-
Notifications
You must be signed in to change notification settings - Fork 0
/
appendix.tex
467 lines (435 loc) · 30 KB
/
appendix.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
\chapter{Additional Results} \label{app:results}
\section{Experiment 1: Feasibility} %
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_T2_IR_VS}
\caption[Boxplot for the \glsdesc{vs} for Feasibility]{Boxplot for the \acrlong{vs}.}
\label{fig:results_boxplot_T2_IR_vs}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_T2_IR_AVD}
\caption[Boxplot for the \glsdesc{avd} for Feasibility]{Boxplot for the \acrlong{avd}.}
\label{fig:results_boxplot_T2_IR_avd}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_T2_IR_HD95}
\caption[Boxplot for the \glsdesc{hd95} for Feasibility]{Boxplot for the \acrlong{hd95}.}
\label{fig:results_boxplot_T2_IR_hd95}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_T2_IR_HD}
\caption[Boxplot for the \glsdesc{hd} for Feasibility]{Boxplot for the \acrlong{hd}.}
\label{fig:results_boxplot_T2_IR_hd}
\end{figure}
\section{Experiment 2: 3-D Context} % ==============================================================================
\begin{sidewaystable}[htbp]
\centering
\caption[Detailed Results for 3D Context]{}
\begin{tabular}{l*{6}{l}}
\toprule
Cohort & Neural Network & DICE & VS & AVD & HD95 & HD \\
& & & & (mm) & (mm) & (mm) \\
\midrule
Patient & Base & $0.704 \pm 0.139$ & $0.882 \pm 0.098$ & $2.880 \pm 2.671$ & $16.526 \pm 17.025$ & $63.696 \pm 20.903$ \\
& Stack\_3to1 & $0.749 \pm 0.139$ & $0.896 \pm 0.123$ & $\mathbf{1.979 \pm 2.190}$ & $\mathbf{10.807 \pm 13.393}$ & $\mathbf{56.262 \pm 23.958}$ \\
& Stack\_5to1 & $\mathbf{0.765 \pm 0.123}$ & $\mathbf{0.898 \pm 0.110}$ & $2.001 \pm 2.401$ & $12.418 \pm 19.104$ & $56.304 \pm 28.746$ \\
& Stack\_5to3 & $0.717 \pm 0.135$ & $0.883 \pm 0.105$ & $2.995 \pm 3.148$ & $19.312 \pm 22.545$ & $65.740 \pm 22.811$ \\
& Stack\_Proj & $0.712 \pm 0.136$ & $0.889 \pm 0.094$ & $3.119 \pm 3.038$ & $19.878 \pm 21.613$ & $60.762 \pm 22.985$ \\
& Patch & $0.682 \pm 0.128$ & $0.876 \pm 0.101$ & $3.581 \pm 2.724$ & $24.241 \pm 20.896$ & $70.737 \pm 26.853$ \\
\midrule
Volunteer & Base & $0.861 \pm 0.057$ & $0.921 \pm 0.056$ & $0.643 \pm 0.866$ & $1.644 \pm 2.321 $ & $\mathbf{35.380 \pm 32.720}$ \\
& Stack\_3to1 & $0.860 \pm 0.073$ & $0.925 \pm 0.043$ & $0.734 \pm 0.859$ & $2.260 \pm 2.336 $ & $39.327 \pm 29.429$ \\
& Stack\_5to1 & $\mathbf{0.878 \pm 0.048}$ & $\mathbf{0.928 \pm 0.048}$ & $ 0.509 \pm 0.393 $ & $1.537 \pm 1.784 $ & $46.515 \pm 30.853$ \\
& Stack\_5to3 & $0.855 \pm 0.052$ & $0.928 \pm 0.062$ & $\mathbf{0.473 \pm 0.435}$ & $\mathbf{1.350 \pm 1.365} $ & $39.033 \pm 30.589$ \\
& Stack\_Proj & $0.842 \pm 0.051$ & $0.917 \pm 0.067$ & $0.861 \pm 0.813$ & $1.642 \pm 1.549 $ & $40.584 \pm 30.139$ \\
& Patch & $0.806 \pm 0.068$ & $0.888 \pm 0.085$ & $1.166 \pm 1.164$ & $7.992 \pm 13.474$ & $39.796 \pm 27.201$ \\
\bottomrule
\end{tabular}
\label{tab:results_3d_context}
\end{sidewaystable}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_VS}
\caption[Boxplot for the \glsdesc{vs} for 3-D Context]{Boxplot for the \acrlong{vs}.}
\label{fig:results_boxplot_vs}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_AVD}
\caption[Boxplot for the \glsdesc{avd} for 3-D Context]{Boxplot for the \acrlong{avd}.}
\label{fig:results_boxplot_avd}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_HD95}
