Depósi o de in es igación de la Uni e sidad de Se illa
h ps://idus.us.es/
Es a es la e sión acep ada del a ículo publicado en: Jou nal o Clinical
Ul asound (2025)
This is a accep ed manusc ip o a pape published in: 25 Ap il 2025
DOI: 10.1002/jcu.24050
Copy igh : © 2025 Wiley Pe iodicals LLC.
El acceso a la e sión publicada del a ículo puede eque i la susc ipción de la e is a.
Access o he published e sion may equi e subsc ip ion.
“This is he pee e iewed e sion o he ollowing a icle: 2025. Jou nal o Clinical
Ul asound , Ma ín-Mo án M, Bo e o-Gonzalez C, Fe nández-Gomez A, Fe nández-
Palacín F, Sainz-Bueno JA. A i icial In elligence o De e mine Co ec Midsagi al
Plane in Dynamic T anspe ineal Ul asound. J Clin Ul asound.
h ps://doi.o g/10.1002/jcu.24050 which has been published in inal o m a
10.1002/jcu.24050. This a icle may be used o non-comme cial pu poses in
acco dance wi h Wiley Te ms and Condi ions o Use o Sel -A chi ed Ve sions. This
a icle may no be enhanced, en iched o o he wise ans o med in o a de i a i e wo k,
wi hou exp ess pe mission om Wiley o by s a u o y igh s unde applicable
legisla ion. Copy igh no ices mus no be emo ed, obscu ed o modi ied. The a icle
mus be linked o Wiley’s e sion o eco d on Wiley Online Lib a y and any
embedding, aming o o he wise making a ailable he a icle o pages he eo by hi d
pa ies om pla o ms, se ices and websi es o he han Wiley Online Lib a y mus be
p ohibi ed."
1
A i icial in elligence o de e mine co ec midsagi al plane in 1
dynamic anspe ineal ul asound 2
3
A icle ype: O iginal Resea ch 4
5
Au ho s: 6
José An onio Ga cía-Mejido1, Juan Galán-Paez2, Da id Solis-Ma ín2, Ma ina Ma ín-7
Mo án1, Ca lo a Bo e o-Gonzalez1, Al onso Fe nández-Gomez1, Fe nando Fe nández-8
Palacín3 , José An onio Sainz-Bueno1 9
10
A ilia ion: 11
1Depa men o su ge y, Facul y o Medicine, Uni e si y o Se ille, Spain. 12
2Depa men o compu e science and a i icial in elligence, Facul y o Ma hema ics, 13
Uni e si y o Se ille, Spain. 14
3 Depa men o s a is ics and ope a ional esea ch, Uni e si y o Cadiz, Cadiz, Spain. 15
16
Co esponding au ho : 17
José An onio Ga cía Mejido., Depa men o Obs e ics and Gynecology Valme 18
Uni e si y Hospi al, Se ille, Spain. Depa men o su ge y, Facul y o Medicine, 19
Uni e si y o Se ille, Spain. E-mail: [email p o ec ed]. Phone: +34955015385 20
21
Da a a ailabili y s a emen : Da a a ailable on eques om he au ho s. 22
23
Funding s a emen : None. 24
25
Con lic o in e es disclosu e： The au ho s ha e no con lic s o in e es o disclose 26
E hics o app o al s a emen ：The s udy was app o ed by Andalucia’s Boa d o 27
Biomedicine E hics Commi ee, wi h code SICEIA-2024-001928. The s udy was 28
2
conduc ed in acco dance wi h he Decla a ion o Helsinki. All pa ien s ga e hei w i en 29
in o med consen be o e s a ing he s udy. 30
31
Pe mission o ep oduce ma e ial om o he sou ces: None. 32
33
Clinical ial egis a ion： None. 34
35
Wo d coun : 3236 wo ds including abs ac bu excluding e e ences, ables, and 36
igu es. 37
38
39
3
A i icial in elligence o de e mine co ec midsagi al plane in 40
dynamic anspe ineal ul asound 41
42
