File size: 46,457 Bytes
9fe9eb3
c85f5af
 
 
 
 
 
9fe9eb3
 
74c27d3
 
c85f5af
74c27d3
9fe9eb3
 
 
b674bdd
 
c85f5af
 
 
9fe9eb3
9d1f79d
 
 
 
 
9fe9eb3
 
 
9d1f79d
74c27d3
b674bdd
 
 
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
74c27d3
c85f5af
 
 
 
 
 
 
 
b674bdd
 
 
 
 
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
b674bdd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c85f5af
 
 
 
b674bdd
c85f5af
b674bdd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c85f5af
 
b674bdd
c85f5af
 
b674bdd
c85f5af
 
b674bdd
c85f5af
 
b674bdd
c85f5af
b674bdd
9fe9eb3
 
c85f5af
9fe9eb3
 
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9fe9eb3
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9fe9eb3
 
 
 
 
 
c85f5af
 
9fe9eb3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c85f5af
9fe9eb3
c85f5af
9fe9eb3
 
 
 
 
 
 
 
 
 
c85f5af
9fe9eb3
 
 
 
c85f5af
9fe9eb3
c85f5af
 
 
 
9fe9eb3
 
 
 
 
 
 
 
 
 
c85f5af
 
 
9fe9eb3
c85f5af
 
9fe9eb3
c85f5af
9fe9eb3
c85f5af
9fe9eb3
c85f5af
9fe9eb3
 
 
 
 
 
 
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9fe9eb3
 
 
 
c85f5af
 
b674bdd
 
c85f5af
b674bdd
 
 
 
 
 
 
 
 
c85f5af
b674bdd
 
 
 
 
 
 
 
 
 
c85f5af
 
b674bdd
 
 
 
 
 
 
 
 
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b674bdd
 
c85f5af
 
 
 
b674bdd
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b674bdd
 
c85f5af
b674bdd
c85f5af
 
b674bdd
c85f5af
 
b674bdd
c85f5af
b674bdd
 
c85f5af
 
 
b674bdd
c85f5af
 
 
 
b674bdd
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b674bdd
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b674bdd
c85f5af
 
b674bdd
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b674bdd
 
 
c85f5af
b674bdd
 
 
c85f5af
 
b674bdd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c85f5af
b674bdd
 
c85f5af
 
 
b674bdd
 
 
 
c85f5af
b674bdd
 
 
c85f5af
b674bdd
 
 
 
 
c85f5af
b674bdd
 
c85f5af
b674bdd
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b674bdd
c85f5af
b674bdd
c85f5af
 
 
 
 
 
 
 
 
 
b674bdd
 
 
c85f5af
9fe9eb3
 
74c27d3
 
 
 
 
 
c85f5af
74c27d3
 
 
 
 
c85f5af
74c27d3
c85f5af
74c27d3
c85f5af
74c27d3
c85f5af
74c27d3
c85f5af
74c27d3
c85f5af
74c27d3
 
 
 
 
 
 
 
c85f5af
74c27d3
 
 
 
c85f5af
 
 
74c27d3
 
 
 
c85f5af
 
74c27d3
 
9fe9eb3
 
 
9d1f79d
c85f5af
 
 
 
 
74c27d3
 
c85f5af
 
74c27d3
c85f5af
 
74c27d3
 
 
c85f5af
 
9d1f79d
c85f5af
 
 
9d1f79d
 
9fe9eb3
c85f5af
 
9fe9eb3
 
c85f5af
9d1f79d
c85f5af
 
 
 
9fe9eb3
 
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
9d1f79d
 
c85f5af
9fe9eb3
c85f5af
 
 
 
 
 
 
9fe9eb3
 
c85f5af
 
b674bdd
c85f5af
 
 
b674bdd
 
c85f5af
 
 
 
 
b674bdd
c85f5af
 
 
 
 
 
b674bdd
c85f5af
 
b674bdd
c85f5af
b674bdd
 
 
c85f5af
b674bdd
 
 
 
c85f5af
 
 
b674bdd
c85f5af
b674bdd
 
 
 
 
 
c85f5af
b674bdd
 
c85f5af
b674bdd
 
 
 
 
 
 
c85f5af
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9fe9eb3
9d1f79d
c85f5af
 
9fe9eb3
 
 
 
 
 
9d1f79d
 
c85f5af
9fe9eb3
c85f5af
 
9fe9eb3
c85f5af
74c27d3
 
 
 
 
c85f5af
9fe9eb3
74c27d3
c85f5af
74c27d3
 
 
9fe9eb3
74c27d3
c85f5af
74c27d3
 
 
 
 
 
 
9fe9eb3
74c27d3
c85f5af
74c27d3
 
 
 
c85f5af
74c27d3
 
 
9fe9eb3
74c27d3
c85f5af
74c27d3
 
 
 
 
 
 
9fe9eb3
74c27d3
c85f5af
 
 
 
 
 
 
74c27d3
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
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
"""
NPH Diagnostic Platform v3.0
Unified Gradio app combining intensity segmentation, YOLO detection,
dual-engine comparison, ensemble scoring, clinical calculator,
multi-slice batch analysis, quality assessment, and report generation.

Author: Matheus Rech, MD
"""

import gradio as gr
import numpy as np
from PIL import Image, ImageFilter, ImageEnhance, ImageOps, ImageDraw
from transformers import pipeline
import cv2
import tempfile
import os
import threading
import logging
import time
import json
from datetime import datetime

from segment_neuroimaging import (
    segment_nph, segment_ventricles, compute_evans_index,
    compute_callosal_angle, compute_temporal_horn_width,
    compute_third_ventricle_width, score_pvh, assess_desh,
    create_overlay, add_annotations, create_comparison,
    preprocess_image, create_roi_mask, morphological_cleanup,
    filter_by_area, Modality, VENTRICLE_THRESHOLDS, CSF_MODE,
    CSFAppearance, COLORS
)

logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# ---- ML models (lazy-loaded) ----
_classifier = None
_detector = None
_segmenter = None

def get_classifier():
    global _classifier
    if _classifier is None:
        _classifier = pipeline("image-classification", model="google/vit-base-patch16-224")
    return _classifier

def get_detector():
    global _detector
    if _detector is None:
        _detector = pipeline("object-detection", model="facebook/detr-resnet-50")
    return _detector

def get_segmenter():
    global _segmenter
    if _segmenter is None:
        _segmenter = pipeline("image-segmentation", model="facebook/detr-resnet-50-panoptic")
    return _segmenter


# ---- YOLO model ----
YOLO_MODEL_PATH = "best.pt"
_yolo_model = None
_yolo_lock = threading.Lock()

YOLO_COLORS = {
    "ventricle":        (0, 150, 255),
    "sylvian_fissure":  (200, 100, 255),
    "tight_convexity":  (255, 150, 100),
    "pvh":              (255, 200, 0),
    "skull_inner":      (200, 200, 200),
}