\caption[Boxplot for the \glsdesc{hd95} for 3-D Context]{Boxplot for the 95\textsuperscript{th} percentile \acrlong{hd}.}
\label{fig:results_boxplot_hd95}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_HD}
\caption[Boxplot for the \glsdesc{hd} for 3-D Context]{Boxplot for the \acrlong{hd}.}
\label{fig:results_boxplot_hd}
\end{figure}
\section{Experiment 3: Post-processing} % ===========================================================================
\begin{sidewaystable}[htbp]
\centering
\caption[Detailed Results for Post-processing]{}
\begin{tabular}{l*{7}{l}}
\toprule
Cohort & Network & Post-processing & DICE & VS & AVD & HD95 & HD \\
& & & & & (mm) & (mm) & (mm) \\
\midrule
Patient & Base & None & $0.705 \pm 0.137$ & $\mathbf{0.883 \pm 0.097}$ & $2.835 \pm 2.655$ & $16.285 \pm 16.896$ & $62.630 \pm 21.803$ \\
& & Volumes only & $0.711 \pm 0.145$ & $0.867 \pm 0.125$ & $3.431 \pm 4.236$ & $20.364 \pm 20.125$ & $50.726 \pm 21.318$ \\
& & Joint volumes & $\mathbf{0.722 \pm 0.136}$ & $0.873 \pm 0.125$ & $\mathbf{1.705 \pm 1.768}$ & $\mathbf{11.812 \pm 12.785}$& $\mathbf{32.159 \pm 20.178}$ \\
\cmidrule{2-8}
& Stack\_5to1 & None & $0.765 \pm 0.123$ & $0.898 \pm 0.110$ & $2.001 \pm 2.401$ & $12.418 \pm 19.104$ & $56.304 \pm 28.746$ \\
& & Volumes only & $0.772 \pm 0.120$ & $0.899 \pm 0.119$ & $1.871 \pm 2.534$ & $11.481 \pm 16.706$ & $40.531 \pm 23.941$ \\
& & Joint volumes & $\mathbf{0.779 \pm 0.123}$ & $\mathbf{0.905 \pm 0.117}$ & $\mathbf{1.106 \pm 1.670}$ & $\mathbf{6.688 \pm 10.332}$ & $\mathbf{28.981 \pm 19.820}$ \\
\midrule
Volunteer & Base & None & $0.861 \pm 0.057$ & $0.921 \pm 0.056$ & $0.643 \pm 0.866$ & $1.644 \pm 2.321 $ & $35.380 \pm 32.720$ \\
& & Volumes only & $0.862 \pm 0.057$ & $0.924 \pm 0.056$ & $0.608 \pm 0.833$ & $2.311 \pm 4.508 $ & $32.943 \pm 30.360$ \\
& & Joint volumes & $\mathbf{0.868 \pm 0.050}$ & $\mathbf{0.929 \pm 0.056}$ & $\mathbf{0.197 \pm 0.173}$ & $\mathbf{1.230 \pm 1.255}$ & $\mathbf{7.894 \pm 5.844}$ \\
\cmidrule{2-8}
& Stack\_5to1 & None & $0.884 \pm 0.046$ & $0.927 \pm 0.051$ & $0.473 \pm 0.399$ & $1.140 \pm 1.344 $ & $46.547 \pm 32.724$ \\
& & Volumes only & $0.883 \pm 0.046$ & $0.933 \pm 0.049$ & $0.349 \pm 0.224$ & $1.357 \pm 1.454 $ & $32.552 \pm 28.627$ \\
& & Joint volumes & $\mathbf{0.894 \pm 0.042}$ & $\mathbf{0.942 \pm 0.050}$ & $\mathbf{0.102 \pm 0.060}$ & $\mathbf{0.655 \pm 0.355}$ & $\mathbf{5.177 \pm 2.088}$ \\
\bottomrule
\end{tabular}
\label{tab:results_pp}
\end{sidewaystable}
\begin{figure}[htbp]
\centering
\subfloat[]
{
\label{fig:subfig:pp_boxplot_base_dice}
\includegraphics[width=0.7\textwidth]{pp_boxplot_base_DICE}
}
\hfill
\subfloat[]
{
\label{fig:subfig:pp_boxplot_5to1_dice}
\includegraphics[width=0.7\textwidth]{pp_boxplot_5to1_DICE}
}
\caption[Boxplots for the \glsdesc{dice} for Post-processing]{Boxplots for the \acrlong{dice} for the \textbf{a)} baseline and best performing \textbf{b)} stack-wise architecture. \textit{n largest volumes} means that only the $n$ largest volumes were kept ($n = 3$ and $n = 2$ for the base and Stack\_5to1 architecture, respectively). \textit{Joint volumes} means that we tried to connect the correctly segmented volumes first, and then only kept the largest one.}
\label{fig:pp_boxplots_dice}
\end{figure}
\begin{figure}[htbp]
\centering
\subfloat[]
{
\label{fig:subfig:pp_boxplot_base_vs}
\includegraphics[width=0.7\textwidth]{pp_boxplot_base_VS}
