current and future applications of machine and deep
TRANSCRIPT
Vol.:(0123456789)1 3
World Journal of Urology https://doi.org/10.1007/s00345-019-03000-5
TOPIC PAPER
Current and future applications of machine and deep learning in urology: a review of the literature on urolithiasis, renal cell carcinoma, and bladder and prostate cancer
Rodrigo Suarez‑Ibarrola1 · Simon Hein1 · Gerd Reis2 · Christian Gratzke1 · Arkadiusz Miernik1
Received: 24 July 2019 / Accepted: 25 October 2019 © Springer-Verlag GmbH Germany, part of Springer Nature 2019
AbstractPurpose The purpose of the study was to provide a comprehensive review of recent machine learning (ML) and deep learning (DL) applications in urological practice. Numerous studies have reported their use in the medical care of various urological disorders; however, no critical analysis has been made to date.Methods A detailed search of original articles was performed using the PubMed MEDLINE database to identify recent English literature relevant to ML and DL applications in the fields of urolithiasis, renal cell carcinoma (RCC), bladder cancer (BCa), and prostate cancer (PCa).Results In total, 43 articles were included addressing these four subfields. The most common ML and DL application in urolithiasis is in the prediction of endourologic surgical outcomes. The main area of research involving ML and DL in RCC concerns the differentiation between benign and malignant small renal masses, Fuhrman nuclear grade prediction, and gene expression-based molecular signatures. BCa studies employ radiomics and texture feature analysis for the distinction between low- and high-grade tumors, address accurate image-based cytology, and use algorithms to predict treatment response, tumor recurrence, and patient survival. PCa studies aim at developing algorithms for Gleason score prediction, MRI computer-aided diagnosis, and surgical outcomes and biochemical recurrence prediction. Studies consistently found the superiority of these methods over traditional statistical methods.Conclusions The continuous incorporation of clinical data, further ML and DL algorithm retraining, and generalizability of models will augment the prediction accuracy and enhance individualized medicine.
Keywords Artificial intelligence · Machine learning · Deep learning · Artificial neural network · Convolutional neural network · Prostate cancer · Bladder cancer · Renal cell carcinoma · Urolithiasis
Introduction
The term artificial intelligence (AI) commonly refers to the computational technologies that mimic or simulate intellec-tual processes typical of human cognitive function, such as reasoning, learning, and problem solving [1]. AI is a branch of computer science and part of a multidisciplinary approach
adopting principles from the fields of mathematics, logic, computation, and biology in an attempt to build intelligent entities often represented as software programs [2, 3]. Given its broad, dynamic, and expanding computational power, AI has been revolutionizing and reshaping our health-care sys-tems, allowing physicians to improve their ability to perform medical tasks [2]. As the medical community´s understand-ing and acceptance of AI grows, so does our imagination in ways to improve diagnostic accuracy, expedite clinical processes, and decrease human resource costs by assisting medical professionals in what once were time-consuming problems [4].
Machine learning (ML) is a subfield of AI involving the development and deployment of dynamic algorithms to analyze data and facilitate the identification of intri-cate patterns [5]. ML tends to improve or ‘learn’ as more
* Rodrigo Suarez-Ibarrola [email protected]
1 Department of Urology, Faculty of Medicine, University of Freiburg-Medical Centre, Hugstetter Str. 55, 79106 Freiburg, Germany
2 Department Augmented Vision, German Research Center for Artificial Intelligence, Kaiserslautern, Germany
World Journal of Urology
1 3
data are incorporated by using decision trees to explicitly learn decision rules [6]. Recent advancements in AI have also been driven by deep learning (DL), which involves the training of artificial neural networks (ANN) with multiple layers on large datasets [7]. An ANN is a collection of indi-vidual information processing units or “artificial neurons” that are arranged and interconnected in network architectural layers to perform computational tasks and recognize com-plex patterns [8]. They are trained with input/output tuples where each input has a specific output assigned to arrange concepts or functions beyond the means of traditional sta-tistical analysis methods. A frequently used ANN that is particularly efficient when applied to pattern recognition in digitized images is the deep convolutional neural network (DCNN). The depth and width of the network determine the complexity and ‘learning potential’ of the network. In health care, ML and DL have been increasingly and successfully applied to preventive medicine, image recognition diagnos-tics, personalized medicine, and clinical decision-making.
The aim of this review article is to address recent ML and DL applications in urolithiasis, renal cell carcinoma, blad-der cancer, and prostate cancer to predict patient outcomes. Their utilization other than in these subfields is not in the scope of this work.
Methodology
A comprehensive review of current literature was performed using the PubMed-Medline database up to May 2019 using the term “urology”, combined with one of the following terms: “machine learning”, “deep learning” and “artificial neural network” in combination with “urolithiasis”, “renal cell carcinoma”, “bladder cancer”, and “prostate cancer”. To capture recent trends in ML and DL applications, the search was limited to articles published within the last 5 years, orig-inally published in English. Review articles and editorials were excluded. Publications relevant to the subject and their cited references were retrieved and appraised independently by two authors (R.S. and A.M.). In accordance with the PRISMA criteria, Fig. 1 was included to delineate our article selection process. After full text evaluation, data were inde-pendently extracted by the authors for further assessment of qualitative and quantitative evidence synthesis. The fol-lowing information was extracted from each study: name of author, journal and year of publication, AI method, number of participants per study, and outcome prediction accuracy.
Urinary stone disease
Despite novel instrumental advancements in urinary stone surgery, decision-making and patient counseling remain a challenge for clinicians. Several investigators have studied
the prognostic role of preoperative parameters on surgical treatment outcomes in terms of stone-free rate (SFR) and the need for secondary procedures [9–13]. An accurate preop-erative outcome prediction would assist urologists in opti-mizing patient selection, choosing ideal treatment options, and personalize patient counselling.
Kadlec et al. developed an ANN that predicted out-comes after various forms of endourologic intervention [9]. Input variables and outcome data from 382 endourologi-cally treated renal units were used to assess SFR (defined by no visible stone on KUB or < 4 mm on CT) and need for secondary procedures. The model predicted SFR with 75.3% sensitivity and 60.4% specificity, and the need for a secondary procedure with 30% sensitivity and 98.3% speci-ficity, yielding a positive and negative predictive value of 60% and 94.2%, respectively. Aminsharifi et al. trained an ANN with pre- and postoperative data from 200 patients and used it to predict various outcomes for 254 patients after percutaneous nephrolithotomy (PCNL) [10]. The accuracy and sensitivity to predict SFR, blood transfusion, and post-PCNL ancillary procedures ranged from 81.0 to 98.2%. Stone burden and morphometry received the highest weight by the ANN as preoperative characteristics affecting post-operative outcomes. Recently, Choo et al. developed and validated a decision-support model using ML algorithms to predict treatment success after a single-session shock wave lithotripsy (SWL) in ureteral stone patients [11]. Using data from 791 patients, a model constructed with 15 variables exhibited 92.3% accuracy to predict SWL outcome. In the decision tree analysis, stone volume, length, and Houns-field units were the top three most important preoperative variables. Similarly, Seckiner et al. collected data from 203 patients and developed an ANN to predict SFR and support SWL treatment planning [14]. ANN analysis demonstrated a prediction accuracy of 99.3% for SFR in the training group, 85.5% in the validation group, and 88.7% in the test group.
Other studies have investigated computer-assisted detec-tion using image features for supporting radiologists in identifying stones. Längkvist et al. developed a DCNN to differentiate ureteral stones from phleboliths in thin slice CT volumes due to their similarity in shape and intensity [15]. The DCNN was evaluated on a database consisting of 465 clinically acquired abdominal CT scans of patients suffering from suspected renal colic. The model achieved 100% sensitivity and an average of 2.68 false positives per patient on a test set of 88 scans. Kazemi et al. derived an ANN for the early detection of kidney stone type and most influential parameters to provide a decision-support system [16]. Information pertaining to 936 patients who underwent treatment for kidney stones was collected and included 42 image features. The model resulted in 97.1% accuracy for predicting kidney stone type and identified gender, calcium level, uric acid condition, hypertension, diabetes, nausea/
World Journal of Urology
1 3
vomiting, flank pain, and urinary tract infection as the most vital parameters for predicting the chance of nephrolithiasis.
A future common goal is for ANNs to be exchanged between institutions to overcome the limitation of having networks trained with data from just one center. To this extent, ML and DL methods hold promise for multi-institu-tional dataset expansion in national registries and the devel-opment of future predictive nomograms. Table 1 provides a summary of studies using ML and DL methods applied in various modalities of urolithiasis therapy.
Renal cell carcinoma
The incidence of renal cell carcinoma (RCC) has steadily increased over the past decades as a result of incidental
small renal mass (SRM) detection via cross-sectional imaging [17]. Surgical series have shown that 20–30% of SRMs ≤ 4 cm are benign, while 20% exhibit potentially aggressive behavior [18]. However, there are currently no clinical or radiographic features that accurately predict histologic analysis. Magnetic resonance imaging (MRI) and computed tomography (CT) have been employed in an attempt to noninvasively differentiate tumors by their degree, pattern, and heterogeneity of enhancement. While promising, these approaches remain suboptimal as clini-cal tools for differentiating SRMs. Recently, powerful ML algorithms are being used to explore complex inter-actions in clinical and imaging data to provide diagno-sis, prognosis, treatment planning and assist in shared decision-making.