YOLO_COLOR_HEX = {
    "ventricle":        "#0096FF",
    "sylvian_fissure":  "#C864FF",
    "tight_convexity":  "#FF9664",
    "pvh":              "#FFC800",
    "skull_inner":      "#C8C8C8",
}


def _get_yolo_model():
    global _yolo_model
    if _yolo_model is None:
        with _yolo_lock:
            if _yolo_model is None and os.path.exists(YOLO_MODEL_PATH):
                try:
                    from ultralytics import YOLO
                    _yolo_model = YOLO(YOLO_MODEL_PATH)
                    logger.info("YOLO model loaded from %s", YOLO_MODEL_PATH)
                except Exception as e:
                    logger.error("Failed to load YOLO model: %s", e)
    return _yolo_model


# ===========================================================================
# Shared: NPH scoring
# ===========================================================================

def _compute_nph_score(data: dict) -> dict:
    """Weighted NPH scoring: VSR(40%) + EI(25%) + CA(20%) + DESH(10%) + Sylvian(5%)."""
    score = 0.0
    evans = data.get("evansIndex") or 0.0
    callosal = data.get("callosalAngle")
    desh = data.get("deshScore") or 0
    sylvian = bool(data.get("sylvianDilation"))
    vsr = data.get("vsr")
    triad = data.get("triad") or []
    atrophy = data.get("corticalAtrophy") or "unknown"

    has_vsr = vsr is not None
    has_callosal = callosal is not None

    if has_vsr:
        if vsr > 2.0:
            score += 40 * min((vsr - 2.0) / 2.0, 1)
        if evans > 0.3:
            score += 25 * min((evans - 0.3) / 0.15, 1)
        if has_callosal and callosal < 90:
            score += 20 * min((90 - callosal) / 50, 1)
        score += (desh / 3) * 10
        if sylvian:
            score += 5
    else:
        scale = 100 / 60
        if evans > 0.3:
            score += 25 * scale * min((evans - 0.3) / 0.15, 1)
        if has_callosal and callosal < 90:
            score += 20 * scale * min((90 - callosal) / 50, 1)
        score += (desh / 3) * 10 * scale
        if sylvian:
            score += 5 * scale

    triad_count = sum(1 for v in triad if v)
    if triad_count == 3:
        score = min(score * 1.15, 100)
    elif triad_count == 2:
        score = min(score * 1.05, 100)

    if atrophy == "significant":
        score *= 0.7
    elif atrophy == "moderate":
        score *= 0.85

    score = int(round(min(score, 100)))

    if score >= 75:
        label, color = "Probable NPH", "#ef4444"
        rec = "Strongly consider CSF tap test and neurosurgical referral for VP shunt evaluation."
    elif score >= 50:
        label, color = "Possible NPH", "#f59e0b"
        rec = "CSF tap test recommended. Consider supplementary MRI for DESH confirmation."
    elif score >= 30:
        label, color = "Low Suspicion", "#3b82f6"
        rec = "NPH less likely. Consider alternative diagnoses. Follow-up imaging in 6 months."
    else:
        label, color = "Unlikely NPH", "#6b7280"
        rec = "Ventriculomegaly likely ex-vacuo or other etiology. Investigate alternative causes."

    return {"score": score, "label": label, "color": color, "recommendation": rec}


# ===========================================================================
# Shared: Image quality assessment
# ===========================================================================

def assess_quality(gray):
    """Return a quality dict: sharpness, contrast, noise, overall grade."""
    laplacian_var = cv2.Laplacian(gray, cv2.CV_64F).var()
    contrast = float(gray.std())
    noise_est = 0
    if gray.shape[0] > 3 and gray.shape[1] > 3:
        kernel = np.array([[1, -2, 1], [-2, 4, -2], [1, -2, 1]])
        sigma = np.abs(cv2.filter2D(gray.astype(np.float64), -1, kernel)).sum()
        noise_est = sigma * np.sqrt(0.5 * np.pi) / (6 * (gray.shape[0] - 2) * (gray.shape[1] - 2))

    sharpness_score = min(100, laplacian_var / 5)
    contrast_score = min(100, contrast * 2)
    noise_score = max(0, 100 - noise_est * 10)
    overall = (sharpness_score * 0.4 + contrast_score * 0.35 + noise_score * 0.25)

    if overall >= 70:
        grade = "Good"
    elif overall >= 40:
        grade = "Acceptable"
    else:
        grade = "Poor"

    return {
        "sharpness": round(sharpness_score, 1),
        "contrast": round(contrast_score, 1),
        "noise": round(noise_score, 1),
        "overall": round(overall, 1),
        "grade": grade,
    }


def compute_symmetry_score(mask):
    """Score left-right symmetry of a binary mask (0-100)."""
    h, w = mask.shape[:2]
    mid = w // 2
    left = mask[:, :mid]
    right = np.fliplr(mask[:, mid:mid + left.shape[1]])
    if left.shape != right.shape:
        min_w = min(left.shape[1], right.shape[1])
        left = left[:, :min_w]
        right = right[:, :min_w]
    intersection = np.logical_and(left > 0, right > 0).sum()
    union = np.logical_or(left > 0, right > 0).sum()
    if union == 0:
        return 0.0
    return round(intersection / union * 100, 1)


# ===========================================================================
# Tab 1: Dual-Engine NPH Analysis (the main innovation)
# ===========================================================================

def _run_intensity_engine(image, modality, sensitivity, pixel_spacing):
    """Run the intensity-based segmentation engine."""
    with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as f:
        Image.fromarray(image).save(f.name)
        temp_path = f.name

    try:
        modality_map = {
            "Axial FLAIR": "FLAIR", "Axial T1": "T1", "Axial T2": "T2",
            "Coronal T2": "T2", "Axial T2 FFE": "T2", "Sagittal T1": "T1",
            "CT Head": "CT_HEAD",
        }
        mod_key = modality_map.get(modality, "T1")
        mod = Modality[mod_key]
        is_coronal = "Coronal" in modality

        img_rgb, gray, _ = preprocess_image(temp_path)
        h, w = gray.shape[:2]
        blurred = cv2.GaussianBlur(gray, (5, 5), 0)
        roi_mask = create_roi_mask(blurred, threshold=30)

        orig_thresh = dict(VENTRICLE_THRESHOLDS[mod])
        sens_adj = (sensitivity - 50) / 50.0
        custom_thresholds = dict(orig_thresh)
        if CSF_MODE[mod] == CSFAppearance.DARK:
            custom_thresholds["csf_high"] = max(20, min(120, int(orig_thresh["csf_high"] + sens_adj * 30)))
        else:
            custom_thresholds["csf_low"] = max(100, min(220, int(orig_thresh["csf_low"] - sens_adj * 30)))