}
\hfill
\subfloat[]
{
\label{fig:subfig:pp_boxplot_5to1_vs}
\includegraphics[width=0.7\textwidth]{pp_boxplot_5to1_VS}
}
\caption[Boxplots for the \glsdesc{vs} for Post-processing]{Boxplots for the \acrlong{vs} for the \textbf{a)} baseline and best performing \textbf{b)} stack-wise architecture. \textit{n largest volumes} means that only the $n$ largest volumes were kept ($n = 3$ and $n = 2$ for the base and Stack\_5to1 architecture, respectively). \textit{Joint volumes} means that we tried to connect the correctly segmented volumes first, and then only kept the largest one.}
\label{fig:pp_boxplots_vs}
\end{figure}
\begin{figure}[htbp]
\centering
\subfloat[]
{
\label{fig:subfig:pp_boxplot_base_avd}
\includegraphics[width=0.7\textwidth]{pp_boxplot_base_AVD}
}
\hfill
\subfloat[]
{
\label{fig:subfig:pp_boxplot_5to1_avd}
\includegraphics[width=0.7\textwidth]{pp_boxplot_5to1_AVD}
}
\caption[Boxplots for the \glsdesc{avd} for Post-processing]{Boxplots for the \acrlong{avd} for the \textbf{a)} baseline and best performing \textbf{b)} stack-wise architecture. \textit{n largest volumes} means that only the $n$ largest volumes were kept ($n = 3$ and $n = 2$ for the base and Stack\_5to1 architecture, respectively). \textit{Joint volumes} means that we tried to connect the correctly segmented volumes first, and then only kept the largest one.}
\label{fig:pp_boxplots_avd}
\end{figure}
\begin{figure}[htbp]
\centering
\subfloat[]
{
\label{fig:subfig:pp_boxplot_base_hd}
\includegraphics[width=0.7\textwidth]{pp_boxplot_base_HD}
}
\hfill
\subfloat[]
{
\label{fig:subfig:pp_boxplot_5to1_hd}
\includegraphics[width=0.7\textwidth]{pp_boxplot_5to1_HD}
}
\caption[Boxplots for the \glsdesc{hd} for Post-processing]{Boxplots for the \acrlong{hd} for the \textbf{a)} baseline and best performing \textbf{b)} stack-wise architecture. \textit{n largest volumes} means that only the $n$ largest volumes were kept ($n = 3$ and $n = 2$ for the base and Stack\_5to1 architecture, respectively). \textit{Joint volumes} means that we tried to connect the correctly segmented volumes first, and then only kept the largest one.}
\label{fig:pp_boxplots_hd}
\end{figure}
\section{Comparison to Human Inter-Rater Performance} % ===========================================================================
\begin{sidewaystable}[htbp]
\centering
\caption[Detailed Results for Comparison to Inter-Rater Performance]{Values for \acrlong{dice}, \acrlong{vs}, \acrlong{avd}, \acrlong{hd95} and \acrlong{hd} achieved when comparing the best performing FCNN (Stack\_5to1 with full post-processing) to the consesus ground truth (FCNN-GT) and rater to rater (R-R).}
\begin{tabular}{l*{7}{l}}
\toprule
Cohort & Comparison & DICE & VS & AVD & HD95 & HD \\
& & & & (mm) & (mm) & (mm) \\
\midrule
Patient & FCNN-GT & $0.779 \pm 0.123$ & $\mathbf{0.905 \pm 0.117}$ & $\mathbf{1.106 \pm 1.670}$ & $\mathbf{6.688 \pm 10.332}$ & $28.981 \pm 19.820$ \\
& R-R & $\mathbf{0.786 \pm 0.093}$ & $0.897 \pm 0.087$ & $1.410 \pm 2.303$ & $11.245 \pm 19.008$ & $\mathbf{28.500 \pm 26.472}$ \\
\midrule
Volunteer & FCNN-GT & $\mathbf{0.894 \pm 0.042}$ & $\mathbf{0.942 \pm 0.050}$ & $\mathbf{0.102 \pm 0.060}$ & $\mathbf{0.655 \pm 0.355} $ & $\mathbf{5.177 \pm 2.088} $ \\
& R-R & $0.869 \pm 0.031$ & $0.937 \pm 0.043$ & $0.110 \pm 0.075$ & $0.703 \pm 0.672 $ & $5.304 \pm 5.902 $ \\
\bottomrule
\end{tabular}
\label{tab:res_fcnn_rater}
