File size: 23,633 Bytes
f8317f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e7117ed
 
 
 
f8317f9
e7117ed
 
f8317f9
e7117ed
f8317f9
e7117ed
 
 
f8317f9
e7117ed
 
f8317f9
e7117ed
 
 
f8317f9
e7117ed
 
 
 
 
f8317f9
e7117ed
 
 
 
f8317f9
e7117ed
f8317f9
e7117ed
f8317f9
 
e7117ed
f8317f9
e7117ed
f8317f9
e7117ed
 
 
f8317f9
e7117ed
f8317f9
e7117ed
 
 
 
 
f8317f9
e7117ed
 
f8317f9
e7117ed
f8317f9
e7117ed
 
f8317f9
e7117ed
 
 
 
 
 
 
 
f8317f9
e7117ed
 
 
f8317f9
e7117ed
 
 
 
 
 
 
 
acf1d8f
 
 
 
f8317f9
e7117ed
 
 
 
 
f8317f9
e7117ed
 
 
f8317f9
e7117ed
f8317f9
 
e7117ed
f8317f9
e7117ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8317f9
 
 
e7117ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8317f9
 
e7117ed
 
 
 
f8317f9
e7117ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8317f9
 
e7117ed
 
 
f8317f9
e7117ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8317f9
 
e7117ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8317f9
 
e7117ed
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f8317f9
 
 
e7117ed
 
 
 
 
 
 
 
 
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
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
import numpy as np
import tensorflow as tf
import tensorflow_addons as tfa
import tensorflow_mri as tfmri
import tqdm
import os
import pydicom as dicom
import glob
from utils.unet3plusnew import *
from utils.custom_unet_code import *
from pydicom.tag import Tag
import imageio
from PIL import Image
import streamlit as st
import zipfile
from collections import defaultdict
import re
from pydicom.tag import Tag
from collections import defaultdict, Counter
import io, shutil, zipfile, time

# Resizes image
def resize(t1, x, y):
    # Adding new axis for the channels
    t1 = tf.expand_dims(t1, -1)

    im1 = tf.image.resize_with_crop_or_pad(t1, x, y)
    return im1

# Function that normalises image
def norm(t1):
    im1 = t1
    im1 = (im1 - np.min(im1)) / np.max(im1)
    return im1

# Applies debanding model to any number of slices
def apply_debanding_model(input_im, frames=32):

    debanding_model = "./models_final/Deband_model"
    debanding = tf.keras.models.load_model(os.path.join(MODELS_BASE, "Deband_model"), compile=False)
    weights = debanding.get_weights()

    inputs = tf.keras.Input(shape=[None, None, None, 1])
    unet = tfmri.models.UNet3D([32, 64, 128], kernel_size=3, out_channels=1, use_global_residual=False)
    DB = unet(inputs)
    de_banding_model = tf.keras.Model(inputs=inputs, outputs=DB)
    de_banding_model.set_weights(weights)

    de_banded = []
    for i in range(frames):
        temp = de_banding_model.predict(tf.expand_dims(tf.expand_dims(input_im[i], 0), -1), verbose=0)
        de_banded.append(temp)

    return de_banded

# Function that applies deformations to 28 slice data
def deformation_28(x):

    sagittal_deformed = []

    for i in range(28):

        input_img = tf.expand_dims(x[0][0, i, :, :], -1)
        dy = tf.expand_dims(tf.expand_dims(x[1][0, i, :, :], -1), 0)
        dx = tf.expand_dims(tf.expand_dims(x[2][0, i, :, :], -1), 0)

        displacement = tf.concat((dy[0, ...], dx[0, ...]), axis=-1)

        img = tf.image.convert_image_dtype(tf.expand_dims(input_img, 0), tf.dtypes.float32)
        displacement = tf.image.convert_image_dtype(displacement, tf.dtypes.float32)
        dense_img_warp = tfa.image.dense_image_warp(img, displacement)
        im_deformed = tf.squeeze(dense_img_warp, 0)
        sagittal_deformed.append(im_deformed)

    sagittal_deformed = tf.image.convert_image_dtype(sagittal_deformed, tf.dtypes.float32)
    sagittal_deformed = tf.expand_dims(sagittal_deformed, axis=0)

