Upload 5 files

DEPLOYMENT_GUIDE.md

@@ -0,0 +1,132 @@
+# 🚀 DEPLOYMENT GUIDE
+
+## Deploy to Hugging Face Spaces (FREE)
+
+### Step 1: Create Hugging Face Account
+1. Go to https://huggingface.co
+2. Sign up (free)
+3. Verify email
+
+### Step 2: Create New Space
+1. Click your profile → "New Space"
+2. Settings:
+   - **Name:** `anemia-screening` (or your choice)
+   - **License:** MIT
+   - **SDK:** Gradio
+   - **Hardware:** CPU Basic (free) — upgrade to GPU if slow
+   - **Visibility:** Public (or Private)
+3. Click "Create Space"
+
+### Step 3: Upload Files
+Upload these files to your Space:
+
+```
+your-space/
+├── app.py                  # Main application (provided)
+├── requirements.txt        # Dependencies (provided)
+├── README.md              # Space config (provided)
+├── model.pkl              # YOUR trained model from Colab
+└── best.pt                # YOUR YOLO weights
+```
+
+**Option A: Via Web Interface**
+- Click "Files" tab → "Add file" → Upload each file
+
+**Option B: Via Git**
+```bash
+git clone https://huggingface.co/spaces/YOUR_USERNAME/anemia-screening
+cd anemia-screening
+# Copy your files here
+git add .
+git commit -m "Initial deployment"
+git push
+```
+
+### Step 4: Wait for Build
+- Space will automatically install dependencies and start
+- Takes 2-5 minutes
+- Check "Logs" tab if errors occur
+
+### Step 5: Test
+- Once running, test with a sample image
+- Share the URL with others
+
+---
+
+## Files You Need from Colab
+
+After Colab finishes, download from your GDrive `outputs/` folder:
+
+| File | What it is |
+|------|------------|
+| `model.pkl` | Trained ML model |
+
+You already have:
+| File | What it is |
+|------|------------|
+| `best.pt` | YOLO segmentation weights |
+
+---
+
+## Common Issues
+
+### "Model not found"
+- Ensure `model.pkl` and `best.pt` are in the root of your Space (not in a subfolder)
+
+### "Out of memory"
+- Upgrade to GPU hardware (still free tier available)
+- Or reduce image size in `app.py`
+
+### "Module not found"
+- Check `requirements.txt` has all dependencies
+- Restart the Space
+
+### YOLO errors
+- Ensure `best.pt` is a segmentation model (not detection-only)
+- Check ultralytics version compatibility
+
+---
+
+## Custom Domain (Optional)
+
+Hugging Face provides URL like:
+`https://huggingface.co/spaces/YOUR_USERNAME/anemia-screening`
+
+For custom domain (e.g., `anemia.yourdomain.com`):
+1. Go to Space Settings
+2. Add custom domain
+3. Configure DNS CNAME record
+
+---
+
+## Data Collection
+
+Collected data saves to `collected_data.csv` in the Space.
+
+To download:
+1. Go to "Files" tab
+2. Click `collected_data.csv`
+3. Download
+
+For production, connect to Google Sheets or a database instead.
+
+---
+
+## Updating the Model
+
+When you retrain with more data:
+1. Download new `model.pkl` from Colab
+2. Upload to Space (replace old file)
+3. Space auto-restarts
+
+---
+
+## Cost
+
+| Tier | Cost | Specs |
+|------|------|-------|
+| CPU Basic | FREE | 2 vCPU, 16GB RAM |
+| CPU Upgrade | $0.03/hr | 8 vCPU, 32GB RAM |
+| GPU T4 | $0.40/hr | Good for heavy YOLO usage |
+
+For most use cases, **FREE tier is sufficient**.

app.py

@@ -0,0 +1,595 @@
+"""
+NON-INVASIVE ANEMIA SCREENING WEBAPP
+=====================================
+Estimates hemoglobin from palpebral conjunctiva image + standardized form inputs.
+
+Deploy on Hugging Face Spaces (free).