Fig. 1 Summary of study selec-tion process
World Journal of Urology
1 3
Tabl
e 1
Mac
hine
and
dee
p le
arni
ng a
pplic
atio
ns in
uro
lithi
asis
AI a
rtific
ial i
ntel
ligen
ce, M
L m
achi
ne le
arni
ng, A
NN
arti
ficia
l neu
ral n
etw
ork,
CN
N c
onvo
lutio
nal n
eura
l net
wor
k, S
WL
shoc
k w
ave
litho
trips
y, fU
RS fl
exib
le u
rete
rosc
opy,
PC
NL
perc
utan
eous
ne
phro
litho
tom
y, S
FR st
one-
free
rate
s, SP
seco
ndar
y pr
oced
ures
, SSD
ston
e-to
-ski
n di
stan
ce, B
MI b
ody
mas
s ind
ex, A
UC
are
a un
der t
he c
urve
, CAD
com
pute
r-aid
ed d
etec
tion
Aut
hor
AI m
etho
dPa
tient
sO
utco
me
pred
ictio
n ac
cura
cy
Kad
lec
et a
l. U
rolit
hias
is 2
014
[8]
Out
com
e pr
edic
tion
in S
WL,
UR
S, P
CN
L38
2 tre
ated
rena
l uni
tsSF
R se
nsiti
vity
75.
3% a
nd sp
ecifi
city
60.
4% (P
PV
75.3
% a
nd N
PV 6
0.4%
)SP
sens
itivi
ty 3
0% a
nd sp
ecifi
city
98.
3% (P
PV 6
0%
and
NPV
94.
2%)
Am
insh
arifi
et a
l. J E
ndou
rol 2
017
[9]
AN
N p
redi
ctio
n of
out
com
e va
riabl
es in
PC
NL
Dat
a fro
m 2
00 p
atie
nts t
o pr
edic
t 254
su
bseq
uent
pat
ient
sO
vera
ll SF
R w
as 7
6.4%
. 54
out o
f 254
(21.
3%)
patie
nts r
equi
red
anci
llary
pro
cedu
res (
SWL
5.9%
, U
RS
10.6
% a
nd re
peat
PC
NL
4.7%
)A
ccur
acy
and
sens
itivi
ty o
f the
syste
m in
pre
dict
ing
diffe
rent
pos
tope
rativ
e va
riabl
es ra
nged
from
81
to 9
8.2%
Cho
o et
al.
J Uro
l 201
8 [1
0]M
L pr
edic
tion
of S
WL
succ
ess i
n ur
eter
al st
ones
791
patie
nts w
ith u
rete
ral s
tone
sA
pply
ing
deci
sion
tree
ana
lysi
s pre
dict
ed tr
eatm
ent
outc
ome
with
92.
29%
acc
urac
y af
ter s
ingl
e se
ssio
n an
d RO
C A
UC
of 0
.951
Man
nil e
t al.
Abd
om R
adio
l 201
8 [1
1]M
L te
xtur
e an
alys
is fo
r suc
cess
ful S
WL
34 u
rinar
y sto
nes i
n in
vitr
o se
tting
Acc
urac
y of
AU
C 0
.84,
sens
itivi
ty o
f 94%
and
spec
i-fic
ity o
f 59%
for s
ucce
ssfu
l sto
ne d
isin
tegr
atio
nM
anni
l et a
l. J U
rol 2
018
[12]
ML
pred
ictio
n of
SW
L su
cces
s in
rena
l sto
nes
51 p
atie
nts w
ith u
ntre
ated
rena
l sto
nes
SWL
succ
ess b
ased
on
SSD
, BM
I, an
d sto
ne si
ze
show
ed A
UC
s of 0
.63,
0.6
8 an
d 0.
58, r
espe
ctiv
ely.
C
ombi
ning
ML
3D te
xtur
e an
alys
is w
ith c
linic
al
varia
bles
impr
oved
acc
urac
y to
an
AU
C o
f 0.8
5 fo
r SS
D, 0
.8 fo
r BM
I, an
d 0.
81 fo
r sto
ne si
zeSe
ckin
er e
t al.
Int B
raz
J Uro
l 201
7 [1
3]A
NN
pre
dict
ing
SFR
afte
r SW
L20
3 pa
tient
s with
rena
l sto
nes
AN
N a
naly
sis d
emon
strat
ed th
at th
e pr
edic
tion
accu
racy
of t
he st
one-
free
rate
was
99.
25%
in th
e tra
inin
g gr
oup,
85.
48%
in th
e va
lidat
ion
grou
p, a
nd
88.7
0% in
the
test
grou
pLä
ngkv
ist e
t al.
Com
put B
iol M
ed 2
018
[14]
CAD
det
ectio
n of
ure
tera
l sto
nes u
sing
CT
volu
mes
on
CN
N46
5 cl
inic
ally
acq
uire
d hi
gh-r
esol
u-tio
n im
ages
of t
he u
rinar
y tra
ctTh
e be
st m
odel
usi
ng 2
.5D
inpu
t dat
a an
d an
atom
i-ca
l inf
orm
atio
n ac
hiev
ed a
sens
itivi
ty o
f 100
% a
nd
an a
vera
ge o
f 2.6
8 fa
lse
posi
tives
per
pat
ient
on
a te
st se
t of 8
8 sc
ans
Kaz
emi e
t al.
Arti
f Int
ell M
ed 2
018
[15]
AN
N fo
r pre
dict
ing
kidn
ey st
one
type
936
patie
nts a
nd 4
2 cl
inic
al fe
atur
esM
odel
pre
dict
ed c
hanc
e of
dev
elop
ing
neph
ro-
lithi
asis
with
97.
1% a
ccur
acy.
Par
amet
ers s
uch
as
gend
er, h
yper
tens
ion,
dia
bete
s, na
usea
, vom
iting
, fla
nk p
ain,
urin
ary
tract
infe
ctio
n, a
nd c
alci
um a
nd
uric
aci
d le
vels
wer
e th
e m
ost v
ital f
or p
redi
ctin
g th
e ch
ance
to d
evel
op n
ephr
olith
iasi
s
World Journal of Urology
1 3
Given the limitations of conventional medical imag-ing, there has been increasing interest in radiomics, which involves automatically extracting quantitative features from medical images. Radiomics may provide a novel approach to develop predictive tools by correlating imaging features to tumor characteristics including histology, tumor grade, genetic patterns and molecular phenotypes, as well as clini-cal outcomes in patients with renal masses. Pixel distribution and pattern-based texture analysis have emerged as practical quantitative methods to build image processing algorithms for the detection of tissue differences that cannot be deter-mined by subjective visual assessments [19].
Several studies have shown that texture analysis has potential in differentiating SRM [20–22]. Yan et al. showed that texture analysis may be a reliable quantitative strategy to differentiate between angiomyolipoma (AML), clear cell RCC (ccRCC), and papillary RCC (pRCC) with an accu-racy in the range of 90.7–100% based on the analysis of three-phase CT scans [23]. Feng et al. achieved a higher accuracy and area under the curve (AUC) of 93.9% using a similar ML strategy [24]. Cui et al. proposed an automatic computer-identification system to differentiate AML from whole-tumor CT images using an over-sampling technique to increase the sample volume of AML [25]. Yu et al. evalu-ated the utility of texture analysis for the distinction of renal tumors, including various RCC subtypes and oncocytoma. The ability of ML to distinguish ccRCC and pRCC from oncocytoma was excellent with AUC of 0.93 and 0.99, respectively [22]. Coy et al. investigated the diagnostic value and feasibility of a DL-based renal lesion classifier to differ-entiate ccRCC from oncocytoma in 179 patients with patho-logically confirmed renal masses on routine four-phase mul-tiple detector CT scans [26]. When using the entire tumor volume, the excretory phase showed the best classification performance with 74.4% accuracy, 85.8% sensitivity, and PPV of 80.1%.
Furthermore, the nuclear grade of a tumor is widely recognized as one of the most important independent prog-nostic factors [27]. Determination of the Fuhrman grade by percutaneous renal biopsy suffers from significant sam-pling bias, making the preoperative recognition of biologi-cal aggressiveness challenging. Studies have shown that ML models constructed from CT imaging texture features can accurately distinguish between ccRCC high and low grades, with accuracy ranging from 0.73 to 0.93 [19, 28–32]. Ding et al. showed high prediction accuracy in identifying ccRCC grade and their results were superior to those obtained from CT image features or the RENAL nephrometry score for high- and low-grade ccRCC predictions [29].
In recent years, biomarkers and multiple gene expression-based signatures have been developed to predict survival and disease prognosis in ccRCC. Li et al. developed a prognostic model based on 15 survival-related genes from The Cancer
Genome Atlas and showed that patients in the model´s high-risk group had significantly worse survival than those in the low-risk group. Risk group was independent of age and sex, but was significantly associated with hemoglobin level, primary tumor size, and grade [33]. Radiogenomics is a field investigating the potential associations between a disease’s imaging features and the underlying genetic pat-terns or molecular phenotype. Kocak et al. evaluated the potential of quantitative CT scan texture analysis to predict the presence of PBRM1 mutations, which is the second most commonly identified mutation in ccRCC, using ANNs and ML algorithms [34]. Overall, the ANN correctly classified 88.2% of ccRCC with regard to PBRM1 mutation status, while the random forest ML algorithm correctly classified 95% of ccRCC.
These are promising results for developing noninvasive imaging biomarkers of histopathologic subtypes, progno-sis, and treatment response. Moreover, they demonstrate that noninvasive ML and DL models constructed from radiomics features have comparable performance to percutaneous renal biopsy in predicting the International Society of Urological Pathology (ISUP) grading. Accurate preoperative nuclear grading may substantially aid in risk assessment, patient stratification, and treatment planning of RCC patients. Table 2 summarizes the most significant findings of MDL applications in the field of RCC.
