Initial Commit

README.md

@@ -1,14 +1,126 @@
 ---
-title: Brain Tumor Detection
-emoji: 🏃
-colorFrom: red
-colorTo: red
+title: Brain Tumor MRI Classifier
+emoji: 🧠
+colorFrom: indigo
+colorTo: purple
 sdk: gradio
-sdk_version: 6.10.0
+sdk_version: 5.29.0
 app_file: app.py
 pinned: false
 license: mit
-short_description: Takes a MRI Scan Image and classifies it to tumor detection
+tags:
+  - medical-imaging
+  - brain-tumor
+  - efficientnet
+  - image-classification
+  - pytorch
+  - mri
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Brain Tumor MRI Classifier — EfficientNet-B3
+
+A fine-tuned **EfficientNet-B3** model for 4-class brain tumor classification from MRI scans, achieving **98.98% validation accuracy** and **0.9896 macro F1**.
+
+## Classes
+
+| Class | Description |
+|---|---|
+| Glioma | Tumor originating in glial cells of the brain or spine |
+| Meningioma | Tumor arising from the meninges surrounding the brain |
+| Pituitary Tumor | Tumor in the pituitary gland at the base of the brain |
+| No Tumor | No tumor detected in the MRI scan |
+
+## Model
+
+- **Architecture**: EfficientNet-B3 (ImageNet pretrained) with custom classification head
+- **Head**: `Dropout → Linear(1536, 512) → SiLU → Dropout → Linear(512, 4)`
+- **Input size**: 300 × 300
+- **Training**: Two-phase — backbone frozen for 5 epochs (head LR 1e-3), then full fine-tune with differential LR (backbone 1e-4, head 1e-3)
+- **Schedule**: Cosine decay with 3-epoch linear warmup
+- **Loss**: Class-weighted cross-entropy
+
+## Weights
+
+The model weights (`model.pt`) are hosted in this repository and downloaded automatically on first run via `huggingface_hub`.
+
+To download manually:
+
+```python
+from huggingface_hub import hf_hub_download
+ckpt_path = hf_hub_download(repo_id="your-hf-username/brain-tumor-efficientnet-b3", filename="model.pt")
+```
+
+## Dataset
+
+Trained on a merged dataset from two sources:
+
+- **Figshare Brain Tumor Dataset** — glioma, meningioma, pituitary MRI scans
+- **Kaggle Brain Tumor MRI Dataset** — 4-class dataset with glioma, meningioma, pituitary, no tumor
+
+| Split | Images |
+|---|---|
+| Train | 8,211 |
+| Validation | 2,053 |
+
+## Results
+
+| Metric | Score |
+|---|---|
+| Accuracy | 0.9898 |
+| Macro F1 | 0.9896 |
+| Weighted F1 | 0.9898 |
+
+Per-class F1: Glioma 0.9915 · Meningioma 0.9832 · No Tumor 0.9903 · Pituitary 0.9935
+
+## Usage
+
+```python
+import torch
+import torch.nn as nn
+from torchvision import transforms
+from torchvision.models import efficientnet_b3
+from huggingface_hub import hf_hub_download
+from PIL import Image
+
+class EfficientNetClassifier(nn.Module):
+    def __init__(self, num_classes=4, dropout=0.4):
+        super().__init__()
+        self.backbone = efficientnet_b3(weights=None)
+        in_features = self.backbone.classifier[1].in_features
+        self.backbone.classifier = nn.Sequential(
+            nn.Dropout(p=dropout, inplace=True),
+            nn.Linear(in_features, 512),
+            nn.SiLU(),
+            nn.Dropout(p=dropout / 2),
+            nn.Linear(512, num_classes),
+        )
+    def forward(self, x):
+        return self.backbone(x)
+
+# Load
+ckpt_path = hf_hub_download("your-hf-username/brain-tumor-efficientnet-b3", "model.pt")
+ckpt = torch.load(ckpt_path, map_location="cpu", weights_only=False)
+id_to_label = {int(k): v for k, v in ckpt["id_to_label"].items()}
+
+model = EfficientNetClassifier()
+model.load_state_dict(ckpt["model"])
+model.eval()
+
+# Infer
+transform = transforms.Compose([
+    transforms.Resize((300, 300)),
+    transforms.ToTensor(),
+    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
+])
+
+img = Image.open("mri_scan.jpg").convert("RGB")
+with torch.no_grad():
+    probs = torch.softmax(model(transform(img).unsqueeze(0)), dim=-1)[0]
+    pred  = id_to_label[probs.argmax().item()]
+
+print(pred)
+```
+
+## Disclaimer
+
+This model is intended for **research purposes only** and is not a certified medical diagnostic tool. Do not use for clinical decision-making.