+Collects anonymized data for model improvement.
+
+Author: JIPMER Anemia Screening Study
+"""
+
+import gradio as gr
+import numpy as np
+import pandas as pd
+import pickle
+import cv2
+import uuid
+from datetime import datetime
+from pathlib import Path
+
+# ============================================================
+# CONFIGURATION
+# ============================================================
+MODEL_PATH = "model.pkl"
+YOLO_PATH = "best.pt"
+DATA_LOG_PATH = "collected_data.csv"  # Local fallback; use Google Sheets in production
+
+# ============================================================
+# LOAD MODELS (done once at startup)
+# ============================================================
+def load_models():
+    """Load ML model and YOLO segmentation model."""
+    global ml_model, feature_names, yolo_model
+    
+    # Load ML model
+    with open(MODEL_PATH, 'rb') as f:
+        model_data = pickle.load(f)
+    ml_model = model_data['model']
+    feature_names = model_data['features']
+    print(f"✅ Loaded ML model: {model_data.get('name', 'Unknown')}")
+    print(f"   Features: {len(feature_names)}")
+    
+    # Load YOLO
+    from ultralytics import YOLO
+    yolo_model = YOLO(YOLO_PATH)
+    print(f"✅ Loaded YOLO model")
+    
+    return True
+
+# ============================================================
+# IMAGE PROCESSING
+# ============================================================
+def calibrate_image(img):
+    """White balance calibration using grey card."""
+    try:
+        if img is None:
+            return None
+        
+        # Convert to RGB if needed
+        if len(img.shape) == 2:
+            img_rgb = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)
+        elif img.shape[2] == 4:
+            img_rgb = cv2.cvtColor(img, cv2.COLOR_RGBA2RGB)
+        else:
+            img_rgb = img.copy()
+        
+        gray = cv2.cvtColor(img_rgb, cv2.COLOR_RGB2GRAY)
+        _, thresh = cv2.threshold(gray, 180, 255, cv2.THRESH_BINARY)
+        contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+        
+        if not contours:
+            _, thresh = cv2.threshold(gray, 150, 255, cv2.THRESH_BINARY)
+            contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+        
+        if not contours:
+            return img_rgb
+        
+        c = max(contours, key=cv2.contourArea)
+        x, y, w, h = cv2.boundingRect(c)
+        
+        if w * h < 1000:
+            return img_rgb
+        
+        patch = img_rgb[y+h//4:y+3*h//4, x+w//4:x+3*w//4]
+        if patch.size == 0:
+            return img_rgb
+        
+        avg = np.mean(patch, axis=(0, 1))
+        if any(v < 10 for v in avg):
+            return img_rgb
+        
+        scale = np.array([128, 128, 128]) / avg
+        return np.clip(img_rgb * scale, 0, 255).astype(np.uint8)
+    except Exception as e:
+        print(f"Calibration error: {e}")
+        return img
+
+
+def extract_image_features(img):
+    """Run YOLO segmentation and extract color features."""