    return sagittal_deformed

# Applies respiratory correction model
def apply_resp_model_28(input_im, frames=32):

    inputs = tf.keras.Input(shape=[None, 256, 128, 1])
    unet = build_3d_unet_resp([None, 256, 128, 1], 2)  # Acts as a deformation field generator
    deformation_fields = unet(inputs)  # Outputs the deformation fields
    lambda_deformation = tf.keras.layers.Lambda(deformation_28)
    out_2 = lambda_deformation([inputs[:, :, :, :, 0], deformation_fields[:, :, :, :, 0], deformation_fields[:, :, :, :, 1]])  # Outputs the deformed volume
    outputs = [deformation_fields, out_2]
    complete_model = tf.keras.Model(inputs=inputs, outputs=outputs)
    complete_model.load_weights(os.path.join(MODELS_BASE, "Resp_Correction_model", "variables", "variables"))

    resp_corrected = []
    deformations = []
    for i in range(frames):

        def_fields, resp_cor = complete_model.predict(input_im[i][:, :, :, :, :], verbose=0)
        resp_corrected.append(resp_cor)
        deformations.append(def_fields)

    return deformations, resp_corrected

# Applies super resolution model
def apply_SR_model(input_im, frames=32):
    E2E_model = os.path.join(MODELS_BASE, "E2E_SR_model")
    E2E = tf.keras.models.load_model(E2E_model, compile=False)
    weights = E2E.get_weights()
    sr_weights = weights[22:]

    inputs = tf.keras.Input(shape=[None, None, None, 1])
    SR_model = build_3d_unet(input_shape=(None, None, None, 1), num_classes=1)
    SR = SR_model(inputs)
    SR_model_done = tf.keras.Model(inputs=inputs, outputs=SR)
    SR_model_done.set_weights(sr_weights)

    super_resed = []
    for i in range(frames):
        super_resed.append(SR_model_done.predict(input_im[i], verbose=0))

    return super_resed

t = Tag(0x0019, 0x10D7)
# Reads in example RT sagittal stack
def load_data_samples(path_to_data):
    sag_volumes = []
    filename = f"{path_to_data}/*"
    if not os.path.exists(path_to_data):
        raise Exception("Error with file path.")
    else:

        clean_ims_1 = []
        locations_1 = []

        clean_ims_final = []
        locations_final = []
        test = sorted(glob.glob(filename))
        for file in test:
            ds = dicom.dcmread(file)

            locations_1.append(ds.SliceLocation)
            clean_ims_1.append(ds.pixel_array)
            if ds[t].value == 30:
                clean_ims_final.append(np.array(clean_ims_1))
                locations_final.append(locations_1)
                clean_ims_1 = []
                locations_1 = []

        final = np.array(clean_ims_final)
        final = np.transpose(final, (1, 0, 2, 3))

    return final


def load_data_samples_from_folder(base_dir, number_of_scans=32):
    """
    Recursively find all DICOM files under the first valid subfolder of base_dir,
    group them by InstanceNumber (time), sort by SliceLocation (z), and return
    a NumPy array of shape (time, z, H, W).
    """
    # 1. Find the real nested folder (skip macOS junk)
    candidates = [
        d for d in os.listdir(base_dir)
        if os.path.isdir(os.path.join(base_dir, d))
        and not d.startswith("._")
        and "__MACOSX" not in d
    ]
    if not candidates:
        st.error("No valid data folder found in ZIP.")
        return np.array([])
    nested_base = os.path.join(base_dir, candidates[0])

    # 2. Recursively collect every file; we'll filter for DICOMs next
    all_paths = glob.glob(os.path.join(nested_base, "**", "*"), recursive=True)
    all_paths = [p for p in all_paths if os.path.isfile(p) and not os.path.basename(p).startswith("._")]

    # 3. Filter valid DICOMs
    dicom_files = []
    for p in all_paths:
        try:
            ds = dicom.dcmread(p, force=True, stop_before_pixels=True)
            if hasattr(ds, "InstanceNumber"):
                dicom_files.append(p)
        except:
            continue

    st.write(f"🧾 Found {len(dicom_files)} DICOM files.")

    if not dicom_files:
        st.error("No valid DICOMs found.")
        return np.array([])

    # 4. Group by InstanceNumber (temporal frames)
    grouped = defaultdict(list)
    for p in dicom_files:
        try:
            ds = dicom.dcmread(p, force=True)
            inst = ds.InstanceNumber
            loc = getattr(ds, "SliceLocation", 0.0)
            grouped[inst].append((loc, ds.pixel_array))
        except:
            continue