app.py

@@ -0,0 +1,417 @@
+import os
+import torch
+import torch.nn as nn
+from torchvision import transforms
+from torchvision.models import efficientnet_b3
+from PIL import Image
+import gradio as gr
+from huggingface_hub import hf_hub_download
+
+# ── Config ────────────────────────────────────────────────────────
+HF_REPO_ID  = "your-hf-username/brain-tumor-efficientnet-b3"   # <- change to your repo
+CKPT_FILE   = "model.pt"
+DEVICE      = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+MEAN        = [0.485, 0.456, 0.406]
+STD         = [0.229, 0.224, 0.225]
+
+ID_TO_LABEL = {
+    0: "Glioma",
+    1: "Meningioma",
+    2: "Pituitary Tumor",
+    3: "No Tumor",
+}
+
+CLASS_INFO = {
+    "Glioma": {
+        "color": "#ef4444",
+        "desc": "A tumor that originates in the glial cells of the brain or spine. Gliomas account for about 30% of all brain tumors.",
+    },
+    "Meningioma": {
+        "color": "#f97316",
+        "desc": "A tumor that arises from the meninges, the membranes surrounding the brain and spinal cord. Usually benign and slow-growing.",
+    },
+    "Pituitary Tumor": {
+        "color": "#a855f7",
+        "desc": "A tumor in the pituitary gland at the base of the brain. Most are benign and can affect hormone regulation.",
+    },
+    "No Tumor": {
+        "color": "#22c55e",
+        "desc": "No tumor detected in the MRI scan. The brain tissue appears within normal parameters.",
+    },
+}
+
+# ── Model definition (must match training code) ───────────────────
+class EfficientNetClassifier(nn.Module):
+    def __init__(self, num_classes=4, dropout=0.4):
+        super().__init__()
+        self.backbone = efficientnet_b3(weights=None)
+        in_features   = self.backbone.classifier[1].in_features
+        self.backbone.classifier = nn.Sequential(
+            nn.Dropout(p=dropout, inplace=True),
+            nn.Linear(in_features, 512),
+            nn.SiLU(),
+            nn.Dropout(p=dropout / 2),
+            nn.Linear(512, num_classes),
+        )
+
+    def forward(self, x):
+        return self.backbone(x)
+
+
+# ── Load model (cached after first download) ──────────────────────
+def load_model():
+    ckpt_path = hf_hub_download(repo_id=HF_REPO_ID, filename=CKPT_FILE)
+    ckpt      = torch.load(ckpt_path, map_location=DEVICE, weights_only=False)
+
+    n_classes   = ckpt.get("num_classes", 4)
+    img_size    = ckpt.get("img_size",    300)
+    id_to_label = {int(k): v for k, v in ckpt["id_to_label"].items()}
+
+    model = EfficientNetClassifier(n_classes).to(DEVICE)
+    model.load_state_dict(ckpt["model"])
+    model.eval()
+    return model, img_size, id_to_label
+
+
+print("Loading model...")
+model, IMG_SIZE, id_to_label = load_model()
+print(f"Model ready on {DEVICE}")
+
+transform = transforms.Compose([
+    transforms.Resize((IMG_SIZE, IMG_SIZE)),
+    transforms.ToTensor(),
+    transforms.Normalize(MEAN, STD),
+])
+
+
+# ── Inference ─────────────────────────────────────────────────────
+@torch.no_grad()
+def predict(image: Image.Image):
+    if image is None:
+        return None, None
+
+    tensor = transform(image.convert("RGB")).unsqueeze(0).to(DEVICE)
+    logits = model(tensor)
+    probs  = torch.softmax(logits, dim=-1)[0]
+
+    results = {
+        id_to_label[i]: round(probs[i].item(), 4)
+        for i in range(len(id_to_label))
+    }
+
+    top_label = max(results, key=results.get)
+    top_prob  = results[top_label]
+
+    # Normalised label for CLASS_INFO lookup
+    label_key = top_label.replace("pituitary", "Pituitary Tumor").strip()
+    if label_key not in CLASS_INFO:
+        # fallback: title-case match
+        for k in CLASS_INFO:
+            if k.lower() == top_label.lower():
+                label_key = k
+                break
+
+    info  = CLASS_INFO.get(label_key, CLASS_INFO.get(top_label, {}))
+    color = info.get("color", "#ffffff")
+    desc  = info.get("desc",  "")
+
+    confidence_html = f"""
+    <div style="
+        background: #0f0f0f;
+        border: 1px solid #1e1e1e;
+        border-radius: 12px;
+        padding: 24px;
+        font-family: 'DM Sans', sans-serif;
+    ">
+        <div style="margin-bottom: 20px;">
+            <span style="
+                font-size: 11px;
+                font-weight: 600;
+                letter-spacing: 0.12em;
+                color: #555;
+                text-transform: uppercase;
+            ">Diagnosis</span>
+            <div style="
+                font-size: 28px;
+                font-weight: 700;