\end{sidewaystable}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_eval_VS}
\caption[Boxplot for the \glsdesc{vs} for Comparison to Inter-Rater Performance]{Boxplot for \acrlong{vs} by comparing the best performing FCNN (Stack\_5to1 with full post-processing) to the consesus ground truth (FCNN-GT) and rater to rater (R-R).}
\label{fig:results_eval_boxplot_vs}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_eval_AVD}
\caption[Boxplot for the \glsdesc{avd} for Comparison to Inter-Rater Performance]{Boxplot for \acrlong{avd} by comparing the best performing FCNN (Stack\_5to1 with full post-processing) to the consesus ground truth (FCNN-GT) and rater to rater (R-R).}
\label{fig:results_eval_boxplot_avd}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_eval_HD95}
\caption[Boxplot for the \glsdesc{hd95} for Comparison to Inter-Rater Performance]{Boxplot for \acrlong{hd95} by comparing the best performing FCNN (Stack\_5to1 with full post-processing) to the consesus ground truth (FCNN-GT) and rater to rater (R-R).}
\label{fig:results_eval_boxplot_hd95}
\end{figure}
\begin{figure}[htbp]
\centering
\includegraphics[width=0.8\textwidth]{boxplot_eval_HD}
\caption[Boxplot for the \glsdesc{hd} for Comparison to Inter-Rater Performance]{Boxplot for \acrlong{hd} by comparing the best performing FCNN (Stack\_5to1 with full post-processing) to the consesus ground truth (FCNN-GT) and rater to rater (R-R).}
\label{fig:results_eval_boxplot_hd}
\end{figure}
\chapter{Network Architectures} % =================================================================================
\begin{sidewaystable}[htbp]
\centering
\caption[Architecture of Base]{Detailed architecture of the baseline neural network.}
\begin{tabular}{l*{4}{l}}
\toprule
Level & Layer & Properties & In & Out \\
& & & $C \times D \times H \times W$ & $C \times D \times H \times W$ \\
\midrule
1D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $2 \times 1 \times 300 \times 300$ & $32 \times 1 \times 300 \times 300$ \\
1D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $32 \times 1 \times 300 \times 300$ & $32 \times 1 \times 300 \times 300$ \\
2D & Max Pooling & K: $2 \times 2$, P0, S2 & $32 \times 1 \times 300 \times 300$ & $32 \times 1 \times 150 \times 150$ \\
2D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $32 \times 1 \times 150 \times 150$ & $64 \times 1 \times 150 \times 150$ \\
2D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $64 \times 1 \times 150 \times 150$ & $64 \times 1 \times 150 \times 150$ \\
3D & Max Pooling & K: $2 \times 2$, P0, S2 & $64 \times 1 \times 150 \times 150$ & $64 \times 1 \times 75 \times 75$ \\
3D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $64 \times 1 \times 75 \times 75$ & $128 \times 1 \times 75 \times 75$ \\
3D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $128 \times 1 \times 75 \times 75$ & $128 \times 1 \times 75 \times 75$ \\
4D & Max Pooling & K: $2 \times 2$, P0, S2 & $128 \times 1 \times 75 \times 75$ & $128 \times 1 \times 37 \times 37$ \\
4D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $128 \times 1 \times 37 \times 37$ & $256 \times 1 \times 37 \times 37$ \\
4D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $256 \times 1 \times 37 \times 37$ & $256 \times 1 \times 37 \times 37$ \\
5D & Max Pooling & K: $2 \times 2$, P0, S2 & $256 \times 1 \times 37 \times 37$ & $256 \times 1 \times 18 \times 18$ \\