        vent_mask = segment_ventricles(gray, mod, roi_mask, custom_thresholds=custom_thresholds)

        if pixel_spacing is None:
            pixel_spacing = round(180.0 / max(w, 256), 2)

        ei_data = compute_evans_index(vent_mask, image_width=w, pixel_spacing_mm=pixel_spacing)
        th_data = compute_temporal_horn_width(vent_mask, pixel_spacing)
        tv_data = compute_third_ventricle_width(vent_mask, pixel_spacing)
        desh_data = assess_desh(vent_mask, gray, roi_mask, mod, pixel_spacing)
        pvh_data = score_pvh(gray, vent_mask) if mod == Modality.FLAIR else None
        ca_data = compute_callosal_angle(vent_mask) if is_coronal else {}

        vent_area = int((vent_mask > 0).sum())
        brain_area = int((roi_mask > 0).sum())
        vb_ratio = round(vent_area / brain_area, 4) if brain_area > 0 else 0

        quality = assess_quality(gray)
        symmetry = compute_symmetry_score(vent_mask)

        # Build overlay
        display_masks = {"ventricles": vent_mask}
        parenchyma = cv2.bitwise_and(roi_mask, cv2.bitwise_not(vent_mask))
        display_masks["parenchyma"] = parenchyma
        if pvh_data and mod == Modality.FLAIR:
            display_masks["pvh"] = pvh_data["pvh_mask"]
        if "sylvian_mask" in desh_data:
            display_masks["sylvian_fissures"] = desh_data["sylvian_mask"]
        if "convexity_mask" in desh_data:
            display_masks["high_convexity_sulci"] = desh_data["convexity_mask"]

        overlay = create_overlay(img_rgb, display_masks, alpha=0.45)
        biomarkers = dict(ei_data)
        biomarkers.update(th_data)
        if pvh_data:
            biomarkers["pvh_grade"] = pvh_data["pvh_grade"]
        biomarkers["is_desh_positive"] = desh_data["is_desh_positive"]
        if ca_data.get("callosal_angle_deg") is not None:
            biomarkers["callosal_angle_deg"] = ca_data["callosal_angle_deg"]

        annotated = add_annotations(overlay, display_masks, f"{modality} -- Intensity Engine", biomarkers)

        # Draw Evans' index line
        row = ei_data.get("measurement_row", 0)
        if row > 0:
            cols = np.where(vent_mask[row, :] > 0)[0]
            if len(cols) > 0:
                minX, maxX = int(cols[0]), int(cols[-1])
                cv2.line(annotated, (minX, row), (maxX, row), (255, 220, 0), 2)
                skull_d = ei_data.get("skull_diameter_px", w)
                cx = w // 2
                hs = skull_d // 2
                cv2.line(annotated, (cx - hs, row + 8), (cx + hs, row + 8), (200, 200, 200), 1)

        return {
            "annotated": annotated,
            "evans_index": ei_data.get("evans_index", 0),
            "frontal_horn_mm": ei_data.get("frontal_horn_width_mm"),
            "skull_diameter_mm": ei_data.get("skull_diameter_mm"),
            "temporal_horn_px": th_data.get("temporal_horn_width_px", 0),
            "temporal_horn_mm": th_data.get("temporal_horn_width_mm"),
            "third_ventricle_px": tv_data.get("third_ventricle_width_px", 0),
            "third_ventricle_mm": tv_data.get("third_ventricle_width_mm"),
            "vb_ratio": vb_ratio,
            "vent_area": vent_area,
            "brain_area": brain_area,
            "desh_positive": desh_data.get("is_desh_positive", False),
            "desh_score": desh_data.get("total_score", 0),
            "desh_ventriculomegaly": desh_data.get("ventriculomegaly_score", 0),
            "desh_sylvian": desh_data.get("sylvian_dilation_score", 0),
            "desh_convexity": desh_data.get("convexity_tightness_score", 0),
            "pvh_grade": pvh_data["pvh_grade"] if pvh_data else None,
            "pvh_ratio": pvh_data["pvh_ratio"] if pvh_data else None,
            "callosal_angle": ca_data.get("callosal_angle_deg"),
            "quality": quality,
            "symmetry": symmetry,
        }
    finally:
        os.unlink(temp_path)


def _run_yolo_engine(image, conf_threshold=0.25):
    """Run the YOLO detection engine."""
    model = _get_yolo_model()
    if model is None:
        return None

    with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as f:
        Image.fromarray(image).save(f.name)
        temp_path = f.name

    try:
        results = model(temp_path, verbose=False)[0]
        h, w = image.shape[:2]
        annotated_img = image.copy()
        boxes = []

        for box in results.boxes:
            conf = float(box.conf[0])
            if conf < conf_threshold:
                continue
            x1, y1, x2, y2 = [int(round(v)) for v in box.xyxy[0].tolist()]
            cls_id = int(box.cls[0])
            cls_name = model.names.get(cls_id, str(cls_id))
            color = YOLO_COLORS.get(cls_name, (255, 255, 255))

            boxes.append({
                "class": cls_name, "x1": x1, "y1": y1, "x2": x2, "y2": y2,
                "confidence": round(conf, 4),
            })
            cv2.rectangle(annotated_img, (x1, y1), (x2, y2), color, 2)
            label = f"{cls_name} {conf:.0%}"
            (lw, lh), _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
            cv2.rectangle(annotated_img, (x1, y1 - lh - 8), (x1 + lw + 4, y1), color, -1)
            cv2.putText(annotated_img, label, (x1 + 2, y1 - 4),
                        cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 1, cv2.LINE_AA)

        # Derive metrics
        ventricle = next((b for b in boxes if b["class"] == "ventricle"), None)
        skull = next((b for b in boxes if b["class"] == "skull_inner"), None)

        if ventricle and skull:
            vent_w = ventricle["x2"] - ventricle["x1"]
            skull_w = skull["x2"] - skull["x1"]
            ei = round(vent_w / skull_w, 4) if skull_w > 0 else 0
        elif ventricle:
            ei = round((ventricle["x2"] - ventricle["x1"]) / w, 4)
        else:
            ei = 0

        desh_classes = {"tight_convexity", "sylvian_fissure", "pvh"}
        detected_desh = {b["class"] for b in boxes if b["class"] in desh_classes}
        sylvian = any(b["class"] == "sylvian_fissure" for b in boxes)
        pvh = any(b["class"] == "pvh" for b in boxes)

        return {
            "annotated": annotated_img,
            "boxes": boxes,
            "evans_index": ei,
            "desh_score": len(detected_desh),
            "sylvian_dilation": sylvian,
            "pvh_detected": pvh,
            "n_detections": len(boxes),
        }
    finally:
        os.unlink(temp_path)


def dual_engine_analyze(image, modality, sensitivity, pixel_spacing_str, yolo_conf):
    """Run BOTH engines and produce comparison + ensemble score."""
    if image is None:
        raise gr.Error("Please upload a brain MRI or CT image first.")