+                color: {color};
+                margin-top: 6px;
+                letter-spacing: -0.02em;
+            ">{top_label}</div>
+            <div style="
+                font-size: 13px;
+                color: #888;
+                margin-top: 8px;
+                line-height: 1.6;
+            ">{desc}</div>
+        </div>
+
+        <div style="margin-bottom: 20px;">
+            <div style="display:flex; justify-content:space-between; margin-bottom:6px;">
+                <span style="font-size:12px; color:#555; letter-spacing:0.08em; text-transform:uppercase;">Confidence</span>
+                <span style="font-size:14px; font-weight:700; color:{color};">{top_prob*100:.1f}%</span>
+            </div>
+            <div style="background:#1a1a1a; border-radius:4px; height:6px; overflow:hidden;">
+                <div style="
+                    height:100%;
+                    width:{top_prob*100:.1f}%;
+                    background:{color};
+                    border-radius:4px;
+                    transition: width 0.6s ease;
+                "></div>
+            </div>
+        </div>
+
+        <div>
+            <span style="font-size:11px; color:#555; letter-spacing:0.1em; text-transform:uppercase;">All class probabilities</span>
+            <div style="margin-top:12px; display:flex; flex-direction:column; gap:10px;">
+    """
+
+    sorted_results = sorted(results.items(), key=lambda x: x[1], reverse=True)
+    for label, prob in sorted_results:
+        lkey  = label
+        for k in CLASS_INFO:
+            if k.lower() == label.lower():
+                lkey = k
+                break
+        c = CLASS_INFO.get(lkey, {}).get("color", "#444")
+        is_top = label == top_label
+        confidence_html += f"""
+                <div>
+                    <div style="display:flex; justify-content:space-between; margin-bottom:4px;">
+                        <span style="
+                            font-size:13px;
+                            color: {'#fff' if is_top else '#888'};
+                            font-weight: {'600' if is_top else '400'};
+                        ">{label}</span>
+                        <span style="font-size:13px; color:{c}; font-weight:600;">{prob*100:.2f}%</span>
+                    </div>
+                    <div style="background:#1a1a1a; border-radius:3px; height:4px; overflow:hidden;">
+                        <div style="
+                            height:100%;
+                            width:{prob*100:.2f}%;
+                            background:{c};
+                            opacity:{'1' if is_top else '0.5'};
+                            border-radius:3px;
+                        "></div>
+                    </div>
+                </div>
+        """
+
+    confidence_html += """
+            </div>
+        </div>
+
+        <div style="
+            margin-top: 20px;
+            padding-top: 16px;
+            border-top: 1px solid #1e1e1e;
+            font-size: 11px;
+            color: #444;
+            text-align: center;
+        ">
+            For research use only. Not a medical diagnostic tool.
+        </div>
+    </div>
+    """
+
+    return results, confidence_html
+
+
+# ── Custom CSS dark theme ─────────────────────────────────────────
+CSS = """
+@import url('https://fonts.googleapis.com/css2?family=DM+Sans:wght@300;400;500;600;700&family=DM+Mono:wght@400;500&display=swap');
+
+:root {
+    --bg-primary:   #080808;
+    --bg-secondary: #0f0f0f;
+    --bg-card:      #111111;
+    --border:       #1e1e1e;
+    --accent:       #6366f1;
+    --text-primary: #f0f0f0;
+    --text-muted:   #555555;
+}
+
+body, .gradio-container {
+    background: var(--bg-primary) !important;
+    font-family: 'DM Sans', sans-serif !important;
+    color: var(--text-primary) !important;
+}
+
+.gradio-container {
+    max-width: 960px !important;
+    margin: 0 auto !important;
+}
+
+/* Header */
+#header {
+    text-align: center;
+    padding: 48px 24px 32px;
+    border-bottom: 1px solid var(--border);
+    margin-bottom: 32px;
+}
+#header h1 {
+    font-size: 32px;
+    font-weight: 700;
+    letter-spacing: -0.04em;
+    color: var(--text-primary);
+    margin: 0 0 10px;
+}
+#header p {
+    font-size: 14px;
+    color: var(--text-muted);
+    margin: 0;
+    line-height: 1.6;
+}
+#header .badge {
+    display: inline-block;
+    font-family: 'DM Mono', monospace;
+    font-size: 10px;
+    letter-spacing: 0.12em;
+    padding: 4px 10px;
+    border: 1px solid #2a2a2a;