Bladder cancer
The diagnosis and tumor staging of bladder cancer (BCa) ultimately depend on cystoscopic examination of the bladder and histological evaluation of sampled tissue by transurethral resection (TURB). The main limitation of cystoscopy is its difficulty in discriminating between areas of malignancy and healthy urothelium given the multifocal nature of the dis-ease and inconspicuous but significant lesions such as CIS. However, CT/MRI image-based 3D texture feature analysis of the bladder wall has demonstrated its potential as a non-invasive, image-based strategy to accurately identify hetero-geneous tumor distribution and preoperatively discriminate BCa from normal wall tissue [35]. MRI textural features extracted from cancerous volumes of interest and incorpo-rated into ML models have further demonstrated their ability to preoperatively distinguish low- and high-grade BCa with 83% accuracy [36]. DCNNs have also been used to clas-sify and predict cystoscopic findings with a high degree of accuracy [37]. Such a DL model can be integrated into an AI-aided imaging diagnostic tool to support urologists dur-ing cystoscopic examinations. ‘AI cystoscopy’ may serve as an adjunct during surgical training and medical education to help differentiate benign from malignant lesions using visual evaluation and thereby reduce the number of unnecessary biopsies. A different approach for image-based diagnosis
World Journal of Urology
1 3
Tabl
e 2
Mac
hine
and
dee
p le
arni
ng a
pplic
atio
ns in
rena
l cel
l car
cino
ma
Aut
hor
Patie
nts
AI m
etho
dPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Bek
tas e
t al.
Eur R
adio
l 201
8 [1
8]53
RC
C p
atie
nts w
ith 5
4 pa
thol
ogi-
cally
pro
ven
ccRC
Cs
Vario
us M
L cl
assi
fiers
a27
9 te
xtur
e fe
atur
es e
xtra
cted
from
co
ntra
st-en
hanc
ed C
T sc
ans
Fuhr
man
nuc
lear
gra
de p
redi
ctio
n S
VM
mod
el sc
ored
hig
hest
with
ov
eral
l acc
urac
y, se
nsiti
vity
for
dete
ctin
g hi
gh-g
rade
cc-
RCC
s, sp
ecifi
city
for d
etec
ting
high
-gra
de
cc-R
CC
s, an
d ov
eral
l AU
C o
f 85
.1%
, 91.
3%, 8
0.6%
, and
0.8
60,
resp
ectiv
ely
Koc
ak e
t al.
Eur J
Rad
iol 2
018
[19]
68 R
CC
pat
ient
sA
NN
and
ML-
base
d SV
Ma c
lass
i-fie
rs27
5 te
xtur
e fe
atur
es e
xtra
cted
and
an
alyz
ed fr
om 3
-pha
se C
T im
ages
Dist
ingu
ishi
ng b
etw
een
thre
e m
ajor
su
btyp
es o
f RC
C
AN
N d
iscr
imin
atio
n of
non
-ccR
CC
fro
m c
cRC
C su
btyp
es w
ith a
n ex
ter-
nal v
alid
atio
n ac
cura
cy, s
ensi
tivity
, an
d sp
ecifi
city
of 8
4.6%
, 69.
2%, a
nd
100%
, res
pect
ivel
y S
VM
dis
crim
inat
ion
of p
RCC
from
ot
her R
CC
subt
ypes
with
an
exte
r-na
l val
idat
ion
accu
racy
, sen
sitiv
ity,
and
spec
ifici
ty o
f 69.
2%, 7
1.4%
, and
10
0%, r
espe
ctiv
ely
Kun
apul
i et a
l. J D
igit
Imag
ing
2018
[2
0]15
0 po
st-re
sect
ed p
atie
nts
Vario
us M
L ra
diom
ics-
base
d m
odel
s20
4 te
xtur
e fe
atur
es e
xtra
cted
from
pr
eope
rativ
e 4-
phas
e C
T im
ages
Iden
tifyi
ng m
alig
nanc
y of
rena
l m
asse
s V
isua
l cla
ssifi
catio
n by
exp
erts
ac
hiev
es a
n A
UC
of 0
.65
usin
g pa
thol
ogy
as g
old
stan
dard
. RFG
B
mod
el o
utpe
rform
ed se
vera
l con
-ve
ntio
nal M
L al
gorit
hms (
p = 0.
05)
with
an
accu
racy
of 0
.82,
and
a
prec
isio
n an
d re
call
of 0
.87
Yu
et a
l. A
bdom
Rad
iol 2
017
[21]
119
RCC
pat
ient
sC
T te
xtur
e an
alys
is w
ith M
L-ba
sed
SVM
cla
ssifi
erTu
mor
imag
es fr
om 4
-pha
se C
T m
anua
lly se
gmen
ted
and
43 te
x-tu
re fe
atur
es a
naly
zed
TA fo
r ren
al tu
mor
diff
eren
tiatio
n E
xcel
lent
dis
crim
inat
ors o
f tum
ors
wer
e id
entifi
ed w
ith A
UC
of 0
.91
and
0.93
(p <
0.00
01),
resp
ec-
tivel
y fo
r diff
eren
tiatin
g cc
RCC
fro
m o
ncoc
ytom
a. A
UC
of 0
.99
(p <
0.00
01) f
or d
iffer
entia
ting
pRC
C fr
om o
ncoc
ytom
a an
d an
A
UC
of 0
.92
for d
iffer
entia
ting
onco
cyto
ma
from
oth
er tu
mor
s. Th
e ab
ility
of M
L to
dist
ingu
ish
ccRC
C
from
oth
er tu
mor
s and
pRC
C fr
om
othe
r tum
ors w
as e
xcel
lent
with
A
UC
of 0
.91
and
0.92
, res
pect
ivel
y
World Journal of Urology
1 3
Tabl
e 2
(con
tinue
d)
Aut
hor
Patie
nts
AI m
etho
dPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Yan
et a
l. A
cad
Rad
iol 2
015
[22]
18 A
ML,
18
ccRC
C a
nd 1
4 pR
CC
pa
thol
ogic
ally
pro
ven
patie
nts
AN
N c
lass
ifier
Imag
es fr
om 3
-pha
se C
T sc
an a
na-
lyze
d w
ith T
A so
ftwar
eTA
to d
iffer
entia
te b
etw
een
angi
omy-
olip
oma,
ccR
CC
, and
pRC
C
Exc
elle
nt c
lass
ifica
tion
resu
lts
(0–9
.3%
err
or) w
ere
obta
ined
for a
ll th
ree
grou
ps, i
ndep
ende
ntly
of C
T ph
ase
used
Feng
et a
l. Eu
r Rad
iol 2
018
[23]
58 p
atie
nts w
ith S
RM
: 17
AM
L an
d 41
RC
C
ML-
base
d SM
V c
lass
ifier
for q
uan-
titat
ive
TA42
man
ually
segm
ente
d te
xtur
e fe
atur
es e
xtra
cted
from
3-p
hase
CT
scan
tum
or re
gion
s of i
nter
est
Diff
eren
tiatin
g an
giom
yolip
oma
from
RC
C
Of 4
2 ex
tract
ed fe
atur
es, 1
6 sh
owed
sign
ifica
nt in
terg
roup
di
ffere
nces
(p <
0.05
). Th
e SM
V-
RFE
+ S
MO
TE c
lass
ifier
ach
ieve
d th
e be
st pe
rform
ance
in d
iscr
imi-
natin
g be
twee
n sm
all A
ML
and
RCC
, with
the
high
est a
ccur
acy,
se
nsiti
vity
, spe
cific
it,y
and
AU
C
of 9
3.9%
, 87.
8%, 1
00%
and
0.9
55,
resp
ectiv
ely
Cui
et a
l. A
cta
Rad
iol 2
019
[24]
171
path
olog
ical
ly p
rove
n re
nal
mas
ses
ML-
base
d SM
V e
stab
lishe
d di
f-fe
rent
iatio
n cl
assi
fiers
Text
ure
feat
ures
ext
ract
ed fr
om
who
le tu
mor
imag
es in
3-p
hase
C
T sc
ans
Diff
eren
tiatio
n of
ang
iom
yolip
oma
from
all
RCC
subt
ypes
Diff
eren
tiatin
g A
ML
from
all-
RCC
(AU
C =
0.96
) and
ccR
CC
(A
UC
0.9
7) w
as h
ighe
r tha
n A
ML
from
non
-ccR
CC
(AU
C =
0.89
). M
orph
olog
ical
inte
rpre
tatio
n by
ra
diol
ogist
s ach
ieve
d lo
wer
per
for-
man
ce d
iffer
entia
ting
AM
L fro
m
all-R
CC
(AU
C =
0.06
7), c
cRC
C
(AU
C =
0.68
), an
d no
n-cc
RCC
(A
UC
= 0.
64)
Coy
et a
l. A
bdom
Rad
iol 2
019
[25]
179
patie
nts w
ith 1
79 re
nal l
esio
nsA
NN
trai
ned
with
400
0 ite
ratio
nsFe
atur
e ex
tract
ion
repr
esen
tativ
e of
th
e en
tire
rena
l mas
sD
iffer
entia
ting
onco
cyto
ma
from
cc
RCC
E
xcre
tory
pha
se o
f ent
ire tu
mor
vo
lum
e ac
hiev
ed h
ighe
st 74
.4%
ac
cura
cy, 8
5.8%
sens
itivi
ty, a
nd
80.1
% P
PV. W
hen
com
bine
d w
ith
tum
or m
id-s
lices
a P
PV o
f 82.
5%
World Journal of Urology
1 3
Tabl
e 2
(con
tinue
d)
Aut
hor
Patie
nts
AI m
etho
dPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Hol
dbro
ok e
t al.