    # 5. Build volume up to number_of_scans frames
    vols = []
    for inst in sorted(grouped.keys())[:number_of_scans]:
        slices = grouped[inst]
        # sort along z
        slices.sort(key=lambda x: x[0])
        vols.append([img for _, img in slices])

    volume = np.array(vols)  # shape (T, Z, H, W)
    st.write(f"✅ Found data shape: {volume.shape}")
    return volume


def load_cine_any(
    base_dir: str,
    number_of_scans: int = None,                  # if None, use all detected phases
    private_phase_tag: Tag = Tag(0x0019, 0x10D7), # your private phase tag (if present)
    verbose: bool = True
):
    """
    Universal DICOM cine loader (flat or nested folders).
    Scans recursively from `base_dir`, detects cardiac phases, sorts slices,
    and returns (T, Z, H, W) along with the total number of phases detected.
    Returns:
        volume: np.ndarray with shape (T, Z, H, W)
        num_phases_detected: int (total phases found in the dataset)
    """
    def log(msg):
        if verbose:
            try:
                st.write(msg)
            except Exception:
                print(msg)

    if not os.path.isdir(base_dir):
        raise FileNotFoundError(f"No such directory: {base_dir}")

    # --- Collect candidate files (recursive), skip junk/zips
    candidates = glob.glob(os.path.join(base_dir, "**", "*"), recursive=True)
    candidates = [
        p for p in candidates
        if os.path.isfile(p)
        and "__MACOSX" not in p
        and not os.path.basename(p).startswith("._")
        and not p.lower().endswith(".zip")
    ]
    if not candidates:
        log("No files found under the provided directory.")
        return np.array([]), 0

    # --- Keep only files that parse as DICOM headers
    dicom_files = []
    for p in candidates:
        try:
            _ = dicom.dcmread(p, force=True, stop_before_pixels=True)
            dicom_files.append(p)
        except Exception:
            pass

    log(f"🧾 Candidate DICOM files: {len(dicom_files)}")
    if not dicom_files:
        log("No valid DICOM files found.")
        return np.array([]), 0

    # --- NEW: detect flat folder layout (all files in the same directory)
    dicom_dirs = {os.path.dirname(p) for p in dicom_files}
    is_flat = (len(dicom_dirs) == 1)

    # --- Probe to choose the best phase key
    def _try_get(ds, tag):
        try:
            return ds[tag].value
        except Exception:
            return None

    uniq_priv, uniq_tpi, uniq_inst = set(), set(), set()
    for p in dicom_files[:min(len(dicom_files), 200)]:
        try:
            ds = dicom.dcmread(p, force=True, stop_before_pixels=True)
            v_priv = _try_get(ds, private_phase_tag)
            if v_priv is not None:
                try:
                    uniq_priv.add(int(v_priv))
                except Exception:
                    pass
            if hasattr(ds, "TemporalPositionIdentifier"):
                try:
                    uniq_tpi.add(int(ds.TemporalPositionIdentifier))
                except Exception:
                    pass
            if hasattr(ds, "InstanceNumber"):
                try:
                    uniq_inst.add(int(ds.InstanceNumber))
                except Exception:
                    pass
        except Exception:
            continue

    if len(uniq_priv) > 1:
        phase_key = ("private", private_phase_tag)
    elif len(uniq_tpi) > 1:
        phase_key = ("tpi", None)
    elif len(uniq_inst) > 1:
        phase_key = ("instance", None)
    else:
        log("Could not determine a phase key (no variation in private/TPI/InstanceNumber).")
        return np.array([]), 0

    def _get_phase(ds):
        if phase_key[0] == "private":
            v = _try_get(ds, phase_key[1])
            return int(v) if v is not None else None
        if phase_key[0] == "tpi":
            return int(getattr(ds, "TemporalPositionIdentifier", None)) \
                   if hasattr(ds, "TemporalPositionIdentifier") else None
        if phase_key[0] == "instance":
            return int(getattr(ds, "InstanceNumber", None)) \
                   if hasattr(ds, "InstanceNumber") else None
        return None

    def _get_z(ds):
        z = getattr(ds, "SliceLocation", None)
        if z is None:
            ipp = getattr(ds, "ImagePositionPatient", None)
            if ipp is not None and len(ipp) >= 3:
                try:
                    z = float(ipp[2])
                except Exception:
                    z = 0.0
            else:
                z = 0.0
        return float(z)

