File size: 23,473 Bytes
3c206da
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
app.py β€” BrainScan AI (Gradio Space, standalone, single-file)
================================================================
Model: Hybrid EfficientNet-B3 + Custom ViT (Cross-Modal Attention Fusion)
Repo checkpoint : Marksnb/brain-hybrid-efficientnet-vit
  - hybrid_vit_efficientnet_brain_best.pth  -> model klasifikasi 5 kelas (utama)
  - best_precheck_model.pth                -> model precheck biner
                                               ("apakah gambar ini CT/MRI otak?")

Alur:
  1. Download kedua checkpoint dari Hugging Face Hub saat startup.
  2. Definisikan arsitektur model utama (Hybrid EfficientNet-B3 + ViT).
  3. Load model precheck secara ADAPTIF: mencoba beberapa arsitektur backbone
     kandidat dan memilih yang paling cocok dengan checkpoint (lihat catatan
     di bagian PRECHECK MODEL di bawah -- arsitektur aslinya tidak
     didokumentasikan di repo, jadi ini best-effort & auto-degrade jika
     tidak cocok).
  4. Preprocessing gambar sama seperti saat training (Resize 224 + ImageNet norm).
  5. Inference -> precheck dulu, baru klasifikasi 5 kelas penyakit otak.
  6. Generate attention heatmap (ViT attention block terakhir) sebagai
     visualisasi "area yang difokuskan model" (Explainable AI ringan).

Jalankan lokal:
    pip install -r requirements.txt
    python app.py

Deploy ke HF Space:
    - README.md di root Space (metadata YAML): sdk: gradio, app_file: app.py
    - requirements.txt berisi paket yang dibutuhkan
    - Endpoint REST otomatis tersedia di /gradio_api/call/analyze
      (lihat api_name="analyze" di bagian UI paling bawah)
"""

import os
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision.transforms as T
from PIL import Image
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt

import gradio as gr
from huggingface_hub import hf_hub_download

# ─────────────────────────────────────────────────────────────
# 1. KONFIGURASI
# ─────────────────────────────────────────────────────────────
HF_REPO_ID = "Marksnb/brain-hybrid-efficientnet-vit"
MAIN_CHECKPOINT_FILENAME = "hybrid_vit_efficientnet_brain_best.pth"
PRECHECK_CHECKPOINT_FILENAME = "best_precheck_model.pth"

IMG_SIZE = 224
NUM_CLASSES = 5

CLASSES = [
    "Alzheimer",
    "Intracranial_Hemorrhage",
    "Normal",
    "Stroke_Iskemik",
    "Tumor",
]

CLASS_DISPLAY = {
    "Alzheimer": "Alzheimer",
    "Intracranial_Hemorrhage": "Intracranial Hemorrhage (ICH)",
    "Normal": "Normal",
    "Stroke_Iskemik": "Ischemic Stroke",
    "Tumor": "Brain Tumor",
}

IMAGENET_MEAN = [0.485, 0.456, 0.406]
IMAGENET_STD = [0.229, 0.224, 0.225]

val_transforms = T.Compose([
    T.Resize((IMG_SIZE, IMG_SIZE)),
    T.ToTensor(),
    T.Normalize(mean=IMAGENET_MEAN, std=IMAGENET_STD),
])

# --- Precheck config -------------------------------------------------------
# CATATAN PENTING: arsitektur & urutan kelas model precheck TIDAK
# didokumentasikan di repo HF, jadi ini asumsi. Index 0 = bukan brain scan,
# index 1 = brain scan. Kalau hasil precheck kebalik-balik setelah deploy,
# tinggal tukar dua string ini.
PRECHECK_CLASS_NAMES = ["Bukan_Brain_Scan", "Brain_Scan"]
# Ambang keyakinan minimum supaya precheck menolak gambar (0-1).
PRECHECK_REJECT_THRESHOLD = 0.65

try:
    import spaces  # noqa: F401
    IS_ZEROGPU = True
except ImportError:
    IS_ZEROGPU = False