    pixel_spacing = None
    if pixel_spacing_str and pixel_spacing_str.strip():
        try:
            pixel_spacing = float(pixel_spacing_str.strip())
        except ValueError:
            pass

    # Run intensity engine
    t0 = time.time()
    intensity = _run_intensity_engine(image, modality, sensitivity, pixel_spacing)
    t_intensity = round(time.time() - t0, 2)

    # Run YOLO engine
    t0 = time.time()
    yolo = _run_yolo_engine(image, yolo_conf)
    t_yolo = round(time.time() - t0, 2)

    # Ensemble: average the Evans' Index from both engines
    ei_intensity = intensity["evans_index"]
    ei_yolo = yolo["evans_index"] if yolo else 0
    if yolo:
        ei_ensemble = round((ei_intensity * 0.6 + ei_yolo * 0.4), 4)
    else:
        ei_ensemble = ei_intensity

    # DESH ensemble
    desh_intensity_score = intensity["desh_score"]
    desh_yolo_score = yolo["desh_score"] if yolo else 0
    desh_ensemble = max(desh_intensity_score, desh_yolo_score)
    sylvian_ensemble = intensity["desh_sylvian"] > 0 or (yolo and yolo["sylvian_dilation"])

    # Compute ensemble NPH score
    score_input = {
        "evansIndex": ei_ensemble,
        "callosalAngle": intensity.get("callosal_angle"),
        "deshScore": desh_ensemble,
        "sylvianDilation": sylvian_ensemble,
        "vsr": None,
        "triad": [],
        "corticalAtrophy": "unknown",
    }
    nph_result = _compute_nph_score(score_input)

    # Build report
    q = intensity["quality"]
    lines = []
    lines.append("# Dual-Engine NPH Analysis Report\n")
    lines.append(f"**Date:** {datetime.now().strftime('%Y-%m-%d %H:%M')} | **Modality:** {modality}\n")

    lines.append("---\n## Image Quality Assessment\n")
    lines.append(f"| Metric | Score |")
    lines.append(f"|---|---|")
    lines.append(f"| Sharpness | {q['sharpness']}/100 |")
    lines.append(f"| Contrast | {q['contrast']}/100 |")
    lines.append(f"| Noise | {q['noise']}/100 |")
    lines.append(f"| **Overall** | **{q['overall']}/100 ({q['grade']})** |")
    lines.append(f"| Symmetry | {intensity['symmetry']}% |")

    lines.append("\n---\n## Engine Comparison\n")
    lines.append("| Metric | Intensity Engine | YOLO Engine | Ensemble |")
    lines.append("|---|---|---|---|")
    ei_i_status = "abnormal" if ei_intensity > 0.3 else "normal"
    ei_y_status = ("abnormal" if ei_yolo > 0.3 else "normal") if yolo else "N/A"
    ei_e_status = "ABNORMAL" if ei_ensemble > 0.3 else "normal"
    lines.append(f"| Evans' Index | {ei_intensity:.3f} ({ei_i_status}) | {ei_yolo:.3f} ({ei_y_status}) | **{ei_ensemble:.3f} ({ei_e_status})** |")
    lines.append(f"| DESH Score | {desh_intensity_score}/6 | {desh_yolo_score}/3 | {desh_ensemble} (max) |")

    desh_pos_i = "Yes" if intensity["desh_positive"] else "No"
    desh_pos_y = ("Yes" if yolo and yolo["desh_score"] >= 2 else "No") if yolo else "N/A"
    lines.append(f"| DESH Positive | {desh_pos_i} | {desh_pos_y} | -- |")

    if yolo:
        lines.append(f"| Detections | -- | {yolo['n_detections']} objects | -- |")

    pvh_str = f"Grade {intensity['pvh_grade']}/3" if intensity["pvh_grade"] is not None else "N/A (not FLAIR)"
    pvh_y_str = ("Yes" if yolo and yolo["pvh_detected"] else "No") if yolo else "N/A"
    lines.append(f"| PVH | {pvh_str} | {pvh_y_str} | -- |")

    lines.append(f"| Processing Time | {t_intensity}s | {t_yolo}s | -- |")

    lines.append("\n---\n## Intensity Engine Details\n")
    if intensity.get("frontal_horn_mm"):
        lines.append(f"- Frontal horn width: {intensity['frontal_horn_mm']} mm")
    if intensity.get("skull_diameter_mm"):
        lines.append(f"- Skull diameter: {intensity['skull_diameter_mm']} mm")
    if intensity["temporal_horn_px"] > 0:
        mm_str = f" ({intensity['temporal_horn_mm']} mm)" if intensity.get("temporal_horn_mm") else ""
        lines.append(f"- Temporal horn width: {intensity['temporal_horn_px']} px{mm_str}")
    if intensity["third_ventricle_px"] > 0:
        mm_str = f" ({intensity['third_ventricle_mm']} mm)" if intensity.get("third_ventricle_mm") else ""
        lines.append(f"- Third ventricle width: {intensity['third_ventricle_px']} px{mm_str}")
    lines.append(f"- Ventricle/Brain ratio: {intensity['vb_ratio']:.4f} ({intensity['vent_area']}/{intensity['brain_area']} px)")
    lines.append(f"- DESH breakdown: Vent={intensity['desh_ventriculomegaly']}/2, Sylvian={intensity['desh_sylvian']}/2, Convexity={intensity['desh_convexity']}/2")
    if intensity.get("callosal_angle") is not None:
        ca = intensity["callosal_angle"]
        ca_str = "Suggestive" if ca < 90 else ("Indeterminate" if ca < 120 else "Normal")
        lines.append(f"- Callosal angle: {ca:.1f} deg ({ca_str})")
    if intensity["pvh_grade"] is not None:
        lines.append(f"- PVH: Grade {intensity['pvh_grade']}/3 (ratio: {intensity['pvh_ratio']:.4f})")

    if yolo:
        lines.append("\n---\n## YOLO Detection Details\n")
        for b in yolo["boxes"]:
            bw = b["x2"] - b["x1"]
            bh = b["y2"] - b["y1"]
            lines.append(f"- **{b['class']}**: {b['confidence']:.1%} conf, {bw}x{bh} px at ({b['x1']},{b['y1']})")

    lines.append(f"\n---\n## Ensemble NPH Score: **{nph_result['score']}/100 -- {nph_result['label']}**\n")
    lines.append(f"*{nph_result['recommendation']}*")

    report = "\n".join(lines)
    yolo_img = yolo["annotated"] if yolo else np.zeros_like(image)

    return intensity["annotated"], yolo_img, report


# ===========================================================================
# Tab 2: Multi-Slice Batch Analysis
# ===========================================================================

def batch_analyze(files, modality, sensitivity):
    """Analyze multiple slices and aggregate results."""
    if not files:
        raise gr.Error("Please upload one or more brain scan images.")

