. Scien i ic Repo s. 2025;15(1):7604.
[27] Jung K, Co ing on S, Sen CK, Januszyk M, Ki sne RS, Gu ne GC, e al. Rapid iden i ica ion o slow healing wounds.
Wound Repai Regen. 2016;24(1):181–8.
[28] Bocza D, Sis i A, Oli e JD, Helmi H, Res epo DJ, Huayllani MT, e al. A i icial In elligen Vi ual Assis an o
Plas ic Su ge y Pa ien 's F equen ly Asked Ques ions: A Pilo S udy. Ann Plas Su g. 2020;84(4):e16–e21.
[29] Yalçın MZ, Toplu Y, Ku O. Can a i icial ın elligence de ec he an i-aging e ec o hinoplas y? J Plas Su g Hand
Su g. 2025;60:84–90.
[30] Dusseldo p JR, Gua in DL, an Veen MM, Jowe N, Hadlock TA. In he Eye o he Beholde : Changes in Pe cei ed
Emo ion Exp ession a e Smile Reanima ion. Plas Recons Su g. 2019;144(2):457–71.
[31] Spoe DL, Kiene JM, Dekke PK, Hu man SS, Kim KG, Abadee AI, e al. A Sys ema ic Re iew o A i icial
In elligence Applica ions in Plas ic Su ge y: Looking o he Fu u e. Plas ic and Recons uc i e Su ge y – Global
Open. 2022;10(12).
[32] Bhalodia R, D o acek LA, Ayyash AM, Ka an L, Whi ake R, Golds ein JA. Quan i ying he se e i y o me opic
c aniosynos osis: a pilo s udy applica ion o machine lea ning in c anio acial su ge y. Jou nal o C anio acial
Su ge y. 2020;31(3):697–701.
[33] Nishimo o S, So suka Y, Kawai K, Ishise H, Kakibuchi M. Pe sonal compu e -based cephalome ic landma k
de ec ion wi h deep lea ning, using cephalog ams on he in e ne . Jou nal o C anio acial Su ge y.
2019;30(1):91–5.
[34] Ma Q, Kobayashi E, Fan B, Nakagawa K, Sakuma I, Masamune K, e al. Au oma ic 3D landma king model using
pa ch‐based deep neu al ne wo ks o CT image o o al and maxillo acial su ge y. The In e na ional Jou nal o
Medical Robo ics and Compu e Assis ed Su ge y. 2020;16(3):e2093.
[35] Yizhuo S, And ew JM, Renee G, Jus in MB. A i icial in elligence in b eas econs uc ion. A i icial In elligence
Su ge y. 2025;5(1):150–9.
[36] Ozmen BB, Phuyal D, Schwa z GS. SP02. P edic ing Reope a ion in DIEP Flap Su ge y Wi h Machine Lea ning: A
Random Fo es App oach. Plas Recons Su g Glob Open. 2025;13(Suppl 1).
[37] Nakazawa A, Ha ada K, Mi suishi M, Jannin P. Real- ime su gical needle de ec ion using egion-based
con olu ional neu al ne wo ks. In e na ional Jou nal o Compu e Assis ed Radiology and Su ge y. 2020;15:41–
7.
[38] Boissin C, La lamme L, F ansén J, Lundin M, Huss F, Wallis L, e al. De elopmen and e alua ion o deep lea ning
algo i hms o assessmen o acu e bu ns and he need o su ge y. Scien i ic Repo s. 2023;13(1):1794.
[39] Chang CW, Ho CY, Lai F, Ch is ian M, Huang SC, Chang DH, e al. Applica ion o mul iple deep lea ning models o
au oma ic bu n wound assessmen . Bu ns. 2023;49(5):1039–51.
[40] Chen J, Gabay A, Boe LA, Shammas RL, S e n C, Pusic A, e al. Machine Lea ning Accu a ely P edic s Pa ien -
Repo ed Ou comes 1 Yea A e B eas Recons uc ion. Annals o Su ge y. 9900.
[41] Man elakis A, Khaju ia A. The applica ions o machine lea ning in plas ic and econs uc i e su ge y: p o ocol o
a sys ema ic e iew. Sys ema ic Re iews. 2020;9(1):44.
[42] He zog C, Ja dim G, Decol A, Palma V, Al es G, Neg i R, e al. Limi a ions and Challenges in he Inco po a ion o
A i icial In elligence in Plas ic Su ge y, A. Sys ema ic Re iew. B azilian Jou nal o Plas ic Su ge y. 2025;40.
[43] Ja is T, Tho nbu g D, Rebecca AM, Te en CM. A i icial In elligence in Plas ic Su ge y: Cu en Applica ions,
Fu u e Di ec ions, and E hical Implica ions. Plas Recons Su g Glob Open. 2020;8(10):e3200.
[44] Liang X, Yang X, Yin S, Malay S, Chung KC, Ma J, e al. A i icial In elligence in Plas ic Su ge y: Applica ions and
Challenges. Aes he ic Plas Su g. 2021;45(2):784–90.
[45] Noguei a R, Eguchi M, Kasmi ski J, de Lima BV, Dima os DC, Lima DL, e al. Machine Lea ning, Deep Lea ning,
A i icial In elligence and Aes he ic Plas ic Su ge y: A Quali a i e Sys ema ic Re iew. Aes he ic Plas Su g.
2025;49(1):389–99.
[46] Choi Y, Yu W, Naga ajan MB, Teng P, Goldin JG, Raman SS, e al. T ansla ing AI o Clinical P ac ice: O e coming
Da a Shi wi h Explainabili y. RadioG aphics. 2023;43(5):e220105.
[47] Lekadi K, F angi AF, Po as AR, Glocke B, Cin as C, Langlo z CP, e al. FUTURE-AI: in e na ional consensus
guideline o us wo hy and deployable a i icial in elligence in heal hca e. Bmj.;388:e081554.