# Di ZeroGPU Space: GPU baru "muncul" saat fungsi ber-@spaces.GPU dipanggil,
# jadi startup HARUS di CPU dulu. Pindah ke cuda dilakukan per-request.
if IS_ZEROGPU:
    DEVICE = torch.device("cpu")
    RUNTIME_DEVICE = torch.device("cuda")
else:
    DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    RUNTIME_DEVICE = DEVICE


# ─────────────────────────────────────────────────────────────
# 2. ARSITEKTUR MODEL UTAMA (Hybrid EfficientNet-B3 + Custom ViT)
#    (persis sama dengan classifier_model.py di repo Space asli,
#     supaya checkpoint bisa di-load tanpa error missing/unexpected key)
# ─────────────────────────────────────────────────────────────
try:
    from torchvision.models import efficientnet_b3, EfficientNet_B3_Weights
    HAS_WEIGHTS_ENUM = True
except ImportError:
    from torchvision.models import efficientnet_b3
    HAS_WEIGHTS_ENUM = False


class PatchEmbedding(nn.Module):
    def __init__(self, in_channels=1536, patch_size=1, embed_dim=768):
        super().__init__()
        self.proj = nn.Conv2d(in_channels, embed_dim, kernel_size=patch_size, stride=patch_size)

    def forward(self, x):
        x = self.proj(x)
        x = x.flatten(2).transpose(1, 2)
        return x


class MultiHeadSelfAttention(nn.Module):
    def __init__(self, embed_dim=768, num_heads=12, dropout=0.1):
        super().__init__()
        assert embed_dim % num_heads == 0
        self.num_heads = num_heads
        self.head_dim = embed_dim // num_heads
        self.scale = self.head_dim ** -0.5
        self.qkv = nn.Linear(embed_dim, embed_dim * 3)
        self.proj = nn.Linear(embed_dim, embed_dim)
        self.drop = nn.Dropout(dropout)

    def forward(self, x, return_attn: bool = False):
        B, N, C = x.shape
        qkv = (self.qkv(x)
               .reshape(B, N, 3, self.num_heads, self.head_dim)
               .permute(2, 0, 3, 1, 4))
        q, k, v = qkv[0], qkv[1], qkv[2]
        attn = (q @ k.transpose(-2, -1)) * self.scale
        attn = attn.softmax(dim=-1)
        attn = self.drop(attn)
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        if return_attn:
            return x, attn
        return x


class TransformerBlock(nn.Module):
    def __init__(self, embed_dim=768, num_heads=12, mlp_ratio=4.0, dropout=0.1):
        super().__init__()
        self.norm1 = nn.LayerNorm(embed_dim)
        self.attn = MultiHeadSelfAttention(embed_dim, num_heads, dropout)
        self.norm2 = nn.LayerNorm(embed_dim)
        hidden = int(embed_dim * mlp_ratio)
        self.mlp = nn.Sequential(
            nn.Linear(embed_dim, hidden),
            nn.GELU(),
            nn.Dropout(dropout),
            nn.Linear(hidden, embed_dim),
            nn.Dropout(dropout),
        )

    def forward(self, x, return_attn: bool = False):
        if return_attn:
            attn_out, attn_weights = self.attn(self.norm1(x), return_attn=True)
            x = x + attn_out
            x = x + self.mlp(self.norm2(x))
            return x, attn_weights
        x = x + self.attn(self.norm1(x))
        x = x + self.mlp(self.norm2(x))
        return x


class CrossModalAttentionFusion(nn.Module):
    def __init__(self, cnn_dim=1536, vit_dim=768, fusion_dim=512, dropout=0.3):
        super().__init__()
        self.cnn_proj = nn.Linear(cnn_dim, fusion_dim)
        self.vit_proj = nn.Linear(vit_dim, fusion_dim)
        self.attn = nn.Sequential(
            nn.Linear(fusion_dim * 2, fusion_dim),
            nn.ReLU(),
            nn.Linear(fusion_dim, 2),
            nn.Softmax(dim=-1),
        )
        self.norm = nn.LayerNorm(fusion_dim)
        self.drop = nn.Dropout(dropout)

