README.md

---
title: BrainScan AI
emoji: 🧠
colorFrom: red
colorTo: blue
sdk: gradio
app_file: app.py
pinned: false
license: mit
---

# BrainScan AI — Hybrid EfficientNet-ViT

Klasifikasi CT-Scan/MRI otak (Alzheimer, Intracranial Hemorrhage, Normal,
Ischemic Stroke, Brain Tumor) menggunakan model hybrid EfficientNet-B3 +
custom Vision Transformer dengan Cross-Modal Attention Fusion.

Model checkpoint diambil otomatis dari:
https://huggingface.co/Marksnb/brain-hybrid-efficientnet-vit

## Cara Pakai lewat Endpoint (API)

Space ini otomatis punya REST API bawaan Gradio di endpoint `/analyze`
(nama ini diatur lewat `api_name="analyze"` di `app.py`). Ganti `SPACE_URL`
di bawah dengan URL Space kamu, contoh: `https://username-brainscan-ai.hf.space`.

### Opsi 1 — Pakai `gradio_client` (Python, paling mudah)

```bash
pip install gradio_client
```

```python
from gradio_client import Client, handle_file

client = Client("SPACE_URL")  # atau "username/brainscan-ai" kalau Space public

result = client.predict(
    handle_file("path/ke/gambar_otak.jpg"),
    api_name="/analyze"
)

label_scores, heatmap_path, summary = result
print(label_scores)   # dict probabilitas tiap kelas
print(summary)        # ringkasan teks prediksi
print(heatmap_path)   # path lokal file heatmap hasil download otomatis
```

Kalau Space kamu private, tambahkan token:

```python
client = Client("SPACE_URL", hf_token="hf_xxxxxxxxxxxxxxxxxxxx")
```

### Opsi 2 — REST API langsung (curl / bahasa apa pun)

Gradio (versi 5+) memakai pola *submit lalu poll*: POST dulu untuk submit job,
lalu GET untuk stream hasilnya pakai `event_id` yang didapat.

**Langkah 1 — Submit gambar (base64):**

```bash
curl -X POST "SPACE_URL/gradio_api/call/analyze" \
  -H "Content-Type: application/json" \
  -d '{
    "data": [
      {
        "path": null,
        "url": "data:image/jpeg;base64,'"$(base64 -w0 gambar_otak.jpg)"'",
        "meta": {"_type": "gradio.FileData"}
      }
    ]
  }'
```

Response berisi `event_id`, contoh:
```json
{"event_id": "abc123..."}
```

**Langkah 2 — Ambil hasil pakai event_id:**

```bash
curl -N "SPACE_URL/gradio_api/call/analyze/abc123..."
```

Response berupa Server-Sent Events (SSE), baris terakhir bertipe `event: complete`
berisi array JSON hasil: `[label_scores, heatmap_fileinfo, summary_markdown]`.

### Opsi 3 — Upload file lewat endpoint upload bawaan Gradio

Kalau gambar berupa file (bukan base64), upload dulu ke endpoint `/gradio_api/upload`,
lalu pakai path hasil upload itu di payload `data` pada langkah submit di atas:

```bash
curl -X POST "SPACE_URL/gradio_api/upload" \
  -F "files=@gambar_otak.jpg"
# -> mengembalikan array path, misal: ["/tmp/xxx/gambar_otak.jpg"]

curl -X POST "SPACE_URL/gradio_api/call/analyze" \
  -H "Content-Type: application/json" \
  -d '{
    "data": [
      {"path": "/tmp/xxx/gambar_otak.jpg", "meta": {"_type": "gradio.FileData"}}
    ]
  }'
```

### Format Output

Endpoint `/analyze` mengembalikan 3 nilai (urut sesuai `outputs=` di `app.py`):

| # | Nilai            | Tipe                          | Keterangan                                   |
|---|-------------------|--------------------------------|-----------------------------------------------|
| 1 | `label_scores`    | `dict[str, float]`             | Probabilitas tiap kelas (0–1)                 |
| 2 | `heatmap`         | file gambar (FileData)         | Overlay peta atensi ViT                       |
| 3 | `summary`         | `string` (markdown)            | Ringkasan prediksi + confidence + disclaimer  |

### Cara paling praktis melihat contoh payload persis

Buka `SPACE_URL/?view=api` di browser (halaman "Use via API" bawaan Gradio) —
di situ tersedia contoh kode Python, JavaScript, dan cURL yang sudah otomatis
disesuaikan dengan skema input/output Space ini.