JCO
Clin
Can
cer
Info
rm 2
018
[27]
Hist
opat
holo
gic
slid
es o
f 59
ccRC
C
patie
nts
ML-
base
d SM
V a
nd lo
gisti
c re
gres
-si
on m
odel
sIm
age-
base
d te
xtur
e fe
atur
e ex
trac-
tion
for n
ucle
ar a
naly
sis
Qua
ntifi
catio
n of
nuc
lear
ple
omor
phic
pa
ttern
s of c
lear
cel
l RC
C
Obj
ectiv
e an
d fu
lly a
utom
ated
pr
oces
s to
anal
yze
path
olog
ic sl
ides
. A
utom
ated
imag
e cl
assi
ficat
ion
pipe
line
scor
e co
rrel
ated
(R =
0.59
) w
ith e
xisti
ng m
ultig
ene
assa
y-ba
sed
scor
ing
syste
m w
hich
has
dem
on-
strat
ed to
be
a str
ong
indi
cato
r of
prog
nosi
sD
ing
et a
l. Eu
r J R
adio
l 201
8 [2
8]11
4 cc
RCC
pat
ient
s tre
ated
by
par-
tial o
r tot
al n
ephr
ecto
my
ML
mod
els f
or n
on-te
xtur
e an
d te
xtur
e fe
atur
e va
riabl
esSi
x no
n-te
xtur
e fe
atur
es: p
seud
ocap
-su
le, r
ound
ness
, max
. dia
met
er,
intra
tum
oral
arte
ry, e
nhan
cem
ent
valu
e, a
nd re
lativ
e en
hanc
emen
t. 18
4 te
xtur
e fe
atur
es e
xtra
cted
from
3-
phas
e C
T sc
ans
Hig
h-gr
ade
ccRC
C p
redi
ctio
n T
extu
re sc
ore-
base
d m
odel
s with
or
with
out n
on-te
xtur
e fe
atur
es
show
ed a
sign
ifica
nt d
iscr
imin
atio
n of
the
high
- fro
m lo
w-g
rade
ccR
CC
(p
< 0.
05)
Koc
ak e
t al.
AJR
201
9 [2
9]81
ccR
CC
pat
ient
s (56
hig
h an
d 25
lo
w g
rade
)A
NN
and
bin
ary
logi
stic
regr
essi
on
mod
els
744
text
ure
feat
ures
ext
ract
ed fr
om
unen
hanc
ed C
T sc
an im
ages
ccRC
C h
istop
atho
logi
c nu
clea
r gra
de
pred
ictio
n A
NN
cor
rect
ly c
lass
ified
81.
5% o
f cc
RCC
s (A
UC
= 0.
714)
. Log
istic
re
gres
sion
cor
rect
ly c
lass
ified
75.
3%
of a
ll tu
mor
s (A
UC
= 0.
656)
. AN
N
outp
erfo
rmed
logi
stic
regr
essi
on
mod
el (p
< 0.
05)
Lin
et a
l. A
bdom
Rad
iol 2
019
[30]
231
patie
nts w
ith 2
32 p
atho
logi
cally
pr
oven
lesi
ons
Vario
us M
L cl
assi
fiers
134
text
ure
feat
ures
ext
ract
ed fr
om
3-ph
ase
CT
imag
es. S
ingl
e an
d 3-
phas
e C
T im
ages
com
pare
d w
ith
each
oth
er
ccRC
C F
uhrm
an g
rade
pre
dict
ion
ML
mod
el b
ased
on
3-ph
ase
CT
imag
es a
chie
ved
best
diag
nosti
c pe
rform
ance
for d
iffer
entia
ting
low
- fro
m h
igh-
grad
e cc
RCC
(A
UC
= 0.
87),
follo
wed
by
sing
le
neph
roge
nic
phas
e (A
UC
= 0.
84),
corti
com
edul
lary
pha
se
(AU
C =
0.80
), an
d pr
e-co
ntra
st ph
ase
(AU
C =
0.82
)Su
n et
al.
Med
icin
e 20
19 [3
1]22
7 pa
tient
s with
ISU
P co
nfirm
ed
ccRC
C
ML-
base
d SM
V m
odel
1029
imag
ing
feat
ures
ext
ract
ed
from
regi
ons o
f int
eres
t ext
ract
ed
from
3-p
hase
CT
scan
imag
es
Pred
ictio
n of
ISU
P gr
adin
g of
ccR
CC
S
VM
mod
el e
ffect
ivel
y di
sting
uish
es
betw
een
high
- and
low
-gra
de
ccRC
C. A
UC
trai
ning
and
val
ida-
tion
valu
es w
ere
0.88
and
0.9
1,
resp
ectiv
ely
World Journal of Urology
1 3
Tabl
e 2
(con
tinue
d)
Aut
hor
Patie
nts
AI m
etho
dPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Li e
t al.
Med
icin
e 20
18 [3
2]15
surv
ival
- sel
ecte
d ge
nes f
rom
TC
GA
dat
aset
ML-
base
d ra
ndom
fore
sta v
aria
ble
hunt
ing
Ris
k sc
ore
mod
el to
pre
dict
surv
ival
of
ccR
CC
pat
ient
s in
the
TCG
A
data
set
Fifte
en g
ene
expr
essi
on su
rviv
al
pred
ictio
n P
atie
nts i
n th
e hi
gh-r
isk
grou
p ha
d si
gnifi
cant
ly w
orse
ove
rall
surv
ival
th
an th
ose
in th
e lo
w-r
isk
grou
p (p
= 5.
6e−
16);
recu
rren
ce-f
ree
sur-
viva
l sho
wed
a si
mila
r pat
tern
. Ris
k w
as in
depe
nden
t of a
ge a
nd se
x, b
ut
was
sign
ifica
ntly
ass
ocia
ted
with
he
mog
lobi
n le
vel,
prim
ary
tum
or
size
, and
gra
deK
ocak
et a
l. A
JR A
m J
Roen
tgen
ol
2019
[33]
45 c
cRC
C p
atie
nts:
16
with
and
29
with
out P
BR
M1
mut
atio
nM
L-ba
sed
high
-dim
ensi
onal
qua
nti-
tativ
e C
T te
xtur
e an
alys
is82
8 te
xtur
e fe
atur
es e
xtra
cted
per
le
sion
PBR
M1
mut
atio
n pr
edic
tion
in
ccRC
C
759
text
ure
feat
ures
(91.
7%) h
ad
exce
llent
repr
oduc
ibili
ty. A
NN
al
gorit
hm c
orre
ctly
cla
ssifi
ed 8
8.2%
of
ccR
CC
s in
term
s of P
BR
M1
mut
atio
n st
atus
. The
sens
itivi
ty,
spec
ifici
ty, a
nd p
reci
sion
for p
redi
ct-
ing
the
ccRC
Cs w
ith th
e PB
RM1
mut
atio
n w
ere
87.8
%, 8
8.5%
, and
86
.7%
, res
pect
ivel
y. R
ando
m fo
rest
algo
rithm
cor
rect
ly c
lass
ified
95%
of
ccR
CC
s with
AU
C v
alue
of
0.98
7. T
he se
nsiti
vity
, spe
cific
ity,
and
prec
isio
n fo
r pre
dict
ing
the
ccRC
Cs w
ith th
e PB
RM1
mut
atio
n w
ere
94.6
%, 9
5.4%
, and
94.
6%,
resp
ectiv
ely
ccRC
C c
lear
cel
l ren
al c
ell c
arci
nom
a, p
RCC
pap
illar
y RC
C, A
ML
angi
omyo
lipom
a, S
RM s
mal
l ren
al m
ass,
TA te
xtur
e an
alys
is, C
T co
mpu
ted
tom
ogra
phy,
ML
mac
hine
lear
ning
, SM
V su
ppor
t ve
ctor
mac
hine
, AN
N a
rtific
ial n
eura
l net
wor
k, IS
UP
Inte
rnat
iona
l Soc
iety
of U
rolo
gica
l Pat
holo
gy, T
CG
A Th
e C
ance
r Gen
ome
Atla
sa C
lass
ifier
s in
ML
refe
r to
the
inse
rtion
of a
new
obs
erva
tion
into
the
appr
opria
te c
ateg
ory
amon
g ot
hers
that
wer
e ba
sed
on tr
aine
d da
ta se
ts o
f kno
wn
obse
rvat
ions
. Sup
port
vect
or m
achi
nes a
re
supe
rvis
ed le
arni
ng m
etho
ds fo
r cla
ssifi
catio
n th
at le
arn
the
optim
al d
iffer
ence
bet
wee
n fe
atur
es o
f eac
h cl
ass.
Ran
dom
fore
st is
sup
ervi
sed
lear
ning
met
hods
for c
lass
ifica
tion
that
is b
ased
on
deci
sion
tree
s
World Journal of Urology
1 3
has focused on the nanoscale-resolution scanning of cell sur-faces collected from urine [38]. Atomic force microscopy coupled to ML analysis has been shown as a noninvasive method to detect BCa with 94% accuracy when five cells per patient’s urine sample are examined. Moreover, it demon-strated a statistically significant improvement in diagnostic accuracy compared to cystoscopy alone. ML-based methods have been further applied to accurately quantify tumor buds from immunofluorescence-labeled slides of muscle-invasive BCa (MIBC) patients [39]. Tumor budding was found to correlate with TNM staging and patients of all stages were stratified into three new staging criteria based on disease-specific death. Tumor bud quantification through automated slide analysis may provide an alternate staging model with prognostic value for MIBC patients.