    def forward(self, cnn_feat, vit_feat):
        c = self.cnn_proj(cnn_feat)
        v = self.vit_proj(vit_feat)
        w = self.attn(torch.cat([c, v], dim=-1))
        fused = w[:, 0:1] * c + w[:, 1:2] * v
        fused = self.norm(fused)
        fused = self.drop(fused)
        return fused


class BrainHybridModel(nn.Module):
    def __init__(self, num_classes: int = NUM_CLASSES,
                 vit_embed_dim: int = 768,
                 vit_num_heads: int = 12,
                 vit_num_layers: int = 6,
                 fusion_dim: int = 512,
                 dropout: float = 0.3,
                 freeze_backbone: bool = True,
                 pretrained_backbone: bool = True):
        super().__init__()

        if pretrained_backbone:
            if HAS_WEIGHTS_ENUM:
                backbone = efficientnet_b3(weights=EfficientNet_B3_Weights.DEFAULT)
            else:
                backbone = efficientnet_b3(pretrained=True)
        else:
            backbone = efficientnet_b3(weights=None) if HAS_WEIGHTS_ENUM else efficientnet_b3(pretrained=False)

        self.features = backbone.features
        self.cnn_out = 1536

        self.patch_embed = PatchEmbedding(self.cnn_out, patch_size=1, embed_dim=vit_embed_dim)
        self.cls_token = nn.Parameter(torch.zeros(1, 1, vit_embed_dim))
        nn.init.trunc_normal_(self.cls_token, std=0.02)
        num_patches = (IMG_SIZE // 32) ** 2
        self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, vit_embed_dim))
        nn.init.trunc_normal_(self.pos_embed, std=0.02)
        self.pos_drop = nn.Dropout(dropout)
        self.blocks = nn.ModuleList([
            TransformerBlock(vit_embed_dim, vit_num_heads, dropout=dropout)
            for _ in range(vit_num_layers)
        ])
        self.vit_norm = nn.LayerNorm(vit_embed_dim)

        self.fusion = CrossModalAttentionFusion(
            cnn_dim=self.cnn_out, vit_dim=vit_embed_dim,
            fusion_dim=fusion_dim, dropout=dropout)

        self.classifier = nn.Sequential(
            nn.Linear(fusion_dim, 256),
            nn.GELU(),
            nn.BatchNorm1d(256),
            nn.Dropout(dropout),
            nn.Linear(256, num_classes),
        )

        if freeze_backbone:
            for param in self.features.parameters():
                param.requires_grad = False

    def _encode(self, x):
        feat_map = self.features(x)
        cnn_feat = F.adaptive_avg_pool2d(feat_map, 1).flatten(1)
        patches = self.patch_embed(feat_map)
        cls = self.cls_token.expand(x.size(0), -1, -1)
        tokens = torch.cat([cls, patches], dim=1)
        tokens = tokens + self.pos_embed
        tokens = self.pos_drop(tokens)
        return cnn_feat, tokens

    def forward(self, x):
        cnn_feat, tokens = self._encode(x)
        for blk in self.blocks:
            tokens = blk(tokens)
        tokens = self.vit_norm(tokens)
        vit_feat = tokens[:, 0]
        fused = self.fusion(cnn_feat, vit_feat)
        logits = self.classifier(fused)
        return logits

    def forward_with_attention(self, x):
        cnn_feat, tokens = self._encode(x)
        last_attn = None
        for i, blk in enumerate(self.blocks):
            if i == len(self.blocks) - 1:
                tokens, last_attn = blk(tokens, return_attn=True)
            else:
                tokens = blk(tokens)
        tokens = self.vit_norm(tokens)
        vit_feat = tokens[:, 0]
        fused = self.fusion(cnn_feat, vit_feat)
        logits = self.classifier(fused)
        return logits, last_attn


def _unwrap_state_dict(raw):
    """Beberapa checkpoint disimpan sebagai dict {'model_state_dict': ...} atau
    {'state_dict': ...}. Fungsi ini menormalkannya menjadi state_dict polos,
    dan membuang prefix 'module.' (umum kalau training pakai DataParallel)."""
    if isinstance(raw, dict):
        for key in ("model_state_dict", "state_dict", "model"):
            if key in raw and isinstance(raw[key], dict):
                raw = raw[key]
                break
    cleaned = {}
    for k, v in raw.items():
        cleaned[k.replace("module.", "", 1) if k.startswith("module.") else k] = v
    return cleaned