    # --- Group by phase; sort by z
    grouped = defaultdict(list)
    for p in dicom_files:
        try:
            ds = dicom.dcmread(p, force=True)
            ph = _get_phase(ds)
            if ph is None:
                continue
            grouped[int(ph)].append((_get_z(ds), ds.pixel_array))
        except Exception:
            continue

    if not grouped:
        log("No groups formed (no phase could be read).")
        return np.array([]), 0

    all_phase_ids = sorted(grouped.keys())
    num_phases_detected = len(all_phase_ids)
    phases_to_use = all_phase_ids if number_of_scans is None else all_phase_ids[:number_of_scans]

    stacks_T, slice_counts = [], []
    for ph in phases_to_use:
        pairs = grouped[ph]
        if not pairs:
            continue
        pairs.sort(key=lambda x: x[0])  # sort by z
        stack = [img for _, img in pairs]  # Z × H × W
        stacks_T.append(stack)
        slice_counts.append(len(stack))

    if not stacks_T:
        log("Groups existed but none had readable slices.")
        return np.array([]), num_phases_detected

    # Harmonize Z across phases (trim to the most common Z)
    if len(set(slice_counts)) > 1:
        common_Z = Counter(slice_counts).most_common(1)[0][0]
        stacks_T = [s[:common_Z] for s in stacks_T if len(s) >= common_Z]
        if not stacks_T:
            log("All phases had inconsistent slice counts.")
            return np.array([]), num_phases_detected

    volume = np.array(stacks_T)  # (T, Z, H, W)

    # --- NEW: flip slice order (Z) if data came from a flat single folder
    if is_flat:
        volume = volume[:, ::-1, :, :]

    log(f"✅ Final volume shape: {volume[0, ...].shape} , Phases detected = {num_phases_detected}")
    return volume, num_phases_detected


def load_data_samples_from_flat_folder(
    base_dir: str,
    number_of_scans: int = 32,
    frame_tag: Tag = Tag(0x0019, 0x10D7)  # private phase tag (adjust if needed)
) -> np.ndarray:
    """
    Robust loader when all DICOMs are under one folder (possibly nested).
    - Steps into the single subfolder if present (ignores upload.zip, macOS junk).
    - Recursively finds DICOMs (even without .dcm extension).
    - Groups by phase from `frame_tag` or fallback (0020,0100).
    - Sorts by SliceLocation/IPPs and returns (Z, T, H, W).
    """
    if not os.path.isdir(base_dir):
        raise FileNotFoundError(f"No such directory: {base_dir}")

    # --- Step 1: if there’s exactly one subfolder (plus upload.zip), dive into it
    entries = [e for e in os.listdir(base_dir) if not e.startswith("._")]
    subdirs = [os.path.join(base_dir, e) for e in entries
               if os.path.isdir(os.path.join(base_dir, e)) and "__MACOSX" not in e]
    # If precisely one subdir, prefer that as root; otherwise use base_dir as-is
    root = subdirs[0] if len(subdirs) == 1 else base_dir

    # --- Step 2: recursively collect candidate files (skip zips and junk)
    candidates = glob.glob(os.path.join(root, "**", "*"), recursive=True)
    candidates = [
        p for p in candidates
        if os.path.isfile(p)
        and not os.path.basename(p).startswith("._")
        and "__MACOSX" not in p
        and not p.lower().endswith(".zip")
    ]
    if not candidates:
        st.error("No files found under the provided directory.")
        return np.array([])

    # --- Step 3: keep only files that parse as DICOM headers
    dicom_files = []
    for p in candidates:
        try:
            ds = dicom.dcmread(p, force=True, stop_before_pixels=True)
            dicom_files.append(p)
        except Exception:
            pass

    st.write(f"🧾 Candidate DICOM files: {len(dicom_files)}")
    if not dicom_files:
        st.error("No valid DICOM files found.")
        return np.array([])