    results = []
    for f in files:
        img = np.array(Image.open(f.name).convert("RGB"))
        try:
            r = _run_intensity_engine(img, modality, sensitivity, None)
            results.append({
                "file": os.path.basename(f.name),
                "evans_index": r["evans_index"],
                "desh_positive": r["desh_positive"],
                "desh_score": r["desh_score"],
                "vb_ratio": r["vb_ratio"],
                "quality_grade": r["quality"]["grade"],
                "symmetry": r["symmetry"],
            })
        except Exception as e:
            results.append({
                "file": os.path.basename(f.name),
                "evans_index": 0,
                "desh_positive": False,
                "desh_score": 0,
                "vb_ratio": 0,
                "quality_grade": "Error",
                "symmetry": 0,
                "error": str(e),
            })

    # Aggregate
    valid = [r for r in results if "error" not in r]
    if not valid:
        return "All slices failed to process."

    ei_values = [r["evans_index"] for r in valid]
    max_ei = max(ei_values)
    max_ei_slice = valid[ei_values.index(max_ei)]["file"]
    mean_ei = np.mean(ei_values)
    any_desh = any(r["desh_positive"] for r in valid)
    max_desh = max(r["desh_score"] for r in valid)
    mean_vb = np.mean([r["vb_ratio"] for r in valid])

    # Score using the max Evans' Index (worst slice = most diagnostic)
    score_input = {
        "evansIndex": max_ei,
        "deshScore": min(max_desh, 3),
        "sylvianDilation": any_desh,
        "corticalAtrophy": "unknown",
    }
    nph_result = _compute_nph_score(score_input)

    lines = ["# Multi-Slice NPH Analysis\n"]
    lines.append(f"**Slices analyzed:** {len(valid)} / {len(results)}\n")

    lines.append("---\n## Per-Slice Results\n")
    lines.append("| Slice | Evans' Index | V/B Ratio | DESH | Quality | Symmetry |")
    lines.append("|---|---|---|---|---|---|")
    for r in results:
        if "error" in r:
            lines.append(f"| {r['file']} | ERROR | -- | -- | -- | -- |")
        else:
            ei_flag = " **" if r["evans_index"] > 0.3 else ""
            desh_flag = "POS" if r["desh_positive"] else f"{r['desh_score']}/6"
            lines.append(f"| {r['file']} | {r['evans_index']:.3f}{ei_flag} | {r['vb_ratio']:.4f} | {desh_flag} | {r['quality_grade']} | {r['symmetry']}% |")

    lines.append(f"\n---\n## Aggregate Summary\n")
    lines.append(f"- **Max Evans' Index:** {max_ei:.3f} (slice: {max_ei_slice})" + (" -- ABNORMAL" if max_ei > 0.3 else ""))
    lines.append(f"- **Mean Evans' Index:** {mean_ei:.3f}")
    lines.append(f"- **Mean V/B Ratio:** {mean_vb:.4f}")
    lines.append(f"- **Max DESH Score:** {max_desh}/6")
    lines.append(f"- **Any DESH Positive:** {'Yes' if any_desh else 'No'}")

    lines.append(f"\n---\n## NPH Score: **{nph_result['score']}/100 -- {nph_result['label']}**\n")
    lines.append(f"*Based on worst-case slice (max EI). {nph_result['recommendation']}*")

    return "\n".join(lines)


# ===========================================================================
# Tab 3: Clinical Scoring Calculator
# ===========================================================================

def compute_clinical_score(
    evans_index, callosal_angle_str, desh_score, sylvian_dilation, vsr_str,
    gait, cognition, urinary, cortical_atrophy
):
    callosal = None
    if callosal_angle_str and callosal_angle_str.strip():
        try:
            callosal = float(callosal_angle_str.strip())
        except ValueError:
            pass

    vsr = None
    if vsr_str and vsr_str.strip():
        try:
            vsr = float(vsr_str.strip())
        except ValueError:
            pass

    triad = [gait, cognition, urinary]
    atrophy_map = {"None/Mild": "none", "Moderate": "moderate", "Significant": "significant"}

    score_data = {
        "evansIndex": evans_index, "callosalAngle": callosal,
        "deshScore": int(desh_score), "sylvianDilation": sylvian_dilation,
        "vsr": vsr, "triad": triad,
        "corticalAtrophy": atrophy_map.get(cortical_atrophy, "unknown"),
    }
    result = _compute_nph_score(score_data)

    lines = [f"# NPH Score: {result['score']}/100", f"## {result['label']}\n", f"{result['recommendation']}\n"]
    lines.append("---\n### Input Summary\n")
    lines.append(f"- **Evans' Index:** {evans_index:.3f}" + (" (>0.3 = abnormal)" if evans_index > 0.3 else ""))
    if callosal is not None:
        lines.append(f"- **Callosal Angle:** {callosal:.1f} deg")
    lines.append(f"- **DESH Score:** {int(desh_score)}/3")
    lines.append(f"- **Sylvian Dilation:** {'Yes' if sylvian_dilation else 'No'}")
    if vsr is not None:
        lines.append(f"- **VSR:** {vsr:.2f}" + (" (>2.0 = strong NPH indicator)" if vsr > 2.0 else ""))
    triad_count = sum(triad)
    lines.append(f"- **Hakim Triad:** {triad_count}/3")
    lines.append(f"- **Cortical Atrophy:** {cortical_atrophy}")

    return "\n".join(lines)


# ===========================================================================
# Tab 4: Report Generator
# ===========================================================================

def generate_report(image, modality, sensitivity, pixel_spacing_str,
                    patient_id, patient_age, clinical_history,
                    gait, cognition, urinary):
    """Generate a structured clinical radiology report."""
    if image is None:
        raise gr.Error("Please upload a brain scan first.")