# ─────────────────────────────────────────────────────────────
# 3. MODEL PRECHECK (biner: brain scan vs bukan)
#    ARSITEKTUR ASLINYA TIDAK DIDOKUMENTASIKAN DI REPO -> kita coba beberapa
#    backbone ringan yang umum dipakai untuk precheck/gatekeeper model, dan
#    pilih otomatis yang paling cocok (paling sedikit missing/unexpected key)
#    dengan checkpoint. Kalau tidak ada yang cukup cocok, precheck otomatis
#    dimatikan (app tetap jalan, hanya tanpa langkah precheck).
# ─────────────────────────────────────────────────────────────
def _build_precheck_candidates(num_out=2):
    """Kembalikan list (nama, model) kandidat arsitektur backbone ringan
    dengan output akhir num_out kelas."""
    import torchvision.models as tvm
    candidates = []

    def safe(name, fn):
        try:
            candidates.append((name, fn()))
        except Exception as e:
            print(f"[precheck] Lewati kandidat '{name}': {e}")

    def make_resnet18():
        m = tvm.resnet18(weights=None)
        m.fc = nn.Linear(m.fc.in_features, num_out)
        return m

    def make_resnet34():
        m = tvm.resnet34(weights=None)
        m.fc = nn.Linear(m.fc.in_features, num_out)
        return m

    def make_mobilenet_v2():
        m = tvm.mobilenet_v2(weights=None)
        m.classifier[-1] = nn.Linear(m.classifier[-1].in_features, num_out)
        return m

    def make_efficientnet_b0():
        m = tvm.efficientnet_b0(weights=None)
        m.classifier[-1] = nn.Linear(m.classifier[-1].in_features, num_out)
        return m

    def make_densenet121():
        m = tvm.densenet121(weights=None)
        m.classifier = nn.Linear(m.classifier.in_features, num_out)
        return m

    safe("resnet18", make_resnet18)
    safe("resnet34", make_resnet34)
    safe("mobilenet_v2", make_mobilenet_v2)
    safe("efficientnet_b0", make_efficientnet_b0)
    safe("densenet121", make_densenet121)
    return candidates


def load_precheck_model(checkpoint_path, device):
    """Coba beberapa arsitektur kandidat, pilih yang paling cocok dengan
    checkpoint. Return (model_or_None, info_string)."""
    try:
        raw = torch.load(checkpoint_path, map_location="cpu")
    except Exception as e:
        return None, f"Gagal membaca checkpoint precheck: {e}"

    state_dict = _unwrap_state_dict(raw)
    total_keys = max(len(state_dict), 1)

    best = None  # (score, name, model)
    for name, model in _build_precheck_candidates():
        model_keys = set(model.state_dict().keys())
        missing, unexpected = model.load_state_dict(state_dict, strict=False)
        # `missing`/`unexpected` di sini adalah namedtuple hasil load_state_dict
        n_bad = len(missing) + len(unexpected)
        score = 1.0 - (n_bad / total_keys)
        print(f"[precheck] Kandidat '{name}': score={score:.3f} "
              f"(missing={len(missing)}, unexpected={len(unexpected)})")
        if best is None or score > best[0]:
            best = (score, name, model)

    if best is None:
        return None, "Tidak ada kandidat arsitektur yang bisa dibangun."

    score, name, model = best
    if score < 0.9:
        return None, (f"Precheck dinonaktifkan: arsitektur checkpoint tidak "
                       f"cocok dengan kandidat manapun (skor terbaik={score:.2f}, "
                       f"kandidat={name}). Cek log server untuk detail key yang "
                       f"tidak cocok, lalu sesuaikan _build_precheck_candidates().")

    model.to(device)
    model.eval()
    return model, f"Precheck aktif menggunakan arsitektur '{name}' (skor kecocokan={score:.2f})."