    # --- Helper: determine phase index
    def _get_phase(ds):
        # Preferred: private tag (your dataset)
        if frame_tag in ds:
            try:
                return int(ds[frame_tag].value)
            except Exception:
                pass
        # Fallback: standard TemporalPositionIdentifier (0020,0100)
        if hasattr(ds, "TemporalPositionIdentifier"):
            try:
                return int(ds.TemporalPositionIdentifier)
            except Exception:
                pass
        # Last resort: AcquisitionNumber (not always phase, but useful fallback)
        if hasattr(ds, "AcquisitionNumber"):
            try:
                return int(ds.AcquisitionNumber)
            except Exception:
                pass
        return None

    # --- Step 4: group by phase; sort by z
    grouped = defaultdict(list)
    phase_missing = 0
    for p in dicom_files:
        try:
            ds = dicom.dcmread(p, force=True)
            phase = _get_phase(ds)
            if phase is None:
                phase_missing += 1
                continue
            # z-order: SliceLocation if present else IPP[2] else 0
            z = getattr(ds, "SliceLocation", None)
            if z is None:
                ipp = getattr(ds, "ImagePositionPatient", None)
                if ipp is not None and len(ipp) >= 3:
                    z = float(ipp[2])
                else:
                    z = 0.0
            img = ds.pixel_array
            grouped[int(phase)].append((z, img))
        except Exception:
            continue

    if not grouped:
        st.error(
            "Could not determine a phase tag for any files. "
            "Check for (0019,10D7) or (0020,0100) in your dataset."
        )
        st.write(f"Files missing phase: {phase_missing} / {len(dicom_files)}")
        # Optional: show attributes of one file to discover tags
        try:
            ds0 = dicom.dcmread(dicom_files[0], force=True, stop_before_pixels=True)
            st.write("Sample DICOM attributes:", ds0.dir())
        except Exception:
            pass
        return np.array([])

    # keep up to number_of_scans phases
    phases = sorted(grouped.keys())[:number_of_scans]

    stacks_T = []
    slice_counts = []
    for ph in phases:
        pairs = grouped[ph]
        if not pairs:
            continue
        pairs.sort(key=lambda x: x[0])  # sort by z
        stack = [img for _, img in pairs]  # Z × H × W
        stacks_T.append(stack)
        slice_counts.append(len(stack))

    if not stacks_T:
        st.error("No phases contained readable slices.")
        return np.array([])

    # Harmonize Z across phases (trim to the most common slice count)
    if len(set(slice_counts)) > 1:
        common_Z = Counter(slice_counts).most_common(1)[0][0]
        stacks_T = [s[:common_Z] for s in stacks_T if len(s) >= common_Z]
        if not stacks_T:
            st.error("All phases had inconsistent slice counts.")
            return np.array([])

    vol = np.array(stacks_T)  # (T, Z, H, W)
    st.write(f"✅ Final volume shape: {vol.shape} (T, S, H, W)")
    return vol


def extract_zip(zip_path, extract_to):
    with zipfile.ZipFile(zip_path, 'r') as zip_ref:
        # Filter out __MACOSX and dotfiles
        valid_files = [
            f for f in zip_ref.namelist()
            if "__MACOSX" not in f and not os.path.basename(f).startswith("._")
        ]
        zip_ref.extractall(extract_to, members=valid_files)


def make_gif(path, timepoints, axis=-1, slice=60, frame_rate=30):
    # 1. Locate all .npy files
    all_files = glob.glob(os.path.join(path, "*.npy"))
    print("Found NPY files:", all_files)

    # 2. Group them by prefix
    scan_keys = ['raw', 'debanded', 'resp_cor', '3D_cine']
    groups = {k: [] for k in scan_keys}

    pattern = re.compile(r'(?P<prefix>raw|debanded|resp_cor|3D_cine)_(?P<index>\d+)\.npy')

    for p in all_files:
        fn = os.path.basename(p)
        match = pattern.match(fn)
        if match:
            prefix = match.group("prefix")
            t_idx = int(match.group("index"))
            groups[prefix].append((t_idx, p))

    # 3. Sanity check: do all groups exist & have equal lengths?
    Ts = [len(v) for v in groups.values()]
    print("Group counts:", Ts)
    if not all(T == timepoints for T in Ts):
        raise ValueError(f"Mismatch in timepoints across groups. Expected {timepoints}, got {Ts}")

    for k in groups:
        groups[k].sort(key=lambda x: x[0])  # sort by t_idx

    # 4. Determine normalization range per group
    stats = {}
    for k in scan_keys:
        mins, maxs = [], []
        for _, p in groups[k]:
            vol = np.load(p)
            if axis == -1:
                slice_ = vol[:, :, slice]
            else:
                slice_ = vol[:, slice, :] if axis == 1 else vol[slice, :, :]
            mins.append(slice_.min())
            maxs.append(slice_.max())
        stats[k] = (min(mins), max(maxs))