    pixel_spacing = None
    if pixel_spacing_str and pixel_spacing_str.strip():
        try:
            pixel_spacing = float(pixel_spacing_str.strip())
        except ValueError:
            pass

    intensity = _run_intensity_engine(image, modality, sensitivity, pixel_spacing)
    yolo = _run_yolo_engine(image, 0.25)

    ei = intensity["evans_index"]
    ei_y = yolo["evans_index"] if yolo else None
    ei_ensemble = round((ei * 0.6 + ei_y * 0.4), 4) if ei_y else ei

    triad = [gait, cognition, urinary]
    triad_count = sum(triad)

    score_input = {
        "evansIndex": ei_ensemble,
        "callosalAngle": intensity.get("callosal_angle"),
        "deshScore": intensity["desh_score"],
        "sylvianDilation": intensity["desh_sylvian"] > 0,
        "triad": triad,
        "corticalAtrophy": "unknown",
    }
    nph_result = _compute_nph_score(score_input)

    lines = []
    lines.append("# NEURORADIOLOGY REPORT")
    lines.append("## Normal Pressure Hydrocephalus Assessment\n")
    lines.append("---\n")
    lines.append(f"**Patient ID:** {patient_id or 'Anonymous'}")
    lines.append(f"**Age:** {patient_age or 'Not specified'}")
    lines.append(f"**Date:** {datetime.now().strftime('%Y-%m-%d')}")
    lines.append(f"**Modality:** {modality}")
    lines.append(f"**Clinical History:** {clinical_history or 'Not provided'}\n")

    lines.append("---\n## CLINICAL PRESENTATION\n")
    symptoms = []
    if gait: symptoms.append("gait disturbance")
    if cognition: symptoms.append("cognitive impairment")
    if urinary: symptoms.append("urinary incontinence")
    if symptoms:
        lines.append(f"Patient presents with {', '.join(symptoms)} ({triad_count}/3 Hakim triad components).")
    else:
        lines.append("No specific Hakim triad symptoms reported.")

    lines.append("\n---\n## FINDINGS\n")
    lines.append("### Ventricular System")
    ei_word = "abnormally enlarged" if ei_ensemble > 0.3 else "within normal limits"
    lines.append(f"The lateral ventricles are {ei_word} with an Evans' Index of **{ei_ensemble:.3f}** (normal < 0.3).")
    if intensity.get("frontal_horn_mm"):
        lines.append(f"Frontal horn width measures {intensity['frontal_horn_mm']} mm with a biparietal skull diameter of {intensity['skull_diameter_mm']} mm.")
    lines.append(f"Ventricle-to-brain parenchyma ratio is {intensity['vb_ratio']:.4f}.")

    if intensity["temporal_horn_px"] > 0:
        mm_str = f" ({intensity['temporal_horn_mm']} mm)" if intensity.get("temporal_horn_mm") else ""
        lines.append(f"\nThe temporal horns measure {intensity['temporal_horn_px']} px{mm_str}.")

    if intensity["third_ventricle_px"] > 0:
        mm_str = f" ({intensity['third_ventricle_mm']} mm)" if intensity.get("third_ventricle_mm") else ""
        lines.append(f"Third ventricle width is {intensity['third_ventricle_px']} px{mm_str}.")

    if intensity.get("callosal_angle") is not None:
        ca = intensity["callosal_angle"]
        ca_word = "acutely narrowed, consistent with NPH" if ca < 90 else "within normal range"
        lines.append(f"\nThe callosal angle measures {ca:.1f} degrees ({ca_word}).")

    lines.append("\n### DESH Assessment")
    desh_word = "present" if intensity["desh_positive"] else "not fully met"
    lines.append(f"DESH pattern is **{desh_word}** (score: {intensity['desh_score']}/6).")
    lines.append(f"- Ventriculomegaly: {intensity['desh_ventriculomegaly']}/2")
    lines.append(f"- Sylvian fissure dilation: {intensity['desh_sylvian']}/2")
    lines.append(f"- High convexity tightness: {intensity['desh_convexity']}/2")

    if intensity["pvh_grade"] is not None:
        pvh_desc = {0: "absent", 1: "pencil-thin periventricular rim", 2: "smooth periventricular halo", 3: "irregular extension into deep white matter"}
        lines.append(f"\n### Periventricular Changes")
        lines.append(f"PVH Grade **{intensity['pvh_grade']}/3**: {pvh_desc.get(intensity['pvh_grade'], '')}.")

    lines.append(f"\n### Image Quality")
    q = intensity["quality"]
    lines.append(f"Image quality is {q['grade'].lower()} (score: {q['overall']}/100). Symmetry index: {intensity['symmetry']}%.")

    if yolo:
        lines.append(f"\n### AI Structure Detection (YOLO)")
        lines.append(f"{yolo['n_detections']} structures detected:")
        for b in yolo["boxes"]:
            lines.append(f"- {b['class']}: {b['confidence']:.0%} confidence")

    lines.append(f"\n---\n## IMPRESSION\n")
    lines.append(f"**NPH Assessment Score: {nph_result['score']}/100 -- {nph_result['label']}**\n")

    findings = []
    if ei_ensemble > 0.3:
        findings.append(f"ventriculomegaly (EI={ei_ensemble:.3f})")
    if intensity["desh_positive"]:
        findings.append("DESH pattern")
    if intensity["pvh_grade"] is not None and intensity["pvh_grade"] >= 2:
        findings.append(f"periventricular hyperintensities (Grade {intensity['pvh_grade']})")
    if intensity.get("callosal_angle") is not None and intensity["callosal_angle"] < 90:
        findings.append(f"acute callosal angle ({intensity['callosal_angle']:.0f} deg)")
    if triad_count >= 2:
        findings.append(f"clinical Hakim triad ({triad_count}/3)")

    if findings:
        lines.append(f"Key findings: {', '.join(findings)}.")
    lines.append(f"\n{nph_result['recommendation']}")

    lines.append(f"\n---\n*This report was generated by the NPH Diagnostic Platform (v3.0). ")
    lines.append(f"Measurements from JPEG/PNG images are approximate. For clinical decisions, ")
    lines.append(f"correlate with DICOM-derived measurements and clinical examination.*")
    lines.append(f"\n*Matheus Rech, MD | {datetime.now().strftime('%Y-%m-%d %H:%M')}*")

    return intensity["annotated"], "\n".join(lines)