5D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $256 \times 1 \times 18 \times 18$ & $512 \times 1 \times 18 \times 18$ \\
5D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $512 \times 1 \times 18 \times 18$ & $512 \times 1 \times 18 \times 18$ \\
4U & ConvTranspose2D & K: $2 \times 2$, P0, S2 & $512 \times 1 \times 18 \times 18$ & $256 \times 1 \times 37 \times 37$ \\
4U & Concat Skip Feature Map & & & $512 \times 1 \times 37 \times 37$ \\
4U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $512 \times 1 \times 37 \times 37$ & $256 \times 1 \times 37 \times 37$ \\
4U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $256 \times 1 \times 37 \times 37$ & $256 \times 1 \times 37 \times 37$ \\
3U & ConvTranspose2D & K: $2 \times 2$, P0, S2 & $256 \times 1 \times 37 \times 37$ & $128 \times 1 \times 75 \times 75$ \\
3U & Concat Skip Feature Map & & & $256 \times 1 \times 75 \times 75$ \\
3U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $256 \times 1 \times 75 \times 75$ & $128 \times 1 \times 75 \times 75$ \\
3U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $128 \times 1 \times 75 \times 75$ & $128 \times 1 \times 75 \times 75$ \\
2U & ConvTranspose2D & K: $2 \times 2$, P0, S2 & $128 \times 1 \times 75 \times 75$ & $64 \times 1 \times 150 \times 150$ \\
2U & Concat Skip Feature Map & & & $128 \times 1 \times 150 \times 150$ \\
2U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $128 \times 1 \times 150 \times 150$ & $64 \times 1 \times 150 \times 150$ \\
2U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $64 \times 1 \times 150 \times 150$ & $64 \times 1 \times 150 \times 150$ \\
1U & ConvTranspose2D & K: $2 \times 2$, P0, S2 & $128 \times 1 \times 150 \times 150$ & $32 \times 1 \times 300 \times 300$ \\
1U & Concat Skip Feature Map & & & $64 \times 1 \times 300 \times 300$ \\
1U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $64 \times 1 \times 300 \times 300$ & $32 \times 1 \times 300 \times 300$ \\
1U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3$, P1, S1 & $32 \times 1 \times 300 \times 300$ & $32 \times 1 \times 300 \times 300$ \\
Out & Conv2D & K: $1 \times 1$, P0, S1 & $32 \times 1 \times 300 \times 300$ & $1 \times 1 \times 300 \times 300$ \\
\bottomrule
\end{tabular}
\label{tab:architecture_fcnn_base}
\end{sidewaystable}
\begin{sidewaystable}[htbp]
\centering
\caption[Architecture of Stack]{Detailed architecture of the stack-wise neural network. $I$, $O$ correspond to the chosen number of input and output slices, respectively. The following combinations have been trained: $I = 3$, $O = 1$ (3-to-1), $I = 5$, $O = 1$ (5-to-1), $I = 5$, $O = 3$ (5-to-3).}
\begin{tabular}{l*{4}{l}}
\toprule
Level & Layer & Properties & In & Out \\
& & & $C \times D \times H \times W$& $C \times D \times H \times W$ \\
\midrule
1D & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $2 \times I \times 300 \times 300$ & $32 \times I \times 300 \times 300$ \\
1D & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $32 \times I \times 300 \times 300$ & $32 \times I \times 300 \times 300$ \\
2D & Max Pooling & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $32 \times I \times 300 \times 300$ & $32 \times I \times 150 \times 150$ \\
2D & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $32 \times I \times 150 \times 150$ & $64 \times I \times 150 \times 150$ \\