    # 5. Create frames
    frames = []
    for t in range(timepoints):
        imgs_t = []
        for k in scan_keys:
            _, p = groups[k][t]
            vol = np.load(p).astype(np.float32)

            if axis == 2:
                img = vol[::-1, :, slice]
            elif axis == 1:
                img = vol[:, slice, :]
            elif axis == 0:
                img = vol[slice, :, :]
                img = np.transpose(img[:, ::-1])

            mn, mx = stats[k]
            img = np.clip(img, mn, mx)
            img8 = ((img - mn) / (mx - mn) * 255).astype(np.uint8)
            img8 = img8.T[:, ::-1]  # flip + transpose
            imgs_t.append(img8)

        # Stitch side-by-side
        composite = np.concatenate(imgs_t, axis=1)
        resized = Image.fromarray(composite).resize((composite.shape[1] * 3, composite.shape[0] * 3), Image.NEAREST)
        frames.append(np.array(resized))

    # 6. Save and return
    out_path = os.path.join(path, f"temp.gif")
    imageio.mimsave(out_path, frames, duration=1000 / frame_rate, loop=0)

    return out_path


def to_dicom(cardiac_frames, patient_number):

    filename = "./utils/dicom_headerfile.dcm"
    for file in glob.glob(filename):
        ds = dicom.read_file(file)

    for i in range(cardiac_frames):

        volume = np.load(f'./out_dir/3D_cine_{i}.npy')
        print(f"Volume: {i}")
        for j in range(volume.shape[0]):

            PixelData = volume[j, :, :]

            PixelData = (PixelData * 255).astype(np.uint16)

            Dicoms = ds.copy()

            Dicoms.InstanceNumber = j
            Dicoms.PatientID = 'Mark'
            Dicoms.PatientName = 'Mark'
            Dicoms.StudyDescription = '3D Cine'
            Dicoms.SeriesDescription = 'HR'
            Dicoms.StudyInstanceUID = '1.3.12.2.1107.5.2.41.169828.3001002301121546102500000000' + str(patient_number)

            Dicoms.SliceThickness = 1.5
            Dicoms.Rows = 256
            Dicoms.Columns = 128
            Dicoms.AcquisitionMatrix = [0, 256, 128, 0]
            Dicoms.ImageOrientationPatient = [1.0, 0.0, 0.0, 0.0, 0.0, -1.0]
            Dicoms.SliceLocation = -100.0 + ((j - 1) * Dicoms.SliceThickness)
            Dicoms.SamplesPerPixel = 1

            Dicoms.BitsAllocated = 16
            Dicoms.BitsStored = 12
            Dicoms.HighBit = 11
            Dicoms.PixelRepresentation = 0
            Dicoms.AcquisitionNumber = i
            Dicoms.SeriesNumber = i
            Dicoms.PixelSpacing = [1.5, 1.5]

            Dicoms.SmallestImagePixelValue = 0
            Dicoms.LargestImagePixelValue = 255

            Dicoms.PixelData = PixelData.tobytes()

            Dicoms.SeriesInstanceUID = '1.3.12.2.1107.5.2.41.169828.300100230112154610250000001' + str(i)
            Dicoms.SOPInstanceUID = dicom.uid.generate_uid()
            Dicoms.AcquisitionTime = str(i)
            Dicoms.SeriesTime = str(i)

            dicom.filewriter.dcmwrite(filename=f'./out_dicoms/MARK_PATIENT_{patient_number}_VOL_{i}_SLICE_{j}.dcm', dataset=Dicoms)

    return 42


def zip_dir_to_memory(dir_path: str) -> io.BytesIO:
    buf = io.BytesIO()
    with zipfile.ZipFile(buf, "w", compression=zipfile.ZIP_DEFLATED) as zf:
        for root, _, files in os.walk(dir_path):
            for f in files:
                full = os.path.join(root, f)
                arc = os.path.relpath(full, dir_path)  # keep relative paths in zip
                zf.write(full, arc)
    buf.seek(0)
    return buf