Abs ac : 43
Pu pose:To c ea e and alida e a machine lea ning(ML) model ha allows iden i ying 44
he co ec cap u e o he midsagi al plane in a dynamic ul asound s udy, as well as 45
es ablishing i s conco dance wi h a senio explo e and a junio explo e . 46
47
Me hods:Obse a ional and p ospec i e s udy wi h 90 pa ien s wi hou pel ic loo 48
pa hology. Each pa ien was gi en an ul asound ideo whe e he midsagi al plane o he 49
pel ic loo was eco ded a es and du ing he Valsal a maneu e .A segmen a ion model 50
was used ha was ained on a p e iously published a icle, gene a ing he segmen a ions 51
o he 90 new ideos o c ea e he model.The algo i hm selec ed o build he model in his 52
p ojec was XGBoos (G adien Boos ing).To ob ain a abula da ase on which o ain 53
he model ea u e enginee ing was ca ied ou on he aw segmen a ion da a.The 54
conco dance o he model, o a junio examine and a senio examine ,wi h he expe 55
examine was s udied using he kappa index. 56
57
Resul s:The i s 60 ideos we e used o ain he model and he las 30 ideos we e 58
ese ed o he es se .The model p esen ed a kappa index 0.930(p<0.001) wi h e y 59
good ag eemen o de ec ion o he co ec midsagi al plane. The junio explo e 60
p esen ed a e y good ag eemen (kappa index=0.930(p<0.001)).The senio explo e 61
p esen ed a kappa index 0.789(p<0.001)(good ag eemen ) o de ec ion o he co ec 62
midsagi al plane. 63
64
Conclusion: We ha e de eloped a model ha allows de e mining he co ec midsagi al 65
plane cap u ed h ough dynamic anspe ineal ul asound wi h a le el o ag eemen 66
compa able o o g ea e han ha o a junio o senio examine , using expe examine 67
assessmen as he gold s anda d. 68
69
4
Keywo ds: Machine lea ning, pel ic loo , ul asonog aphy, g adien boos ing, 70
XGBoos , a i icial in elligence, le a o ani muscle. 71
72
5
In oduc ion. 73
74
Pel ic loo ul asound has ep esen ed a b eak h ough in he s udy and diagnosis o 75
pel ic loo dys unc ions. One o he cha ac e is ics o anspe ineal pel ic loo 76
ul asound is ha i is s anda dized om he midsagi al plane [1] o wo-dimensional 77
ul asound. F om he midsagi al plane we can simul aneously s udy he pubic symphysis, 78
he u e h a, he u ina y bladde , he agina, he u e us, he anal canal, he ec um and he 79
le a o ani muscle [1] and he e o e mos o he pa hology ha co e s each compa men . 80
Co ec cap u e o he midsagi al plane equi es lea ning ha some imes depends on he 81
skills o he examine . In addi ion, anspe ineal ul asound is associa ed wi h manual 82
measu emen s, which in ol e a ime consump ion in he consul a ion and depend on he 83
expe ience o he examine , which will di ec ly in luence he a ia ions in he sco e [2]. 84
To hese aspec s we mus add ha pel ic loo ul asound allows a dynamic s udy o he 85
di e en s uc u es, making i s s udy mo e complica ed, especially in he case o pel ic 86
o gan p olapse. In ac , i has been desc ibed ha in o de no o lose he midsagi al plane 87
du ing he Valsal a maneu e , he o a ional mo emen o he ansduce mus be 88
a oided, p ese ing he o iginal alignmen o he pel ic loo in he ul asound image [3]. 89
90