ML algorithms have been employed to create recurrence and survival predictive models from imaging and operative data. Patient recurrence and survival at 1, 3 and 5 years after cystectomy was predicted with greater than 70% sensitivity and specificity [40]. Such predictive models may help dic-tate patients´ follow-up schedules, adjuvant treatments, and also provide opportunities for improving care by optimally utilizing operative data collection. ML algorithms used to identify genes at initial presentation that are most predictive of recurrence can be applied as molecular signatures to pre-dict the risk of recurrence within 5 years after TURB [41]. Whole genome profiling from frozen non-muscle-invasive BCa specimens was integrated into a genetic programming algorithm to generate classifier mathematical models for outcome prediction. The model identified 21 key genes that are associated with recurrence from which an optimal three-gene rule [TMEM205 × (NFKBIA × KRT17)] was developed to predict recurrence with 70.6% sensitivity and 66.7% spec-ificity on the test set.
An unmet need in BCa treatment is the early assessment of chemotherapeutic efficacy and prediction of treatment failure at an early phase to reduce unnecessary morbidity, improve patients´ quality of life, and reduce costs. There-fore, the development of accurate predictive models to determine the effectiveness of neoadjuvant chemotherapy is of critical importance in BCa management. Computerized decision support systems (CDSS) have been developed to provide noninvasive, objective, and reproducible decision support for identifying non-responders, so that treatment may be suspended early to preserve their physical condi-tion or to distinguish full responders for organ preservation. Wu et al. compared the performance of different DCNN models and showed that they effectively predicted a bladder lesion´s response to chemotherapy and compared favorably to radiologists´ performance [42]. Cha et al. developed a CT-based CDSS to improve the identification of patients who responded completely to neoadjuvant chemotherapy and found that physicians´ diagnostic accuracy significantly
increased with the aid of CDSS [43]. Thus, computer-aided treatment prediction using DL algorithms may prove to be invaluable to medical professionals as a decision support tool for improving the selection of patients considering blad-der-sparing therapy for MIBC and avoiding adverse effects in non-responders.
Despite several ML and DL research efforts in predict-ing BCa patients’ outcomes, there is scarce adoption of such models in clinical practice. The main challenges ahead before such models can be deployed successfully in a clini-cal setting are the inclusion of standardized parameters, adjusting for the equipment variance, and the collection of multi-institutional data to ensure the generalizability of the models. Once these issues are addressed, ML and DL models can be trained using BCa datasets to accurately pre-dict an individual patient´s outcome using pre-, peri-, and postoperative data. Table 3 summarizes the most significant findings of MDL applications in the field of BCa.
Prostate cancer
There is an unmet need for definitive diagnosis besides tran-srectal imaging and biopsy for men with prostate cancer (PCa). Although a biopsy is necessary for a conclusive diag-nosis, patients with low cancer risk could avoid this proce-dure due to the potential complications that may arise. To achieve this goal, prediction models have been developed to determine patients’ cancer risk on the basis of clinical characteristics. Multilayer ANNs have predicted patients’ prostate biopsy results more accurately while assessing large numbers of variables than traditional statistical meth-ods ranging from non-linear relationships to logistic regres-sion [44, 45]. Despite MRI having improved PCa detection and thereby reducing the number of unnecessary biopsies, excessive variation in its performance and interpretation is a major barrier for global standardization. Computer-aided diagnostic (CAD) systems with DL architecture have been applied to diminish variation in the interpretation of prostatic MRI. Among the advantages of this approach are consist-ent diagnoses, cost-effectiveness, and improved efficiency. Ishioka et al. developed DCNN algorithms that estimate the area in which a targeted biopsy may detect the presence of cancer and in its execution decrease the number of patients mistakenly diagnosed as having cancer [46]. However, other studies have shown no added benefit of radiomic ML when compared with mean apparent diffusion coefficient in differ-entiating benign versus malignant prostate lesions [47]. Pro-vided that the diagnostic precision of CAD systems exceeds that attained by humans, and the pathological diagnosis can be predicted with high accuracy, it may be reasonable to confirm clinically significant PCa solely based on MRI images rather than with biopsy.
World Journal of Urology
1 3
Tabl
e 3
Mac
hine
and
dee
p le
arni
ng a
pplic
atio
ns in
bla
dder
can
cer
Aut
hor
AI m
etho
dPa
tient
sPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Xu
et a
l. In
t J C
AR
S 20
17 [3
4]3D
text
ure
feat
ures
in M
RI
62 B
Ca
patie
nts
VO
Is e
xtra
cted
from
T2
MR
I da
tase
ts3D
text
ure
anal
ysis
to d
iscr
imin
ate
blad
der t
umor
s F
rom
eac
h V
OI,
58 te
xtur
e fe
atur
es
wer
e de
rived
. 37
feat
ures
show
ed
sign
ifica
nt in
terc
lass
diff
eren
ces
(p ≤
0.01
). 29
opt
imal
feat
ures
by
RFE
-SV
M re
sulte
d in
sens
itivi
ty,
spec
ifici
ty, a
ccur
acy,
and
AU
C o
f 0.
90, 0
.85,
0.8
8 an
d 0.
90, r
espe
c-tiv
ely
Zhan
g et
al.
J Mag
n Re
son
Imag
ing
2017
[35]
ML-
base
d SV
M c
lass
ifica
tion
61 B
Ca
patie
nts
DW
I and
AD
C m
aps t
o di
sting
uish
lo
w- f
rom
hig
h-gr
ade
Bca
Text
ural
feat
ures
dist
ingu
ish
low
-gr
ade
from
hig
h-gr
ade
Bca
Fro
m e
ach
VO
I, 10
2 fe
atur
es
wer
e de
rived
. 47
feat
ures
show
ed
sign
ifica
nt in
ter-g
roup
diff
eren
ce
(p ≤
0.05
). 22
opt
imal
feat
ures
by
SV
M-R
FE re
sulte
d in
bes
t pe
rform
ance
in b
ladd
er c
ance
r gr
adin
g w
ith se
nsiti
vity
, spe
cific
ity,
accu
racy
, and
AU
C o
f 0.7
8, 0
.87,
0.
83 a
nd 0
.86,
resp
ectiv
ely
Emin
aga
et a
l. JC
O 2
018
[36]
CN
N m
odel
s47
9 pa
tient
s with
44
cysto
scop
ic
findi
ngs
18,6
81 c
ysto
scop
y im
ages
Dia
gnos
tic c
lass
ifica
tion
of c
ysto
-sc
opic
imag
es O
ver 9
9% a
ccur
acy
for c
lass
ifyin
g im
ages
was
ach
ieve
d by
four
mod
-el
s. 7.
68%
of i
mag
es sh
owin
g bl
ad-
der s
tone
s with
indw
ellin
g ca
thet
er
and
1.43
% o
f div
ertic
ulum
imag
es
wer
e fa
lsel
y cl
assi
fied
Soko
lov
et a
l. PN
AS
2018
[37]
Ato
mic
forc
e m
icro
copy
and
ML
anal
ysis
43 c
ontro
ls a
nd 2
5 ca
ses
Surfa
ce p
aram
eter
s of c
ells
col
lect
ed
from
urin
eD
etec
tion
of b
ladd
er c
ance
r in
void
ed
urin
e cy
tolo
gy u
sing
ato
mic
forc
e m
icro
scop
y 9
4% d
iagn
ostic
acc
urac
y w
hen
exam
inin
g fiv
e ce
lls p
er p
atie
nt´s
ur
ine
sam
ple.
A st
atist
ical
ly
sign
ifica
nt im
prov
emen
t (p <
0.05
) in
dia
gnos
tic a
ccur
acy
of A
FM
(AU
C =
0.92
) com
pare
d to
cys
tos-
copy
(AU
C =
0.77
)
World Journal of Urology
1 3
Tabl
e 3
(con
tinue
d)
Aut
hor
AI m
etho
dPa
tient
sPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Brie
u et
al.
Sci R
ep 2
019
[38]
ML-
base
d m
etho
dolo
gy10
0 M
IBC
pat
ient
sTu
mor
bud
qua
ntifi
catio
n ac
ross
im
mun
ofluo
resc
ence
labe
led
slid
esM
L tu
mor
bud
qua
ntifi
catio
n fo
r TN
M st
agin
g in
MIB
C T
umor
bud
ding
cor
rela
ted
to
TNM
(p =
0.00
09) a
nd p
T st
agin
g (p
= 0.
008)
. Dec
isio
n su
ppor
t tre
e m
odel
repo
rted
that
tum
or b
uddi
ng
was
the
mos
t sig
nific
ant f
eatu
re
(HR
= 2.
59, p
= 0.
009)
Has
nain
et a
l. PL
oS O
ne 2
019
[39]
ML-
base
d SM
V m
odel
3499
pos
t-cys
tect
omy
patie
nts
Pre-
and
ope
rativ
e da
taA
djuv
ant t
hera
py c
ycle
sPo
st-cy
stect
omy
1-, 3
-, an
d 5-
year
re
curr
ence
and
surv
ival
pre
dict
ion
Pat
ient
recu
rren
ce a
nd su
rviv
al 1
, 3,
and
5 y
ears
afte
r cys
tect
omy
can
be p
redi
cted
with
gre
ater
than
70
% se
nsiti
vity
and
spec
ifici
ty.
Path
olog
ic st
age
subg
roup
(org
an
confi
ned,
ext
rave
sica
,l an
d no
de+
) an
d TN
M st
age
wer
e ra
nked
the
mos
t im
porta
nt p
redi
ctor
sB
arts
ch e
t al.