+    border-radius: 4px;
+    color: #666;
+    margin-bottom: 16px;
+    text-transform: uppercase;
+}
+
+/* Cards */
+.card {
+    background: var(--bg-card) !important;
+    border: 1px solid var(--border) !important;
+    border-radius: 12px !important;
+}
+
+/* Upload zone */
+.upload-zone {
+    border: 1.5px dashed #2a2a2a !important;
+    border-radius: 12px !important;
+    background: #0a0a0a !important;
+    min-height: 280px !important;
+    transition: border-color 0.2s ease;
+}
+.upload-zone:hover {
+    border-color: var(--accent) !important;
+}
+
+/* Button */
+#run-btn {
+    background: var(--accent) !important;
+    border: none !important;
+    border-radius: 8px !important;
+    color: #fff !important;
+    font-family: 'DM Sans', sans-serif !important;
+    font-size: 14px !important;
+    font-weight: 600 !important;
+    letter-spacing: 0.04em !important;
+    padding: 12px 28px !important;
+    cursor: pointer !important;
+    transition: opacity 0.2s !important;
+    width: 100% !important;
+}
+#run-btn:hover { opacity: 0.88 !important; }
+
+/* Examples */
+.gr-samples-table td, .gr-samples-table th {
+    background: var(--bg-secondary) !important;
+    border-color: var(--border) !important;
+    color: var(--text-primary) !important;
+}
+
+/* Labels */
+label span {
+    font-family: 'DM Sans', sans-serif !important;
+    font-size: 11px !important;
+    font-weight: 600 !important;
+    letter-spacing: 0.1em !important;
+    text-transform: uppercase !important;
+    color: var(--text-muted) !important;
+}
+
+/* Hide default label on HTML output */
+.result-panel > label { display: none !important; }
+
+/* Footer */
+#footer {
+    text-align: center;
+    padding: 24px;
+    border-top: 1px solid var(--border);
+    margin-top: 32px;
+    font-size: 12px;
+    color: var(--text-muted);
+}
+"""
+
+# ── UI ────────────────────────────────────────────────────────────
+with gr.Blocks(css=CSS, theme=gr.themes.Base(), title="Brain Tumor MRI Classifier") as demo:
+
+    gr.HTML("""
+    <div id="header">
+        <div class="badge">EfficientNet-B3 · 98.98% Val Acc</div>
+        <h1>Brain Tumor MRI Classifier</h1>
+        <p>Upload a brain MRI scan to classify into Glioma, Meningioma, Pituitary Tumor, or No Tumor.<br>
+        Trained on Figshare + Kaggle Brain Tumor datasets · 8,211 training images.</p>
+    </div>
+    """)
+
+    with gr.Row(equal_height=True):
+        with gr.Column(scale=1):
+            image_input = gr.Image(
+                type="pil",
+                label="MRI Scan",
+                elem_classes=["upload-zone"],
+                height=300,
+            )
+            run_btn = gr.Button("Run Classification", elem_id="run-btn")
+
+        with gr.Column(scale=1):
+            result_html = gr.HTML(
+                label="Result",
+                elem_classes=["result-panel"],
+                value="""
+                <div style="
+                    background:#0f0f0f;
+                    border:1px solid #1e1e1e;
+                    border-radius:12px;
+                    padding:24px;
+                    height:300px;
+                    display:flex;
+                    align-items:center;
+                    justify-content:center;
+                    flex-direction:column;
+                    gap:12px;
+                ">
+                    <div style="font-size:32px; opacity:0.15;">⬆</div>
+                    <div style="font-size:13px; color:#444; text-align:center; line-height:1.6;">
+                        Upload an MRI scan and click<br>Run Classification
+                    </div>
+                </div>
+                """,
+            )
+
+    # Hidden label output (used internally, not shown)
+    label_output = gr.Label(visible=False)
+
+    run_btn.click(
+        fn=predict,
+        inputs=[image_input],
+        outputs=[label_output, result_html],
+    )
+    image_input.change(
+        fn=predict,
+        inputs=[image_input],
+        outputs=[label_output, result_html],
+    )
+
+    gr.HTML("""
+    <div id="footer">
+        EfficientNet-B3 fine-tuned for brain tumor classification ·
+        <a href="https://huggingface.co/your-hf-username/brain-tumor-efficientnet-b3"
+           style="color:#6366f1; text-decoration:none;">Model on Hugging Face</a>
+        · For research use only
+    </div>
+    """)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,6 @@
+torch==2.10.0
+torchvision==0.25.0
+gradio==5.29.0
+huggingface_hub==1.4.1
+pillow==12.1.1
+numpy==2.4.2