On he o he hand, he de elopmen o a i icial in elligence (AI) in he ield o 91
u ogynecology is p og essi ely expanding, showing i s use ulness in iden i ying di e en 92
s uc u es o he pel ic loo [4-7]. AI allows a compu e p og am o pe o m easoning 93
p ocesses simila o he human b ain, wi h deep lea ning (DL) being a subca ego y ha 94
can ecognize medical images, classi y hem, and de ec objec s [8]. Cu en ly, CNN 95
(con olu ional neu al ne wo k) has been used in pel ic loo ul asound o analysis o 96
he le a o ani muscle [5, 8-10], o measu emen s o he le a o hia us [5,6,11,12], 97
measu emen o he u ogeni al hia us [13], he assessmen o u odynamic s ess 98
incon inence wi h ul asound [4] and o he s udy o pel ic o gan p olapse [14]. 99
Howe e , all hese s udies a e based on s a ic images, and o ob ain he ul asound 100
diagnosis o he di e en pel ic loo dys unc ions we need a dynamic ul asound s udy 101
[3]. Recen ly, i has been desc ibed ha i is possible o apply deep lea ning o iden i y 102
he di e en pel ic loo o gans in a dynamic ul asound s udy in he co ec ly cap u ed 103
midsagi al plane [15]. Howe e , some imes i is possible ha he midsagi al plane is no 104
well de ined, ei he due o he condi ion o he pa ien 's issues o due o he lack o 105
6
aining o he examine who pe o ms he echnique. These de ec s in he cap u e o he 106
image can lead o diagnos ic e o s and in e p e a ion o he ul asound. Based on hese 107
aspec s, we conside ha AI can be o g ea help i i can help us de ine he co ec 108
midsagi al plane o he ul asound s udy o he pel ic loo . The e o e, ou objec i e is 109
o c ea e and alida e a p edic i e model ha allows us o iden i y he co ec midsagi al 110
plane in a dynamic ul asound s udy, as well as es ablish i s conco dance wi h a senio 111
examine and a junio examine . 112
113
114
Ma e ials 115
116
An obse a ional and p ospec i e s udy was conduc ed wi h 90 pa ien s. The included 117
pa ien s had no pel ic loo pa hology and we e ec ui ed consecu i ely in he gene al 118
gynecology clinic om May 1, 2024 o June 31, 2024. Pa ien s wi h di icul y pe o ming 119
he Valsal a maneu e o a his o y o pel ic loo dys unc ion we e excluded. The 120
ollowing clinical pa ame e s we e collec ed o each pa ien : age, weigh , heigh , body 121
mass index (BMI), pa i y, menopausal s a us, age a menopause. 122
123
The s udy was app o ed by Andalucia’s Boa d o Biomedicine E hics Commi ee, wi h 124
code SICEIA-2024-001928. The s udy was conduc ed in acco dance wi h he Decla a ion 125
o Helsinki. All pa ien s ga e hei w i en in o med consen be o e s a ing he s udy. 126
127
Ul asound examina ion 128
All anspe ineal ul asounds we e pe o med by he same expe pel ic loo ul asound 129
examine using a Canon i700 Aplio® (Canon Medical Sys ems Co p., Tokyo, Japan) 130
ul asound wi h a PVT-675MV 3D abdominal p obe. Images we e acqui ed ollowing 131
guidelines p e iously es ablished in he li e a u e, wi h pa ien s in do sal li ho omy 132
posi ion wi h hips lexed [1]. P io o cap u ing and s o ing he ideo, he pa ien was 133
ained o co ec ly pe o m he Valsal a maneu e . Each pa ien was gi en an ul asound 134