J Uro
l 201
6 [4
0]W
hole
gen
ome
profi
ling
112
NM
IBC
spec
imen
sK
ey g
enes
invo
lved
in B
Ca
recu
r-re
nce
Gen
e ex
pres
sion
pro
filin
g to
pre
dict
re
curr
ent N
MIB
C 2
1 ge
nes p
redi
cted
recu
rren
ce. A
5-
gene
alg
orith
m y
ield
ed 7
7% se
nsi-
tivity
and
85%
spec
ifici
ty to
pre
dict
re
curr
ence
in a
trai
ning
set,
and
69%
an
d 62
%, r
espe
ctiv
ely,
in a
test
set.
Sing
ular
3-g
ene
pred
icte
d re
cur-
renc
e w
ith 8
0% se
nsiti
vity
and
90%
sp
ecifi
city
in a
trai
ning
set,
and
71%
an
d 67
%, r
espe
ctiv
ely,
in a
test
set
Wu
et a
l. To
mog
raph
y 20
19 [4
1]D
L-C
NN
CT
scan
s of 1
23 p
atie
nts w
ith 1
29
canc
er fo
ciEx
tract
ion
of R
OIs
from
segm
ente
d le
sion
s of p
re- a
nd p
osttr
eatm
ent
scan
s
Ass
essm
ent o
f BC
a tre
atm
ent
resp
onse
usi
ng a
DL
appr
oach
The
bas
e D
L-C
NN
with
pre
-trai
ned
wei
ghts
and
tran
sfer
lear
ning
ac
hiev
ed te
st A
UC
of 0
.79.
The
ra
diol
ogist
s’ A
UC
s wer
e 0.
76 a
nd
0.77
. DL-
CN
N p
erfo
rmed
bet
ter
with
pre
-trai
ned
than
rand
omly
in
itial
ized
wei
ghts
World Journal of Urology
1 3
Automated computational methods applied to digital pathology images have shown the ability to overcome Gleason score ambiguity, convey reproducible results, and generate large amounts of data. Arvanati et al. trained a DCNN as Gleason score annotator and used the model’s predictions to assign patients into low-, intermediate-, and high-risk groups, achieving pathology expert-level stratification results [48]. Accurate post-surgical risk stratification is essential to identify patients at high risk of PCa-specific mortality who would benefit from early intervention. Donovan et al. introduced an innovative platform which accurately discriminates between low-, intermediate-, and high-risk PCa, and predicts the likeli-hood of significant clinical failure within 8 years [49]. By combining ML-guided image analysis with biological attributes, the authors provided a risk assignment that is unbiased, broadly applicable, and independent of interpre-tive histology.
While information from clinical registry data assists physicians to make data-driven decisions, there is limited opportunity for patients to access these registries to help them make informed decisions. Auffenberg et al. utilized data from a prospective cancer registry comprising 7543 men diagnosed with PCa to train an ML model to help newly diagnosed men to view predicted treatment deci-sions based on patients with similar characteristics [50]. Their personalized model was highly accurate with age, followed by number of positive cores and Gleason score resulting as the most important variables that influenced patient treatment decisions.
Treatment response prediction using MRI images has been shown as an efficient clinical decision-making tool. Abdollahi et al. developed various radiomics models based on pre- and post-intensity-modulated radiotherapy (IMRT) MRI data for individualized treatment response predic-tion in PCa patients [51]. Their results showed that the features extracted from pre-treatment MRI images pre-dicted early IMRT response with reliable performance. Moreover, Hung et al. presented a novel ML method of processing automated performance metrics to evaluate surgical performance and predict clinical outcomes after robot-assisted radical prostatectomy (RARP) [52]. Their model predicted length of hospital stay, operative time, Foley catheter duration, and urinary continence with over 85% accuracy [53]. In a recent study, Wong et al. used three ML algorithms for the prediction of early biochemi-cal recurrence and showed with an AUC > 0.95 to outper-form traditional statistical regression models [54]. Such methodology can be employed as potentially more accu-rate for identifying patients at risk and equip patients and physicians alike with prognostic information to provide individualized health care (Table 4).
Tabl
e 3
(con
tinue
d)
Aut
hor
AI m
etho
dPa
tient
sPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Cha
et a
l. A
cad
Rad
iol 2
018
[42]
DL-
CN
NC
T sc
ans o
f 123
subj
ects
with
157
M
IBC
foci
Rad
iom
ics f
eatu
re-b
ased
ana
lysi
sC
DSS
for p
redi
ctin
g B
Ca
treat
men
t re
spon
se T
he m
ean
AU
Cs f
or a
sses
smen
t of
T0 d
isea
se w
ere
0.80
for C
DSS
al
one,
0.7
4 fo
r phy
sici
ans n
ot u
sing
C
DSS
, and
0.7
7 fo
r phy
sici
ans
usin
g C
DSS
. The
incr
ease
in th
e ph
ysic
ians
´ per
form
ance
was
stat
is-
tical
ly si
gnifi
cant
(p <
0.05
)
BCa
blad
der c
ance
r, RC
radi
cal c
yste
ctom
y, M
IBC
mus
cle-
inva
sive
bla
dder
can
cer,
NM
IBC
non
-mus
cle-
inva
sive
bla
dder
can
cer,
ML
mac
hine
lear
ning
, DL
deep
lear
ning
, CN
N d
eep
conv
olu-
tiona
l neu
ral n
etw
ork,
SVM
supp
ort v
ecto
r mac
hine
, VO
I vol
ume
of in
tere
st, R
OI r
egio
n of
inte
rest,
DW
I diff
use-
wei
ghte
d im
ages
, AD
C a
ppar
ent d
iffus
ion
coeffi
cien
t, AU
C a
rea
unde
r the
cur
ve,
CD
SS c
ompu
ter d
ecis
ion
supp
ort s
yste
m
World Journal of Urology
1 3
Tabl
e 4
Mac
hine
and
dee
p le
arni
ng a
pplic
atio
ns in
pro
stat
e ca
ncer
Aut
hor
AI m
etho
dPa
tient
sPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Teke
uchi
et a
l. C
UA
J 201
8 [4
3]M
ultil
ayer
AN
N33
4 pa
tient
s mpM
RI t
rans
rect
al
biop
syTw
enty
-two
sele
cted
var
iabl
esPr
osta
te c
ance
r pre
dict
ion
AN
N p
reve
nted
48%
of p
atie
nts w
ith-
out P
Ca
from
und
ergo
ing
pros
tate
bi
opsy
. It m
isse
d 16
% o
f PC
a an
d 6%
of G
S ≥
7. N
PV w
as 7
6% fo
r any
PC
a an
d 94
% w
ith G
S ≥
7Zh
ang
et a
l. O
ncot
arge
t 201
6 [4
4]M
l-bas
ed S
VM
cla
ssifi
er20
5 PC
a pa
tient
sC
linic
opat
holo
gica
l and
MR
imag
ing
data
sets
Imag
e-ba
sed
clin
ical
out
com
e pr
edic
-tio
n: S
VM
vs.
logi
stic
regr
essi
on a
nd
D`A
mic
o ris
k gr
oups
SV
M h
ad si
gnifi
cant
ly h
ighe
r AU
C
(0.9
59 v
s. 0.
886,
p =
0.00
7), s
ensi
tiv-
ity (9
3.3%
vs 8
3.3%
, p =
0.02
5),
spec
ifici
ty (9
1.7%
vs.
77.2
%,
p = 0.
009)
and
acc
urac
y (9
2.2%
vs
. 79%
, p =
0.00
6) th
an lo
gisti
c re
gres
sion
ana
lysi
s. D
’Am
ico
sche
me
effec
tivel
y im
prov
ed b
y ad
ding
MR
I-de
rived
var
iabl
esIs
hiok
a et
al.
BJU
I 201
8 [4
5]C
NN
alg
orith
m33
5 pa
tient
s with
MR
I and
ext
ende
d sy
stem
atic
pro
stat
e bi
opsy
MR
imag
es la
belle
d as
‘can
cer’
or
‘no
canc
er’
Com
pute
r-aid
ed d
iagn
osis
alg
orith
m
for d
etec
ting
PCa
on M
RI
Gra
phic
s pro
cess
ing
unit
requ
ired
5.5
h to
lear
n to
ana
lyze
2 m
illio
n im
ages
. Alg
orith
ms r
equi
red
30 m
s/im
age
for e
valu
atio
n an
d sh
owed
A
UC
of 0
.645
and
0.6
36. 1
6/17
and
7/
17 p
atie
nts m
istak
enly
dia
gnos
ed
as h
avin
g PC
a in
two
valid
atio
n da
tase
tsB
onek
amp
et a
l. Eu
r Rad
iol 2
018
[46]
ML-
base
d ra
diom
ics m
odel
s31
6 m
en w
ith M
RI-
trans
rect
al U
S fu
sion
bio
psy
Lesi
on se
gmen
tatio
n an
d ra
diom
ics
anal
ysis
Rad
iom
ic c
hara
cter
izat
ion
of p
rost
ate
lesi
ons w
ith M
RI:
ML
vs A
DC
co
mpa
rison
AU
C fo
r the
mea
n A
DC
(AU
C
glob
al =
0.84
; AU
C zon
e-sp
ecifi
c ≤ 0.