# ─────────────────────────────────────────────────────────────
# 4. LOAD MODEL (sekali saat startup)
# ─────────────────────────────────────────────────────────────
print(f"[startup] Downloading '{MAIN_CHECKPOINT_FILENAME}' dari {HF_REPO_ID} ...")
main_checkpoint_path = hf_hub_download(repo_id=HF_REPO_ID, filename=MAIN_CHECKPOINT_FILENAME)
print(f"[startup] Checkpoint utama tersimpan di: {main_checkpoint_path}")

print(f"[startup] Downloading '{PRECHECK_CHECKPOINT_FILENAME}' dari {HF_REPO_ID} ...")
precheck_checkpoint_path = hf_hub_download(repo_id=HF_REPO_ID, filename=PRECHECK_CHECKPOINT_FILENAME)
print(f"[startup] Checkpoint precheck tersimpan di: {precheck_checkpoint_path}")

model = BrainHybridModel().to(DEVICE)
raw_state = torch.load(main_checkpoint_path, map_location=DEVICE)
main_state_dict = _unwrap_state_dict(raw_state)
missing, unexpected = model.load_state_dict(main_state_dict, strict=False)
if missing:
    print(f"[startup] WARNING - model utama, missing keys: {missing}")
if unexpected:
    print(f"[startup] WARNING - model utama, unexpected keys: {unexpected}")
model.eval()
print(f"[startup] Model utama siap. Device: {DEVICE}")

precheck_model, precheck_info = load_precheck_model(precheck_checkpoint_path, DEVICE)
PRECHECK_ENABLED = precheck_model is not None
print(f"[startup] {precheck_info}")


# ─────────────────────────────────────────────────────────────
# 5. FUNGSI INFERENCE + ATTENTION HEATMAP
# ─────────────────────────────────────────────────────────────
def generate_attention_overlay(orig_image: Image.Image, attn: torch.Tensor):
    """Buat gambar overlay heatmap attention (ViT) di atas gambar asli."""
    avg_attn = attn.squeeze(0).mean(dim=0)          # [seq_len, seq_len]
    cls_attn = avg_attn[0, 1:]                       # attention CLS -> semua patch

    num_patches = int(cls_attn.shape[0] ** 0.5)
    heatmap = cls_attn.reshape(num_patches, num_patches).detach().cpu().numpy()

    heatmap = np.maximum(heatmap, 0)
    heatmap = heatmap / (np.max(heatmap) if np.max(heatmap) != 0 else 1.0)

    heatmap_img = Image.fromarray((heatmap * 255).astype(np.uint8))
    heatmap_resized = np.array(
        heatmap_img.resize(orig_image.size, Image.Resampling.BILINEAR)
    ) / 255.0

    fig, ax = plt.subplots(figsize=(5, 5))
    ax.imshow(orig_image)
    ax.imshow(heatmap_resized, cmap="jet", alpha=0.45)
    ax.axis("off")
    ax.set_title("Peta Fokus Atensi AI (ViT Attention)")
    fig.tight_layout()

    fig.canvas.draw()
    # buffer_rgba() kompatibel dengan matplotlib versi baru (tostring_rgb
    # sudah deprecated/dihapus di beberapa versi terbaru).
    buf = np.asarray(fig.canvas.buffer_rgba())
    overlay_img = Image.fromarray(buf).convert("RGB")
    plt.close(fig)
    return overlay_img


def _run_precheck(tensor_image: torch.Tensor):
    """Return (is_brain_scan: bool, confidence: float, label_scores: dict)."""
    with torch.no_grad():
        logits = precheck_model(tensor_image)
        probs = F.softmax(logits, dim=1).squeeze(0).cpu().numpy()
    pred_idx = int(np.argmax(probs))
    label_scores = {PRECHECK_CLASS_NAMES[i]: float(p) for i, p in enumerate(probs)}
    is_brain_scan = (pred_idx == 1)
    confidence = float(probs[pred_idx])
    return is_brain_scan, confidence, label_scores


def _analyze_brain_scan_impl(image: Image.Image):
    if image is None:
        return None, None, "Silakan upload gambar CT-Scan / MRI otak terlebih dahulu."