ideo cap u e showing he midsagi al plane o he pel ic loo a es and he Valsal a 135
maneu e . The ul asound ideos we e o ien ed by placing he c anio en al egion on 136
he le and he do socaudal egion on he igh . The expe examine cap u ed 45 ideos 137
7
o he co ec midsagi al plane and 45 ideos o he inco ec midsagi al plane. A co ec 138
midsagi al plane was de ined as one ha included he iew o he pubic symphysis, 139
u e h a, bladde , agina, u e us, anus, ec um and le a o ani muscle ( igu e 1). An 140
inco ec midsagi al plane was de ined when any o he p e iously desc ibed ana omical 141
s uc u es was missing, ei he due o a displacemen o he p obe in he an e opos e io 142
axis ( igu e 2) o a o a ion o he image ( igu e 3). 143
144
Algo i hm 145
146
The aim o his wo k is o build a model a model ha de e mines whe he he ul asound 147
was pe o med co ec ly o each speci ic o gan. In o de o c ea e such model, he i s 148
60 ideos we e used as aining se , whe eas he las 30 cases we e ese ed as es se . 149
150
Fo his p ojec , a segmen a ion model ained on a p e iously published pape [15], ha 151
aims o iden i y o gan posi ions on images ( ames) ex ac ed om ul asound ideos, 152
was used o gene a e segmen a ions o he 90 new ideos. In his s age, he ideo ames 153
a e ex ac ed and sen one by one o he men ioned segmen a ion model which will 154
gene a e a segmen a ion pe image ( igu e 4). 155
156
F om he segmen a ions, a se o ea u es o s a is ics was ex ac ed o each ame and 157
each segmen ed o gan o cap u e he ela i e con idence in he iden i ica ion o each 158
o gan. The ex ac ed ea u es include he a e age con idence le el o he ull 159
segmen a ion (p edic ion mean), con idence a ia ion o he ull segmen a ion (p edic ion 160
de ia ion), maximum con idence le el o he ull segmen a ion (maximum p edic ion 161
alue), and minimum con idence le el o he ull segmen a ion (minimum p edic ion 162
alue). 163
164
Addi ionally, o each o gan in each ame, hese same ea u es we e calcula ed, bu only 165
conside ing p edic ion alues g ea e han 0.5. This app oach allows cap u ing he 166
absolu e con idence o he model in he iden i ica ion o he o gans, compa ed o he 167
gene al con idence in each ame. 168
169
8
I he alues ob ained o he o al p edic ions a e signi ican ly lowe han hose calcula ed 170
o p edic ions g ea e han 0.5, i indica es ha he con idence assigned o he pixels ha 171
do no belong o he o gan is e y low, e lec ing a high con idence o he model in he 172
iden i ica ion o he o gan. Con e sely, i he alues be ween bo h g oups a e simila , i 173
means ha high le els o con idence ha e been assigned o pixels ou side he o gan, 174
indica ing lowe ce ain y o he model in he segmen a ion. 175
176
Fo each segmen a ion o each o gan, i s bounding box was calcula ed, i.e., he box ha 177
con ains he p edic ed segmen a ion. Fo each bounding box, he cen oid, he maximum 178
and minimum coo dina es on each axis, he wid h, heigh , and a ea we e calcula ed. The 179
a ionale o hese ea u es was ha he model could de e mine whe he he loca ion, size, 180
and aspec a io we e consis en o ha o gan. 181