2D & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $64 \times I \times 150 \times 150$ & $64 \times I \times 150 \times 150$ \\
3D & Max Pooling & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $64 \times I \times 150 \times 150$ & $64 \times I \times 75 \times 75$ \\
3D & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $64 \times I \times 75 \times 75$ & $128 \times I \times 75 \times 75$ \\
3D & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $128 \times I \times 75 \times 75$ & $128 \times I \times 75 \times 75$ \\
4D & Max Pooling & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $128 \times I \times 75 \times 75$ & $128 \times I \times 37 \times 37$ \\
4D & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $128 \times I \times 37 \times 37$ & $256 \times I \times 37 \times 37$ \\
4D & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $256 \times I \times 37 \times 37$ & $256 \times I \times 37 \times 37$ \\
5D & Max Pooling & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $256 \times I \times 37 \times 37$ & $256 \times I \times 18 \times 18$ \\
5D & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $256 \times I \times 18 \times 18$ & $512 \times I \times 18 \times 18$ \\
5D & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $512 \times I \times 18 \times 18$ & $512 \times I \times 18 \times 18$ \\
4U & ConvTranspose3D & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $512 \times I \times 18 \times 18$ & $256 \times I \times 37 \times 37$ \\
4U & Concat Skip Feature Map & & & $512 \times I \times 37 \times 37$ \\
4U & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $512 \times I \times 37 \times 37$ & $256 \times I \times 37 \times 37$ \\
4U & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $256 \times I \times 37 \times 37$ & $256 \times I \times 37 \times 37$ \\
3U & ConvTranspose3D & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $256 \times I \times 37 \times 37$ & $128 \times I \times 75 \times 75$ \\
3U & Concat Skip Feature Map & & & $256 \times I \times 75 \times 75$ \\
3U & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $256 \times I \times 75 \times 75$ & $128 \times I \times 75 \times 75$ \\
3U & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $128 \times I \times 75 \times 75$ & $128 \times I \times 75 \times 75$ \\
2U & ConvTranspose3D & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $128 \times I \times 75 \times 75$ & $64 \times I \times 150 \times 150$ \\
2U & Concat Skip Feature Map & & & $128 \times I \times 150 \times 150$ \\
2U & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $128 \times I \times 150 \times 150$ & $64 \times I \times 150 \times 150$ \\
2U & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $64 \times I \times 150 \times 150$ & $64 \times I \times 150 \times 150$ \\
1U & ConvTranspose2D & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $128 \times I \times 150 \times 150$ & $32 \times I \times 300 \times 300$ \\
1U & Concat Skip Feature Map & & & $64 \times I \times 300 \times 300$ \\
1U & Conv3D, Dropout, BN, ReLU & K: $1 \times 3 \times 3$, P(0, 1, 1), S1 & $64 \times I \times 300 \times 300$ & $32 \times I \times 300 \times 300$ \\
1U & Conv3D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $32 \times I \times 300 \times 300$ & $32 \times I \times 300 \times 300$ \\
Out & Conv3D & K: $I \times 1 \times 1$, P0, S1 & $32 \times I \times 300 \times 300$ & $1 \times O \times 300 \times 300$ \\