87)
vs th
e R
ML
(AU
C glo
bal =
0.88
, p =
0.17
6; A
UC
zone-
spec
ific ≤
0.89
, p ≥
0.49
3) sh
owed
no
sign
ifica
ntly
di
ffere
nt p
erfo
rman
ce
World Journal of Urology
1 3
Tabl
e 4
(con
tinue
d)
Aut
hor
AI m
etho
dPa
tient
sPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Arv
aniti
et a
l. Sc
i Rep
201
8 [4
7]C
NN
alg
orith
mSy
stem
trai
ned
on 6
41 p
atie
nts a
nd
teste
d on
245
pat
ient
sA
nnot
ated
dat
aset
of P
Ca
tissu
e m
icro
arra
ysA
utom
ated
Gle
ason
scor
e gr
adin
g A
gree
men
ts b
etw
een
the
mod
el a
nd
each
pat
holo
gist
wer
e 0.
75 a
nd 0
.71.
M
odel
´s G
S ss
ignm
ent a
chie
ved
path
olog
y ex
pert-
leve
l stra
tifica
tion
of p
atie
nts i
nto
prog
nosti
cally
dis
-tin
ct g
roup
s on
the
basi
s of d
isea
se-
spec
ific
surv
ival
Don
ovan
et a
l. Pr
osta
te C
ance
r Pro
s-ta
tic D
is 2
018
[48]
ML/
mic
rosc
opic
pat
tern
ana
lysi
s89
2 po
st ra
dica
l pro
stat
ecto
my
patie
nts
Dig
ital i
mag
e an
alys
isA
utom
ated
Gle
ason
scor
e gr
ade
and
mol
ecul
ar p
rofil
e In
trai
ning
, mod
el p
redi
cted
sign
ifi-
cant
clin
ical
failu
re w
ith C
-inde
x of
0.
82 a
nd H
R 6
.7. R
esul
ts c
onfir
med
in
val
idat
ion
with
C-in
dex
0.77
and
H
R 5
.4 fo
r dis
crim
inat
ing
low
from
in
term
edia
te h
igh-
risk
pro
stat
e ca
ncer
Auff
enbe
rg e
t al.
Eur U
rol 2
018
[49]
Ran
dom
fore
st M
L7,
543
PCa
patie
nts
3413
(45%
) und
erw
ent R
P, 2
289
(30%
) AS,
128
0 (1
7%) R
T, 4
22
(5.6
%) A
DT,
and
139
(1.8
%) W
W
ML
mod
el to
info
rm p
atie
nts o
f PC
a tre
atm
ents
cho
sen
by si
mila
r men
Per
sona
lized
pre
dict
ion
for p
atie
nts
in th
e va
lidat
ion
coho
rt w
as h
ighl
y ac
cura
te (A
UC
= 0.
81).
Age
, num
ber
of p
ositi
ve c
ores
, and
Gle
ason
scor
e w
ere
the
mos
t infl
uent
ial v
aria
bles
in
fluen
cing
pat
ient
trea
tmen
t dec
i-si
ons
Abd
olla
hi e
t al.
Rad
iol M
ed 2
019
[50]
ML-
base
d ra
diom
ics m
odel
s33
PC
a pa
tient
s inc
lude
dR
adio
mic
feat
ures
ext
ract
ed fr
om
MR
I T2W
and
AD
C im
ages
Pred
ictio
n of
inte
nsity
-mod
ulat
ed
radi
atio
n th
erap
y re
spon
se T
wen
ty h
ighl
y co
rrel
ated
radi
omic
s fe
atur
es w
ith IM
RT re
spon
se. F
or
GS
pred
ictio
n, T
2W ra
diom
ics m
od-
els w
ere
foun
d to
be
mor
e pr
edic
tive
(mea
n A
UC
0.7
39),
whi
le fo
r sta
ge
pred
ictio
n, A
DC
mod
els h
ad a
hig
her
perfo
rman
ce (A
UC
0.6
75)
Hun
g et
al.
J End
ouro
l 201
8 [5
1]3
ML
algo
rithm
s75
RA
RP
case
sA
utom
ated
per
form
ance
met
rics a
nd
hosp
ital l
engt
h of
stay
RA
RP
perfo
rman
ce a
nd o
utco
me
pred
ictio
n A
lgor
ithm
pre
dict
ed le
ngth
of s
tay
with
87.
2% a
ccur
acy.
Pat
ient
out
-co
mes
pre
dict
ed b
y th
e al
gorit
hm
had
sign
ifica
nt a
ssoc
iatio
n w
ith
‘gro
und
truth
’ in
surg
ery
time,
leng
th
of st
ay a
nd F
oley
dur
atio
n
World Journal of Urology
1 3
Tabl
e 4
(con
tinue
d)
Aut
hor
AI m
etho
dPa
tient
sPa
ram
eter
sPr
edic
ted
outc
ome
accu
racy
Hun
g et
al.
BJU
I 201
9 [5
2]D
L m
odel
(Dee
pSur
v)10
0 R
AR
P ca
ses
Aut
omat
ed p
erfo
rman
ce m
etric
s, pa
tient
clin
icop
atho
logi
cal a
nd
cont
inen
ce d
ata
Pred
ictio
n of
urin
ary
cont
inen
ce re
cov-
ery
afte
r RA
RP
Con
tinen
ce w
as a
ttain
ed in
79%
afte
r m
edia
n of
126
day
s. D
L m
odel
ac
hiev
ed C
-inde
x of
0.6
in p
redi
ctin
g co
ntin
ence
. Aut
omat
ed p
erfo
rman
ce
met
rics w
ere
rank
ed h
ighe
r by
the
mod
el th
an c
linic
opat
holo
gica
l fe
atur
esW
ong
et a
l. B
JUI 2
019
[53]
3 M
L al
gorit
hms
338
RA
RP
case
s19
diff
eren
t tra
inin
g va
riabl
esEa
rly b
ioch
emic
al re
curr
ence
pre
dic-
tion
afte
r RA
RP
All
thre
e M
L m
odel
s out
perfo
rmed
th
e co
nven
tiona
l sta
tistic
al re
gres
-si
on m
odel
(AU
C 0
.865
). A
ccur
acy
pred
ictio
n sc
ores
for K
-nea
rest
neig
hbor
, ran
dom
fore
st tre
e, a
nd
logi
stic
regr
essi
on w
ere
0.97
6, 0
.953
an
d 0.
976,
resp
ectiv
ely
PCa
pros
tate
can
cer,
RP ra
dica
l pro
stat
ecto
my,
AS
activ
e su
rvei
llanc
e, R
T ra
diot
hera
py, A
DT
andr
ogen
dep
rivat
ion
ther
apy,
WW
wat
chfu
l wai
ting,
AN
N a
rtific
ial n
eura
l net
wor
k, C
NN
con
volu
-tio
nal n
eura
l net
wor
k, G
S G
leas
on sc
ore,
IMRT
inte
nsity
-mod
ulat
ed ra
diat
ion
ther
apy,
AD
C a
ppar
ent d
iffus
ing
coeffi
cien
t, RA
RP ro
bot-a
ssist
ed ra
dica
l pro
stat
ecto
my,
ML
mac
hine
lear
ning
, DL
deep
lear
ning
, SVM
supp
ort v
ecto
r mac
hine
World Journal of Urology
1 3
ML and DL limitations
AI technologies have been attracting substantial attention in urology; however, their real-life implementation still faces obstacles. Several limitations exist in most studies applying ML and DL methods to urological diseases. First, the varia-bility in study design, algorithms employed, training features used, and observed end points make it difficult to perform quantitative analysis. Second, most algorithms in these stud-ies were validated with their dataset; therefore, they lack external validation and the generalizability of their results across other datasets is not applicable. Third, further algo-rithm development and research are particularly required in the field of urolithiasis to outperform conventional statisti-cal methods as observed in urooncological investigations to reduce procedural costs and maximize patient outcomes. Lastly, some studies did not compare AI with conventional statistical analysis, since these methods only allow a limited number of training features, whereas AI can process big data and can thus be trained with a greater number of training features. For this reason, a comparison between any two techniques is challenging [55].
Future directions
Future research should focus on the construction of larger medical databases and further development of AI tech-niques. Once developed, the use of improved algorithms should not require large computer centers, but be per-formed on mobile devices or by access to cloud services. Specialized AI-based software for image-guided, real-time, intraoperative decisions will require appropriate regulatory approvals to function with robotic platforms and expand to operating rooms worldwide [56]. Issues remain regarding the trustworthiness of a computer´s diagnosis and that pro-gramming biases do not interfere with diagnoses. Human intuition, experience, and common sense will remain to play a crucial role in future AI developments to ensure that these systems are operating as intended and to deal with undesired consequences in a timely fashion.
Conclusion
The predictive precision of ML and DL will continue to provide and enhance personalized medicine with the further inclusion of data and model retraining. Larger patient data-sets and electronic medical records can be semi-automated to provide instant predictive analytics that can be used to obtain insights into a variety of disease processes. Predic-tive accuracy, however, is highly dependent on efficient data integration obtained from different sources to enable it to be generalized. Although the shared decision-making will not
be replaced by these models, it may complement the infor-mation patients obtain from traditional methods. While this is the beginning and further validation is required, there are limitless future applications for artificial intelligence in the field of urology.
Author contributions Project development: RS and AM. Literature review and data extraction: RS. Manuscript drafting: RS, GR, and AM. Manuscript editing: SH, CG, and AM.
Funding This research received no financial or other support.
Compliance with ethical standards
Conflict of interest The authors declare no conflicts of interest.
Human and animal rights statement This research did not involve human subjects or animals.
Ethical approval As this is a review of the literature, no ethical approval was necessary.