182
To build a abula da ase in which each ow ep esen s a pa ien and an o gan, labeled as 183
"co ec " o "inco ec ," i was necessa y o agg ega e he ea u es calcula ed a he ame 184
le el o he pa ien o ul asound le el, i.e., a he ame sequence le el ( igu e 5). This 185
agg ega ion was pe o med by calcula ing he mean, s anda d de ia ion, maximum alue, 186
and minimum alue o he ea u es ob ained o e a sequence o N ames. 187
188
As a esul , a da ase was gene a ed in which each ow ep esen s an o gan and a sequence 189
o N ames, wi h a o al o 68 ea u es pe ow. This da ase was used o ain a model 190
ha de e mines whe he he ul asound was pe o med co ec ly o each speci ic o gan. 191
N = 60 was se , esul ing in a o al o 1,816 ows o he 90 cases analyzed. A e di iding 192
he se in o aining and es subse s, 1,184 ows we e ob ained o he aining se and 193
632 o he es se . Figu e 6 shows he dis ibu ion o he samples in bo h subse s. 194
195
The alue o N = 60 was chosen ins ead o using he comple e da a sequence o each 196
ul asound as a "da a augmen a ion" echnique, use ul when he e is li le da a a ailable. 197
I only one ow pe ul asound had been gene a ed, he da ase would ha e 720 ows, 198
gi en ha he e a e 90 pa ien s and 8 o gans pe pa ien . Howe e , by using windows o 199
N ames, mul iple ows we e gene a ed pe pa ien and o gan, which inc eased he 200
da ase 's size o 1,816 ows. 201
202
15
pel ic le a o hia us segmen a ion om ul asound images. Eu J Radiol Open. 2022; 399
24(9):100412. doi: 10.1016/j.ej o.2022.100412. 400
12) Li X, Hong Y, Kong D, Zhang X. Au oma ic segmen a ion o le a o hia us om 401
ul asound images using U-ne wi h dense connec ions. Phys. Med. Biol. 2019, 64, 402
075015. 403
13) Noo F D, Manzini C, an de Vaa CH, an Limbeek MAJ, Slump CH, G ob ATM. 404
Au oma ic iden i ica ion and segmen a ion o slice o minimal hia al dimensions in 405
anspe ineal ul asound olumes. Ul asound Obs e . Gynecol. 2022, 60, 570–576. 406
14) Duan L, Wang Y, Li J, Zhou N. Explo ing he clinical diagnos ic alue o pel ic loo 407
ul asound images. The Jou nal o Supe compu ing. 2021, 77: 10699–720.38. 408
15) Ga cía-Mejido JA, Solis-Ma ín D, Ma ín-Mo án M, Fe nández-Conde C, 409
Fe nández-Palacín F, Sainz-Bueno JA. Applicabili y o Deep Lea ning o Dynamically 410
Iden i y he Di e en O gans o he Pel ic Floo in he Midsagi al Plane. In U ogynecol 411
J. 2024; 24. doi: 10.1007/s00192-024-05841-0. 412
16) Chen, T. & Gues in, C., 2016. XGBoos : A Scalable T ee Boos ing Sys em. In 413
P oceedings o he 22nd ACM SIGKDD In e na ional Con e ence on Knowledge 414
Disco e y and Da a Mining. KDD '16. New Yo k, NY, USA: ACM, pp. 785–794. 415
A ailable a : h p://doi.acm.o g/10.1145/2939672.2939785. 416
17) G adien Boos ing: F iedman, J.H., 2001. G eedy unc ion app oxima ion: a g adien 417
boos ing machine. Annals o s a is ics, pp.1189–1232. 418
18) Szen im ey Z, Ame i G, Hong CX, Cheung RYK, Ukwa a E, El ahawi A. Au oma ed 419
segmen a ion and measu emen o he emale pel ic loo om he mid-sagi al plane o 420
3D ul asound olumes. Med Phys. 2023;50(10):6215-6227. 421