\bottomrule
\end{tabular}
\label{tab:architecture_fcnn_volumetric}
\end{sidewaystable}
\begin{sidewaystable}[htbp]
\centering
\caption[Architecture of Patch]{Detailed architecture of the patch-wise neural network.}
\begin{tabular}{l*{4}{l}}
\toprule
Level & Layer & Properties & In & Out \\
& & & $C \times D \times H \times W$& $C \times D \times H \times W$ \\
\midrule
1D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $2 \times 12 \times 128 \times 128$ & $32 \times 12 \times 128 \times 128$ \\
1D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $32 \times 12 \times 128 \times 128$ & $32 \times 12 \times 128 \times 128$ \\
2D & Max Pooling & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $32 \times 12 \times 128 \times 128$ & $32 \times 12 \times 64 \times 64$ \\
2D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $32 \times 12 \times 64 \times 64$ & $64 \times 12 \times 64 \times 64$ \\
2D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $64 \times 12 \times 64 \times 64$ & $64 \times 12 \times 64 \times 64$ \\
3D & Max Pooling & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $64 \times 12 \times 64 \times 64$ & $64 \times 12 \times 32 \times 32$ \\
3D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $64 \times 12 \times 32 \times 32$ & $128 \times 12 \times 32 \times 32$ \\
3D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $128 \times 12 \times 32 \times 32$ & $128 \times 12 \times 32 \times 32$ \\
4D & Max Pooling & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $128 \times 12 \times 32 \times 32$ & $128 \times 12 \times 16 \times 16$ \\
4D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $128 \times 12 \times 16 \times 16$ & $256 \times 12 \times 16 \times 16$ \\
4D & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3$, P1, S1 & $256 \times 12 \times 16 \times 16$ & $256 \times 12 \times 16 \times 16$ \\
3U & ConvTranspose2D & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $256 \times 12 \times 16 \times 16$ & $128 \times 12 \times 32 \times 32$ \\
3U & Concat Skip Feature Map & & & $256 \times 12 \times 32 \times 32$ \\
3U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3 \times 3$, P1, S1 & $256 \times 12 \times 32 \times 32$ & $128 \times 12 \times 32 \times 32$ \\
3U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3 \times 3$, P1, S1 & $128 \times 12 \times 32 \times 32$ & $128 \times 12 \times 32 \times 32$ \\
2U & ConvTranspose2D & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $128 \times 12 \times 32 \times 32$ & $64 \times 12 \times 64 \times 64$ \\
2U & Concat Skip Feature Map & & & $128 \times 12 \times 64 \times 64$ \\
2U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3 \times 3$, P1, S1 & $128 \times 12 \times 64 \times 64$ & $64 \times 12 \times 64 \times 64$ \\
2U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3 \times 3$, P1, S1 & $64 \times 12 \times 64 \times 64$ & $64 \times 12 \times 64 \times 64$ \\
1U & ConvTranspose2D & K: $1 \times 2 \times 2$, P0, S(1, 2, 2) & $128 \times 12 \times 64 \times 64$ & $32 \times 12 \times 128 \times 128$ \\
1U & Concat Skip Feature Map & & & $64 \times 12 \times 128 \times 128$ \\
1U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3 \times 3$, P1, S1 & $64 \times 12 \times 128 \times 128$ & $32 \times 12 \times 128 \times 128$ \\