# ===========================================================================
# Tabs 5-8: Filters & ML Models
# ===========================================================================

def apply_filter(image, effect, intensity):
    if image is None:
        raise gr.Error("Please upload an image first.")
    img = Image.fromarray(image)
    if effect == "Grayscale":
        filtered = ImageOps.grayscale(img).convert("RGB")
        if intensity < 1.0: filtered = Image.blend(img, filtered, intensity)
    elif effect == "Sepia":
        gray = ImageOps.grayscale(img)
        sepia = ImageOps.colorize(gray, "#704214", "#C0A080")
        filtered = Image.blend(img, sepia, intensity)
    elif effect == "Blur":
        filtered = img.filter(ImageFilter.GaussianBlur(radius=max(1, int(intensity * 10))))
    elif effect == "Sharpen":
        filtered = ImageEnhance.Sharpness(img).enhance(1 + intensity * 4)
    elif effect == "Edge Detect":
        filtered = Image.blend(img, img.filter(ImageFilter.FIND_EDGES), intensity)
    elif effect == "Invert":
        filtered = Image.blend(img, ImageOps.invert(img.convert("RGB")), intensity)
    elif effect == "Brightness":
        filtered = ImageEnhance.Brightness(img).enhance(0.5 + intensity * 1.5)
    elif effect == "Contrast":
        filtered = ImageEnhance.Contrast(img).enhance(0.5 + intensity * 2)
    else:
        filtered = img
    return np.array(filtered)


def classify_image(image):
    if image is None:
        raise gr.Error("Please upload an image first.")
    return {r["label"]: r["score"] for r in get_classifier()(Image.fromarray(image))}

def detect_objects(image, threshold):
    if image is None:
        raise gr.Error("Please upload an image first.")
    results = get_detector()(Image.fromarray(image), threshold=threshold)
    return (image, [((r["box"]["xmin"], r["box"]["ymin"], r["box"]["xmax"], r["box"]["ymax"]),
                     f"{r['label']} ({r['score']:.0%})") for r in results])

def segment_image(image):
    if image is None:
        raise gr.Error("Please upload an image first.")
    results = get_segmenter()(Image.fromarray(image))
    return (image, [(np.array(r["mask"]), r["label"]) for r in results])


# ===========================================================================
# Build the UI
# ===========================================================================

CUSTOM_CSS = """
.main-title { text-align: center; margin-bottom: 0.2em; }
.subtitle { text-align: center; color: #666; margin-top: 0; font-size: 0.9em; }
.engine-label { font-weight: 700; font-size: 0.85em; text-transform: uppercase; letter-spacing: 0.05em; }
footer { display: none !important; }
"""

with gr.Blocks(theme=gr.themes.Soft(), css=CUSTOM_CSS, title="NPH Diagnostic Platform") as demo:
    gr.Markdown("# NPH Diagnostic Platform", elem_classes="main-title")
    gr.Markdown(
        "Dual-engine analysis (intensity segmentation + YOLO detection), ensemble scoring, "
        "multi-slice batch processing, clinical calculator, and structured report generation.",
        elem_classes="subtitle"
    )

    # ========== Tab 1: Dual-Engine Analysis ==========
    with gr.Tab("Dual-Engine Analysis"):
        gr.Markdown(
            "### Two Engines, One Diagnosis\n"
            "Runs **intensity-based segmentation** AND **YOLO deep learning detection** on the same image, "
            "compares results side-by-side, and produces an **ensemble NPH score** (weighted 60/40)."
        )
        with gr.Row():
            with gr.Column(scale=1):
                de_input = gr.Image(label="Upload Brain Scan", type="numpy")
                de_modality = gr.Dropdown(
                    choices=["Axial FLAIR", "Axial T1", "Axial T2", "Coronal T2",
                             "Axial T2 FFE", "Sagittal T1", "CT Head"],
                    value="Axial FLAIR", label="Modality / Sequence"
                )
                de_sensitivity = gr.Slider(10, 90, value=50, step=5, label="Sensitivity (%)")
                de_yolo_conf = gr.Slider(0.1, 0.95, value=0.25, step=0.05, label="YOLO Confidence Threshold")
                de_spacing = gr.Textbox(label="Pixel Spacing (mm/px)", placeholder="auto-estimate", value="")
                de_btn = gr.Button("Run Dual-Engine Analysis", variant="primary", size="lg")

            with gr.Column(scale=2):
                with gr.Row():
                    with gr.Column():
                        gr.Markdown("**Intensity Engine**", elem_classes="engine-label")
                        de_intensity_out = gr.Image(label="Segmentation Overlay", type="numpy")
                    with gr.Column():
                        gr.Markdown("**YOLO Engine**", elem_classes="engine-label")
                        de_yolo_out = gr.Image(label="Detection Overlay", type="numpy")

        de_report = gr.Markdown(label="Dual-Engine Report")

        de_btn.click(
            fn=dual_engine_analyze,
            inputs=[de_input, de_modality, de_sensitivity, de_spacing, de_yolo_conf],
            outputs=[de_intensity_out, de_yolo_out, de_report]
        )

        with gr.Accordion("How Ensemble Scoring Works", open=False):
            gr.Markdown(
                "The ensemble combines both engines:\n\n"
                "- **Evans' Index**: Weighted average (60% intensity + 40% YOLO)\n"
                "- **DESH Pattern**: Takes the maximum score from either engine\n"
                "- **Sylvian Dilation**: Positive if either engine detects it\n\n"
                "This approach is more robust than either engine alone -- intensity segmentation "
                "is better at precise boundary delineation, while YOLO is better at detecting "
                "spatial patterns and multiple structures simultaneously."
            )

    # ========== Tab 2: Multi-Slice Batch ==========
    with gr.Tab("Multi-Slice Batch"):
        gr.Markdown(
            "### Batch Analysis Across Multiple Slices\n"
            "Upload multiple axial slices from the same patient. Each slice is analyzed individually, "
            "then results are aggregated. The **worst-case slice** (highest Evans' Index) drives the NPH score."
        )
        with gr.Row():
            with gr.Column():
                batch_files = gr.File(
                    label="Upload Multiple Slices",
                    file_count="multiple",
                    file_types=["image"],
                )
                batch_modality = gr.Dropdown(
                    choices=["Axial FLAIR", "Axial T1", "Axial T2", "CT Head"],
                    value="Axial FLAIR", label="Modality"
                )
                batch_sensitivity = gr.Slider(10, 90, value=50, step=5, label="Sensitivity (%)")
                batch_btn = gr.Button("Analyze All Slices", variant="primary", size="lg")

        batch_report = gr.Markdown(label="Batch Report")
        batch_btn.click(fn=batch_analyze, inputs=[batch_files, batch_modality, batch_sensitivity], outputs=batch_report)