19) Ga cía-Mejido JA, Fe nández-Palacín A, Bonomi-Ba by MJ, De la Fuen e Vaque o 422
P, Iglesias E, Sainz JA. Online lea ning o 3D/4D anspe ineal ul asound o he pel ic 423
loo . J Ma e n Fe al Neona al Med. 2020;33(16):2805-2811. 424
20) Mu a M, Takahashi T, Tamai N, Suzuki M, Kawamo o A, Sanada H, Nakagami G. 425
Pel ic loo muscle con ac ion au oma ic e alua ion algo i hm o pel ic loo muscle 426
aining bio eedback using sel -pe o med ul asound. BMC Womens Heal h. 427
2024;4;24(1):219. doi: 10.1186/s12905-024-03041-y.. 428
21) Ni D, Ji X, Gao Y, Cheng JZ, Wang H, Qin J, Lei B, Wang T, Wu G, Shen D. 429
Au oma ic Cys ocele Se e i y G ading in Ul asound by Spa io-Tempo al Reg ession. 430
16
Med Image Compu Compu Assis In e . 2016;9901:247-255. 431
22) Eisenbe g VH, Chan a aso n V, Shek KL, Die z HP. Does le a o ani inju y a ec 432
cys ocele ype? Ul asound Obs e Gynecol 2010;36(5):618-23. 433
23) G een TH J . U ina y s ess incon inence: di e en ial diagnosis, pa hophysiology, 434
and managemen . Am J Obs e Gynecol 1975;122(3):368-400. 435
24) Chan a aso n V, Die z HP. Diagnosis o cys ocele ype by clinical examina ion and 436
pel ic loo ul asound. Ul asound Obs e Gynecol 2012; 39 (6): 710-4. 437
25) Ga cía-Mejido JA, Ramos-Vega Z, A mijo-Sánchez A, Fe nández-Palacín A, Ga cía-438
Jimenez R, Sainz JA. Di e en ial diagnosis o middle compa men pel ic o gan p olapse 439
wi h anspe ineal ul asound. In U ogynecol J 2021;32(8):2219-25. 440
26) Ga cía-Mejido JA, Ramos Vega Z, A mijo Sánchez A, Fe nández-Palacín A, 441
Fe nández CB, Sainz Bueno JA. In e obse e a iabili y o ul asound measu emen s o 442
he di e en ial diagnosis o u e ine p olapse and ce ical elonga ion wi hou u e ine 443
p olapse. In . U ogynecol. J. 2022;33(10):2825-31. 444
27) Die z HP, S eensma AB. Pos e io compa men p olapse on wo-dimensional and 445
h ee- dimensional pel ic loo ul asound: he dis inc ion be ween ue ec ocele, pe ineal 446
hype mo ili y and en e ocele. Ul asound Obs e Gynecol 2005; 26 (1): 73-7. 447
28) Ga cía-Mejido JA, Ramos-Vega Z, Fe nández-Palacín A, Bo e o C, Valdi ia M, 448
Pelayo-Delgado I, Sainz-Bueno JA. P edic i e Model o he Diagnosis o U e ine 449
P olapse Based on T anspe ineal Ul asound . Tomog aphy. 2022;8(4):1716-25. 450
29) Ga cía-Mejido JA, Ma ín-Ma inez A, González-Diaz E, Núñez-Ma as MJ, 451
Fe nández-Palacín A, Ca ballo-Ras illa S, Fe nández-Fe nández C, Ga cía-Jimenez R, 452
Sainz-Bueno JA. Is I Possible o Diagnose Su gical U e ine P olapse Wi h T anspe ineal 453
Ul asound? Mul icen e Valida ion o Diagnos ic So wa e. J Ul asound Med. 2023; 8. 454
doi: 10.1002/jum.16303. 455
30) Ga cía Mejido JA, O ega I, Ga cía-Jiménez R, Sainz-Bueno JA. Di e en ial 456
diagnosis o pos e io compa men p olapse using anspe ineal ul asound Ul asound 457
Obs e Gynecol. 2022;60(1):142-144. 458
459
17
Table 1: Cha ac e is ics o he pa ien s included in he es se . 460
461
!
n: 30!
95% CI
Age!
46.1±13.9!
40.9; 51.3!
Weigh !
68.9±10.1!
65.1; 72.6!
Heigh !
161.5±5.9!
159.3; 163.7!
BMI!
26.3±4.3!
24.6; 27.9!
Pa i y!
1.5±1.1!
1.1; 1.9!
Menopause!
9 (31.0%)!
17.1%; 49.4%!
Menopause age!
53.7±1.5!
52.5: 54.8!