1U & Conv2D, Dropout, BN, ReLU & K: $3 \times 3 \times 3 \times 3$, P1, S1 & $32 \times 12 \times 128 \times 128$ & $32 \times 12 \times 128 \times 128$ \\
Out & Conv2D & K: $1 \times 1 \times 1$, P0, S1 & $32 \times 12 \times 128 \times 128$ & $1 \times 12 \times 128 \times 128$ \\
\bottomrule
\end{tabular}
\label{tab:architecture_fcnn_patches}
\end{sidewaystable}
\chapter{Additional Tables} % =====================================================================================
\begin{table}[htbp]
\centering
\caption[Subject Assignment]{Assignment of subjects to the different folds for the four-fold cross-validation.}
\begin{tabular}{p{3cm}l}
\toprule
\textbf{Fold-0} & $n = 13$ \\
Subject & Cohort \\
\midrule
05B & Volunteer \\
25A & Volunteer \\
012\_V0\_R1 & Patient \\
037\_V0\_R2 & Patient \\
044\_V0\_R4 & Patient \\
084\_V0\_R3 & Patient \\
159\_V0\_R3 & Patient \\
173\_V0\_R4 & Patient \\
174\_V0\_R4 & Patient \\
199\_V0\_R2 & Patient \\
221\_V0\_R3 & Patient \\
229\_V0\_R3 & Patient \\
231\_V0\_R4 & Patient \\
\bottomrule
\end{tabular}
\begin{tabular}{p{3cm}l}
\toprule
\textbf{Fold-1} & $n = 13$ \\
Subject & Cohort \\
\midrule
05A & Volunteer \\
12A & Volunteer \\
013\_V0\_R3 & Patient \\
022\_V0\_R3 & Patient \\
034\_V1\_R3 & Patient \\
161\_V0\_R3 & Patient \\
166\_V0\_R4 & Patient \\
167\_V0\_R3 & Patient \\
192\_V0\_R2 & Patient \\
213\_V0\_R4 & Patient \\
218\_V0\_R3 & Patient \\
220\_V0\_R1 & Patient \\
222\_V0\_R1 & Patient \\
\bottomrule
\end{tabular}
\begin{tabular}{p{3cm}l}
\textbf{Fold-2} & $n = 13$ \\
Subject & Cohort \\
\midrule
12B & Volunteer \\
13A & Volunteer \\
24A & Volunteer \\
049\_V0\_R3 & Patient \\
062\_V0\_R3 & Patient \\
153\_V0\_R3 & Patient \\
158\_V0\_R3 & Patient \\
160\_V0\_R1 & Patient \\
166\_V0\_R5 & Patient \\
171\_V0\_R1 & Patient \\
201\_V0\_R1 & Patient \\
219\_V0\_R4 & Patient \\
222\_V1\_R2 & Patient \\
\bottomrule
\end{tabular}
\begin{tabular}{p{3cm}l}
\textbf{Fold-3} & $n = 13$ \\
Subject & Cohort \\
\midrule
13B & Volunteer \\
24B & Volunteer \\
25B & Volunteer \\
013\_V0\_R2 & Patient \\
019\_V0\_R3 & Patient \\
036\_V1\_R1 & Patient \\
062\_V1\_R3 & Patient \\
172\_V0\_R4 & Patient \\
180\_V0\_R3 & Patient \\
195\_V0\_R1 & Patient \\
215\_V0\_R1 & Patient \\
219\_V0\_R3 & Patient \\
230\_V0\_R3 & Patient \\
\bottomrule
\end{tabular}
\label{tab:fold_assignment}
\end{table}
\begin{sidewaystable}[htbp]
\centering
\caption[Hyperparameters]{The used hyperparemeters for the different network architectures we trained.}
\begin{tabular}{l*{7}{l}}
\toprule
Neural Network & Epochs & Batchsize & Learning Rate & Steps & Momentum & In & Out \\
& & & & & & $D \times H \times W$ & $D \times H \times W$ \\
\midrule
Base & 150 & 32 & 0.001 & 100, 125 & 0.99 & $1 \times 300 \times 300$ & $1 \times 300 \times 300$ \\
Stack\_3to1 & 200 & 8 & 0.0001 & 180, 190 & 0.99 & $3 \times 300 \times 300$ & $1 \times 300 \times 300$ \\
Stack\_5to1 & 200 & 8 & 0.0001 & 180, 190 & 0.99 & $5 \times 300 \times 300$ & $1 \times 300 \times 300$ \\
Stack\_5to3 & 200 & 6 & 0.0001 & 180, 190 & 0.99 & $5 \times 300 \times 300$ & $3 \times 300 \times 300$ \\
Stack\_Proj & 200 & 6 & 0.0001 & 180, 190 & 0.99 & $5 \times 300 \times 300$ & $3 \times 300 \times 300$ \\
Patch & 500 & 16 & 0.0001 & 400, 450 & 0.99 & $12 \times 128 \times 128$ & $12 \times 128 \times 128$ \\
\bottomrule
\end{tabular}
\label{tab:hyperparameters}
\end{sidewaystable}