References
1. Nuffield Council on Bioethics (2018) Bioethics briefing notes: artificial intelligence (AI) in healthcare and research. https ://nuffi eldbi oethi cs.org/wp-conte nt/uploa ds/Artifi cial -Intel ligen ce-AI-in-healt hcare -and-resea rch.pdf. Accessed 21 Dec 2018
2. Frankish K, Ramsey WM (eds) (2014) Introduction. The Cam-bridge handbook of artificial intelligence. Cambridge University Press, Cambridge, pp 1–14
3. Stuart R, Norvig P (eds) (2010) Artificial intelligence—a modern approach, 3rd edn. Prentice Hall, Upper Saddle River
4. Tran BX et al (2019) Global evolution of research in artificial intelligence in health and medicine: a bibliometric study. J Clin Med 8(3):360
5. Goldenberg SL, Nir G, Salcudean SE (2019) A new era: artificial intelligence and machine learning in prostate cancer. Nat Rev Urol 16(7):391–403
6. Yu KH, Beam AL, Kohane IS (2018) Artificial intelligence in healthcare. Nat Biomed Eng 2(10):719–731
7. Curran Associates Inc. (2014) Advances in neural information processing systems 26: 27th annual conference on neural informa-tion processing systems 2014, December 8–13. Curran Associates Inc., vol 1
8. Abbod MF et al (2007) Application of artificial intelligence to the management of urological cancer. J Urol 178(4 Pt 1):1150–1156
9. Kadlec AO et al (2014) Nonlinear logistic regression model for outcomes after endourologic procedures: a novel predictor. Uro-lithiasis 42(4):323–327
10. Aminsharifi A et al (2017) Artificial neural network system to pre-dict the postoperative outcome of percutaneous nephrolithotomy. J Endourol 31(5):461–467
11. Choo MS et al (2018) A prediction model using machine learning algorithm for assessing stone-free status after single session shock wave lithotripsy to treat ureteral stones. J Urol 200(6):1371–1377
12. Mannil M et al (2018) Prediction of successful shock wave litho-tripsy with CT: a phantom study using texture analysis. Abdom Radiol (NY) 43(6):1432–1438
World Journal of Urology
1 3
13. Mannil M et al (2018) Three-dimensional texture analysis with machine learning provides incremental predictive information for successful shock wave lithotripsy in patients with kidney stones. J Urol 200(4):829–836
14. Seckiner I et al (2017) A neural network-based algorithm for predicting stone-free status after ESWL therapy. Int Braz J Urol 43(6):1110–1114
15. Langkvist M et al (2018) Computer aided detection of ureteral stones in thin slice computed tomography volumes using Convo-lutional Neural Networks. Comput Biol Med 97:153–160
16. Kazemi Y, Mirroshandel SA (2018) A novel method for predict-ing kidney stone type using ensemble learning. Artif Intell Med 84:117–126
17. Richard PO et al (2015) Renal tumor biopsy for small renal masses: a single-center 13-year experience. Eur Urol 68(6):1007–1013
18. Mir MC et al (2018) Role of active surveillance for localized small renal masses. Eur Urol Oncol 1(3):177–187
19. Bektas CT et al (2019) Clear cell renal cell carcinoma: machine learning-based quantitative computed tomography texture analysis for prediction of Fuhrman nuclear grade. Eur Radiol 29(3):1153–1163
20. Kocak B et al (2018) Textural differences between renal cell car-cinoma subtypes: machine learning-based quantitative computed tomography texture analysis with independent external validation. Eur J Radiol 107:149–157
21. Kanapuli G et al (2018) A decision-support tool for renal mass classification. J Digit Imaging 31(6):929–939
22. Yu H et al (2017) Texture analysis as a radiomic marker for dif-ferentiating renal tumors. Abdom Radiol (NY) 42(10):2470–2478
23. Yan L et al (2015) Angiomyolipoma with minimal fat: differen-tiation from clear cell renal cell carcinoma and papillary renal cell carcinoma by texture analysis on CT images. Acad Radiol 22(9):1115–1121
24. Feng Z et al (2018) Machine learning-based quantitative texture analysis of CT images of small renal masses: differentiation of angiomyolipoma without visible fat from renal cell carcinoma. Eur Radiol 28(4):1625–1633
25. Cui EM et al (2019) Differentiation of renal angiomyolipoma without visible fat from renal cell carcinoma by machine learn-ing based on whole-tumor computed tomography texture features. Acta Radiol 60(11):1543–1552
26. Coy H et al (2019) Deep learning and radiomics: the utility of Google TensorFlow Inception in classifying clear cell renal cell carcinoma and oncocytoma on multiphasic CT. Abdom Radiol 44(6):2009–2020
27. Minardi D et al (2005) Prognostic role of Fuhrman grade and vascular endothelial growth factor in pT1a clear cell carcinoma in partial nephrectomy specimens. J Urol 174(4 Pt 1):1208–1212
28. Holdbrook DA et al (2018) Automated renal cancer grading using nuclear pleomorphic patterns. JCO Clin Cancer Inform 2:1–12
29. Ding J et al (2018) CT-based radiomic model predicts high grade of clear cell renal cell carcinoma. Eur J Radiol 103:51–56
30. Kocak B et al (2019) Unenhanced CT texture analysis of clear cell renal cell carcinomas: a machine learning-based study for predicting histopathologic nuclear grade. AJR Am J Roentgenol 212:W1–W8
31. Lin F et al (2019) CT-based machine learning model to predict the Fuhrman nuclear grade of clear cell renal cell carcinoma. Abdom Radiol 44(7):2528–2534
32. Sun X et al (2019) Prediction of ISUP grading of clear cell renal cell carcinoma using support vector machine model based on CT images. Medicine (Baltimore) 98(14):e15022
33. Li P et al (2018) Fifteen-gene expression based model predicts the survival of clear cell renal cell carcinoma. Medicine (Baltimore) 97(33):e11839
34. Kocak B et al (2019) Radiogenomics in clear cell renal cell carci-noma: machine learning-based high-dimensional quantitative CT texture analysis in predicting PBRM1 mutation status. AJR Am J Roentgenol 212(3):W55–W63
35. Xu X et al (2017) Three-dimensional texture features from intensity and high-order derivative maps for the discrimination between bladder tumors and wall tissues via MRI. Int J CARS 12(4):645–656
36. Zhang X et al (2017) Radiomics assessment of bladder cancer grade using texture features from diffusion-weighted imaging. J Magn Reson Imaging 46(5):1281–1288
37. Eminaga O et al (2018) Diagnostic classification of cystoscopic images using deep convolutional neural networks. JCO Clin Can-cer Inform 2:1–8
38. Sokolov I et al (2018) Noninvasive diagnostic imaging using machine-learning analysis of nanoresolution images of cell surfaces: detection of bladder cancer. Proc Natl Acad Sci USA 115(51):12920–12925
39. Brieu N et al (2019) Automated tumour budding quantification by machine learning augments TNM staging in muscle-invasive bladder cancer prognosis. Sci Rep 9(1):5174
40. Hasnain Z et al (2019) Machine learning models for predict-ing post-cystectomy recurrence and survival in bladder cancer patients. PLoS ONE 14(2):e0210976
41. Bartsch G Jr et al (2016) Use of artificial intelligence and machine learning algorithms with gene expression profiling to predict recurrent nonmuscle invasive urothelial carcinoma of the blad-der. J Urol 195(2):493–498
42. Wu E et al (2019) Deep learning approach for assessment of blad-der cancer treatment response. Tomography 5(1):201–208
43. Cha KH et al (2018) Diagnostic accuracy of CT for prediction of bladder cancer treatment response with and without computerized decision support. Acad Radiol 26:1137–1145
44. Takeuchi T et al (2019) Prediction of prostate cancer by deep learning with multilayer artificial neural network. Can Urol Assoc J 13(5):E145–E150
45. Zhang YD et al (2016) An imaging-based approach predicts clini-cal outcomes in prostate cancer through a novel support vector machine classification. Oncotarget 7(47):78140–78151
46. Ishioka J et al (2018) Computer-aided diagnosis of prostate can-cer on magnetic resonance imaging using a convolutional neural network algorithm. BJU Int 122(3):411–417
47. Bonekamp D et al (2018) Radiomic machine learning for char-acterization of prostate lesions with MRI: comparison to ADC values. Radiology 289(1):128–137
48. Arvaniti E et al (2018) Automated Gleason grading of prostate cancer tissue microarrays via deep learning. Sci Rep 8(1):12054
49. Donovan MJ et al (2018) Development and validation of a novel automated Gleason grade and molecular profile that define a highly predictive prostate cancer progression algorithm-based test. Prostate Cancer Prostatic Dis 21(4):594–603
50. Auffenberg GB et al (2019) askMUSIC: leveraging a clinical reg-istry to develop a new machine learning model to inform patients of prostate cancer treatments chosen by similar men. Eur Urol 75(6):901–907
51. Abdollahi H et al (2019) Machine learning-based radiomic models to predict intensity-modulated radiation therapy response, Gleason score and stage in prostate cancer. Radiol Med 124(6):555–567
52. Hung AJ et al (2018) Utilizing machine learning and auto-mated performance metrics to evaluate robot-assisted radical prostatectomy performance and predict outcomes. J Endourol 32(5):438–444
53. Hung AJ et al (2019) A deep-learning model using automated performance metrics and clinical features to predict urinary con-tinence recovery after robot-assisted radical prostatectomy. BJU Int 124(3):487–495
World Journal of Urology
1 3
54. Wong NC et al (2019) Use of machine learning to predict early biochemical recurrence after robot-assisted prostatectomy. BJU Int 123(1):51–57
55. Chen J et al (2019) Current status of artificial intelligence applica-tions in urology and their potential to influence clinical practice. BJU Int [Epub ahead of print]
56. Goldenberg SL, Nir G, Salcudean SE (2019) A new era: artificial intelligence and machine learning in prostate cancer. Nat Rev Urol 16(7):391–403
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.