    # ========== Tab 3: NPH Score Calculator ==========
    with gr.Tab("NPH Score Calculator"):
        gr.Markdown(
            "### Clinical NPH Scoring Calculator\n"
            "Enter imaging biomarkers and clinical findings to compute a weighted NPH probability score."
        )
        with gr.Row():
            with gr.Column():
                gr.Markdown("#### Imaging Biomarkers")
                calc_evans = gr.Slider(0.0, 0.6, value=0.30, step=0.01, label="Evans' Index")
                calc_callosal = gr.Textbox(label="Callosal Angle (degrees)", placeholder="e.g. 85", value="")
                calc_desh = gr.Slider(0, 3, value=0, step=1, label="DESH Score (0-3)")
                calc_sylvian = gr.Checkbox(label="Sylvian Fissure Dilation", value=False)
                calc_vsr = gr.Textbox(label="VSR", placeholder="e.g. 2.5", value="")
            with gr.Column():
                gr.Markdown("#### Clinical Findings (Hakim Triad)")
                calc_gait = gr.Checkbox(label="Gait disturbance", value=False)
                calc_cognition = gr.Checkbox(label="Cognitive impairment", value=False)
                calc_urinary = gr.Checkbox(label="Urinary incontinence", value=False)
                gr.Markdown("#### Modifiers")
                calc_atrophy = gr.Radio(["None/Mild", "Moderate", "Significant"], value="None/Mild", label="Cortical Atrophy")
                calc_btn = gr.Button("Calculate NPH Score", variant="primary", size="lg")

        calc_report = gr.Markdown(label="Score Report")
        calc_btn.click(
            fn=compute_clinical_score,
            inputs=[calc_evans, calc_callosal, calc_desh, calc_sylvian, calc_vsr,
                    calc_gait, calc_cognition, calc_urinary, calc_atrophy],
            outputs=calc_report
        )

    # ========== Tab 4: Report Generator ==========
    with gr.Tab("Report Generator"):
        gr.Markdown(
            "### Structured Clinical Report\n"
            "Generates a formal neuroradiology-style NPH assessment report combining imaging analysis "
            "with clinical findings."
        )
        with gr.Row():
            with gr.Column(scale=1):
                rpt_input = gr.Image(label="Upload Brain Scan", type="numpy")
                rpt_modality = gr.Dropdown(
                    choices=["Axial FLAIR", "Axial T1", "Axial T2", "Coronal T2", "CT Head"],
                    value="Axial FLAIR", label="Modality"
                )
                rpt_sensitivity = gr.Slider(10, 90, value=50, step=5, label="Sensitivity (%)")
                rpt_spacing = gr.Textbox(label="Pixel Spacing (mm/px)", placeholder="auto-estimate", value="")
                gr.Markdown("#### Patient Info")
                rpt_id = gr.Textbox(label="Patient ID", placeholder="Anonymous")
                rpt_age = gr.Textbox(label="Age", placeholder="e.g. 72")
                rpt_history = gr.Textbox(label="Clinical History", lines=2, placeholder="e.g. Progressive gait instability...")
                gr.Markdown("#### Hakim Triad")
                rpt_gait = gr.Checkbox(label="Gait disturbance", value=False)
                rpt_cognition = gr.Checkbox(label="Cognitive impairment", value=False)
                rpt_urinary = gr.Checkbox(label="Urinary incontinence", value=False)
                rpt_btn = gr.Button("Generate Report", variant="primary", size="lg")

            with gr.Column(scale=2):
                rpt_overlay = gr.Image(label="Segmentation", type="numpy")
                rpt_text = gr.Markdown(label="Clinical Report")

        rpt_btn.click(
            fn=generate_report,
            inputs=[rpt_input, rpt_modality, rpt_sensitivity, rpt_spacing,
                    rpt_id, rpt_age, rpt_history, rpt_gait, rpt_cognition, rpt_urinary],
            outputs=[rpt_overlay, rpt_text]
        )

    # ========== Tab 5: Browser NPH Detector ==========
    with gr.Tab("NPH Detector (Browser)"):
        gr.Markdown(
            "### Client-Side NPH Pipeline\n"
            "Runs entirely in your browser via JavaScript Canvas API. Zero server dependency."
        )
        gr.HTML(
            value='<iframe src="https://mmrech-nph-detector-js.hf.space" '
                  'width="100%" height="900" frameborder="0" '
                  'allow="clipboard-write" '
                  'style="border-radius: 12px; border: 1px solid #333;"></iframe>',
        )

    # ========== Tab 6: Video Demo ==========
    with gr.Tab("Video Demo"):
        gr.Markdown("### Whole-Brain Segmentation Demo")
        gr.Video(value="examples/hydromorph_whole_brain_segmentation.mp4", label="NPH Segmentation Video", autoplay=False)

    # ========== Tab 7: Filters ==========
    with gr.Tab("Filters & Effects"):
        with gr.Row():
            with gr.Column():
                filter_input = gr.Image(label="Upload Image", type="numpy")
                filter_effect = gr.Dropdown(
                    choices=["Grayscale", "Sepia", "Blur", "Sharpen", "Edge Detect", "Invert", "Brightness", "Contrast"],
                    value="Sepia", label="Effect"
                )
                filter_intensity = gr.Slider(0.0, 1.0, value=0.7, step=0.05, label="Intensity")
                filter_btn = gr.Button("Apply Filter", variant="primary")
            with gr.Column():
                filter_output = gr.Image(label="Result", type="numpy")
        filter_btn.click(fn=apply_filter, inputs=[filter_input, filter_effect, filter_intensity], outputs=filter_output)

    # ========== Tab 8: Classification ==========
    with gr.Tab("Image Classification"):
        with gr.Row():
            with gr.Column():
                cls_input = gr.Image(label="Upload Image", type="numpy")
                cls_btn = gr.Button("Classify", variant="primary")
            with gr.Column():
                cls_output = gr.Label(label="Predictions", num_top_classes=5)
        cls_btn.click(fn=classify_image, inputs=cls_input, outputs=cls_output)

    # ========== Tab 9: Object Detection ==========
    with gr.Tab("Object Detection"):
        with gr.Row():
            with gr.Column():
                det_input = gr.Image(label="Upload Image", type="numpy")
                det_threshold = gr.Slider(0.1, 0.95, value=0.5, step=0.05, label="Confidence Threshold")
                det_btn = gr.Button("Detect Objects", variant="primary")
            with gr.Column():
                det_output = gr.AnnotatedImage(label="Detections")
        det_btn.click(fn=detect_objects, inputs=[det_input, det_threshold], outputs=det_output)

    # ========== Tab 10: Segmentation ==========
    with gr.Tab("Segmentation"):
        with gr.Row():
            with gr.Column():
                seg_input = gr.Image(label="Upload Image", type="numpy")
                seg_btn = gr.Button("Segment", variant="primary")
            with gr.Column():
                seg_output = gr.AnnotatedImage(label="Segmentation Map")
        seg_btn.click(fn=segment_image, inputs=seg_input, outputs=seg_output)

    gr.Markdown(
        "<center style='color: #888; font-size: 0.75em; margin-top: 20px;'>"
        "NPH Diagnostic Platform v3.0 | Matheus Rech, MD | "
        "Built with Gradio + YOLO + Transformers"
        "</center>"
    )

demo.launch()