462
18
Table 2: Kappa index assessmen o he p edic i e model o he co ec iden i ica ion 463
o he di e en pel ic loo o gans using he expe examine 's examina ion as he gold 464
s anda d. 465
466
P edic i e model
(n:30)
P alue
(McNema )
Kappa (p)
Co ec midsagi al plane
19 (63.3%)
1.0
0.930 (<0.001)
Co ec isualiza ion o he pubis
27(90.0%)
1.0
1.0 (<0.001)
Co ec isualiza ion o he u e h a
23(76.7%)
1.0
1.0(<0.001)
Co ec isualiza ion o he u ina y bladde
29(96.7%)
1.0
1.0(<0.001)
Co ec isualiza ion o he agina
30(100%)
---
--- (---)
Co ec isualiza ion o he u e us
27(90.0%)
1.0
0.783(<0.001)
Co ec isualiza ion o he anus
25(83.3%)
1.0
0.760(<0.001)
Co ec isualiza ion o he ec um
26(86.7%)
1.0
0.516(<0.001)
Co ec isualiza ion o he le a o ani muscle
26(86.7%)
1.0
0.870(<0.001)
467
19
Table 3: Kappa index assessmen o he junio examine o he co ec iden i ica ion 468
o he di e en pel ic loo o gans using he expe examine 's examina ion as he gold 469
s anda d. 470
471
Junio examine
(n:30)
P alue
(McNema )
Kappa (p)
Co ec midsagi al plane
19(63.3%)
1.0
0.930(<0.001)
Co ec isualiza ion o he pubis
30(100%)
---
--- (---)
Co ec isualiza ion o he u e h a
24(80.0%)
1.0
0.902(<0.001)
Co ec isualiza ion o he u ina y bladde
28(93.3%)
1.0
-0.047(0.786)
Co ec isualiza ion o he agina
30(100%)
---
--- (---)
Co ec isualiza ion o he u e us
29(96.7%)
1.0
0.651 (<0.001)
Co ec isualiza ion o he anus
26(86.7%)
1.0
0.609(<0.001)
Co ec isualiza ion o he ec um
24(80.0%)
0.250
0.615(<0.001)
Co ec isualiza ion o he le a o ani muscle
24(80.0%)
1.0
0.667(<0.001)
472
20
Table 4: Kappa index assessmen o he senio examine o he co ec iden i ica ion 473
o he di e en pel ic loo o gans using he expe examine 's examina ion as he gold 474
s anda d. 475
476
senio examine
(n:30)
P alue
(McNema )
Kappa (p)
Co ec midsagi al plane
19(63.3%)
1.0
0.789(<0.001)
Co ec isualiza ion o he pubis
28(93.3%)
1.0
0.348(0.051 )
Co ec isualiza ion o he u e h a
24(80.0%)
1.0
0.902(<0.001)
Co ec isualiza ion o he u ina y bladde
30(100%)
---
--- (---)
Co ec isualiza ion o he agina
30(100%)
---
--- (---)
Co ec isualiza ion o he u e us
24(80.0%)
0.125
0.444(0.003)
Co ec isualiza ion o he anus
26(86.7%)
1.0
0.609(<0.001)
Co ec isualiza ion o he ec um
25(83.3%)
0.625
0.429(0.014)
Co ec isualiza ion o he le a o ani muscle
24(80.0%)
1.0
0.667(<0.001)
477
478
21
Figu e 1: Shows he co ec midsagi al plane. Pubis (P), u e h a (U), u ina y bladde 479
(UB), agina (V), u e us (U), anus (AN), ec um (R), le a o ani muscle (L). 480
481
Figu e 2: Inco ec midsagi al plane whe e he anus, ec um and le a o ani muscle 482
a e no isualized due o displacemen in he sagi al axis. Pubis (P), u e h a (U), 483
u ina y bladde (UB), agina (V), u e us (U). 484
485
22
Figu e 3: Inco ec midsagi al plane whe e he u e us, anus, ec um and le a o ani 486
muscle a e no isualized due o a o a ion o he image (C). Pubis (P), u e h a (U), 487
u ina y bladde (UB), agina (V). 488
489
Figu e 4: Ex ac ion o ames om he ideo and sending o he segmen a ion model 490
ha will gene a e segmen a ion by image. 491
492
23
Figu e 5: Agg ega ion o compu ed ea u es a he ame sequence le el. 493
494
Figu e 6: T ain se dis ibu ion (A) Tes se dis iubu ion (B). 495
496
Figu e 7: Hype pa ame e op imiza ion h ough g ouped 5- old c oss- alida ion 497
498