    infer_device = RUNTIME_DEVICE if IS_ZEROGPU else DEVICE
    model.to(infer_device)
    if PRECHECK_ENABLED:
        precheck_model.to(infer_device)

    orig_image = image.convert("RGB")
    tensor_image = val_transforms(orig_image).unsqueeze(0).to(infer_device)

    precheck_note = ""
    if PRECHECK_ENABLED:
        is_brain_scan, pc_conf, _ = _run_precheck(tensor_image)
        if (not is_brain_scan) and pc_conf >= PRECHECK_REJECT_THRESHOLD:
            warning = (
                f"⚠️ **Gambar ini kemungkinan BUKAN CT-Scan/MRI otak** "
                f"(keyakinan precheck {pc_conf * 100:.1f}%).\n\n"
                f"Model klasifikasi utama tidak dijalankan karena gambar tidak "
                f"lolos precheck. Silakan upload ulang dengan gambar CT-Scan "
                f"atau MRI otak yang valid."
            )
            return None, None, warning
        precheck_note = f"βœ… Precheck: gambar terdeteksi sebagai brain scan (keyakinan {pc_conf * 100:.1f}%).\n\n"
    else:
        precheck_note = "ℹ️ Precheck dinonaktifkan (lihat log server untuk detail).\n\n"

    with torch.no_grad():
        logits, attn = model.forward_with_attention(tensor_image)
    probs = F.softmax(logits, dim=1).squeeze(0).cpu().numpy()

    pred_idx = int(np.argmax(probs))
    pred_class = CLASSES[pred_idx]
    pred_label = CLASS_DISPLAY[pred_class]
    confidence = float(probs[pred_idx]) * 100

    label_scores = {CLASS_DISPLAY[c]: float(p) for c, p in zip(CLASSES, probs)}

    overlay_img = generate_attention_overlay(orig_image, attn)

    summary = (
        f"{precheck_note}"
        f"**Prediksi: {pred_label}**  (keyakinan {confidence:.2f}%)\n\n"
        f"Catatan: hasil ini adalah output model AI, BUKAN diagnosis medis resmi. "
        f"Selalu konsultasikan dengan dokter/radiolog untuk keputusan klinis."
    )

    return label_scores, overlay_img, summary


if IS_ZEROGPU:
    @spaces.GPU(duration=60)
    def analyze_brain_scan(image: Image.Image):
        return _analyze_brain_scan_impl(image)
else:
    def analyze_brain_scan(image: Image.Image):
        return _analyze_brain_scan_impl(image)


# ─────────────────────────────────────────────────────────────
# 6. UI GRADIO
# ─────────────────────────────────────────────────────────────
with gr.Blocks(title="BrainScan AI β€” Hybrid EfficientNet-ViT") as demo:
    gr.Markdown(
        """
        # 🧠 BrainScan AI
        Klasifikasi otomatis CT-Scan / MRI otak menggunakan arsitektur
        **Hybrid EfficientNet-B3 + Custom Vision Transformer** dengan
        Cross-Modal Attention Fusion, dilengkapi model **precheck** untuk
        memvalidasi apakah gambar yang diupload benar-benar CT/MRI otak.

        Kelas yang dideteksi: Alzheimer, Intracranial Hemorrhage (ICH),
        Normal, Ischemic Stroke, Brain Tumor.

        ⚠️ **Disclaimer:** alat ini untuk tujuan riset/edukasi, bukan pengganti
        diagnosis medis profesional.
        """
    )

    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="pil", label="Upload CT-Scan / MRI Otak")
            analyze_btn = gr.Button("πŸ” Analisis", variant="primary")
        with gr.Column():
            label_output = gr.Label(num_top_classes=5, label="Probabilitas per Kelas")
            heatmap_output = gr.Image(label="Peta Fokus Atensi AI (Explainability)")

    summary_output = gr.Markdown()

    analyze_btn.click(
        fn=analyze_brain_scan,
        inputs=image_input,
        outputs=[label_output, heatmap_output, summary_output],
        api_name="analyze",
    )

    gr.Examples(
        examples=[],  # tambahkan path gambar contoh di sini kalau ada, mis. "samples/normal_1.jpg"
        inputs=image_input,
    )


if __name__ == "__main__":
    demo.launch()