+    try:
+        # Calibrate
+        img_cal = calibrate_image(img)
+        if img_cal is None:
+            return None
+        
+        # YOLO segmentation
+        results = yolo_model(img_cal, verbose=False)
+        
+        if len(results) > 0 and results[0].masks is not None:
+            masks = results[0].masks.data.cpu().numpy()
+            confs = results[0].boxes.conf.cpu().numpy()
+            if len(confs) > 0:
+                m = masks[np.argmax(confs)]
+                m = cv2.resize(m, (img_cal.shape[1], img_cal.shape[0]))
+                roi_mask = (m > 0.5).astype(np.uint8) * 255
+            else:
+                # Fallback to center crop
+                h, w = img_cal.shape[:2]
+                roi_mask = np.zeros((h, w), dtype=np.uint8)
+                roi_mask[h//4:3*h//4, w//4:3*w//4] = 255
+        else:
+            h, w = img_cal.shape[:2]
+            roi_mask = np.zeros((h, w), dtype=np.uint8)
+            roi_mask[h//4:3*h//4, w//4:3*w//4] = 255
+        
+        # Extract color features from ROI
+        px = img_cal[roi_mask > 0]
+        if len(px) == 0:
+            return None
+        
+        R, G, B = [np.mean(px[:, i]) for i in range(3)]
+        
+        hsv = cv2.cvtColor(img_cal, cv2.COLOR_RGB2HSV)
+        hpx = hsv[roi_mask > 0]
+        H, S, V = [np.mean(hpx[:, i]) for i in range(3)]
+        
+        lab = cv2.cvtColor(img_cal, cv2.COLOR_RGB2LAB)
+        lpx = lab[roi_mask > 0]
+        L, A, Bc = [np.mean(lpx[:, i]) for i in range(3)]
+        
+        return {
+            'R': R, 'G': G, 'B': B,
+            'H': H, 'S': S, 'V': V,
+            'L_color': L, 'A_color': A, 'B_color': Bc
+        }
+    except Exception as e:
+        print(f"Image processing error: {e}")
+        return None
+
+
+# ============================================================
+# FORM INPUT → FEATURE VECTOR MAPPING
+# ============================================================
+def form_to_features(
+    age, income_range, family_size, diet,
+    food_grains, food_leafy, food_cereals, food_pulses,
+    food_fish, food_eggs, food_meat, food_jaggery, food_nuts,
+    fruit_freq, veg_freq, tea_coffee,
+    comorb_dm, comorb_htn, comorb_thyroid,
+    prev_anemia, family_anemia, hookworm, bleeding_disorder, deworming,
+    menstrual, pads_per_day,
+    gravida, para, live_births, abortions, stillbirths,
+    family_dm, family_htn
+):
+    """Convert standardized form inputs to feature vector."""
+    
+    # Income mapping
+    income_map = {
+        "Below ₹5,000": 2500,
+        "₹5,000 - ₹10,000": 7500,
+        "₹10,000 - ₹15,000": 12500,
+        "₹15,000 - ₹25,000": 20000,
+        "₹25,000 - ₹50,000": 37500,
+        "Above ₹50,000": 75000
+    }
+    income = income_map.get(income_range, 20000)
+    income_per_capita = income / max(family_size, 1)
+    
+    # Diet one-hot
+    diet_veg = 1 if diet == "Vegetarian" else 0
+    diet_nonveg = 1 if diet == "Non-vegetarian" else 0
+    diet_mixed = 1 if diet == "Mixed" else 0
+    
+    # Iron foods
+    food_iron_count = sum([food_grains, food_leafy, food_cereals, food_pulses,
+                          food_fish, food_eggs, food_meat, food_jaggery, food_nuts])
+    
+    # Frequency mapping
+    freq_map = {"Daily": 4, "Twice a week": 3, "Once a week": 2, "Rarely": 1, "Never": 0}
+    fruit_freq_val = freq_map.get(fruit_freq, 2)
+    veg_freq_val = freq_map.get(veg_freq, 2)
+    
+    # Binary mappings
+    tea_coffee_val = 1 if tea_coffee == "Yes" else 0
+    comorb_dm_val = 1 if comorb_dm else 0
+    comorb_htn_val = 1 if comorb_htn else 0
+    comorb_thyroid_val = 1 if comorb_thyroid else 0
+    comorb_none = 1 if not any([comorb_dm, comorb_htn, comorb_thyroid]) else 0
+    comorb_count = comorb_dm_val + comorb_htn_val + comorb_thyroid_val
+    
+    prev_anemia_val = 1 if prev_anemia == "Yes" else 0
+    family_anemia_val = 1 if family_anemia == "Yes" else 0
+    hookworm_val = 1 if hookworm == "Yes" else 0
+    bleeding_val = 1 if bleeding_disorder == "Yes" else 0
+    deworming_val = 1 if deworming == "Yes" else 0
+    
+    # Menstrual
+    menstrual_regular = 1 if menstrual == "Regular" else 0
+    menstrual_irregular = 1 if menstrual == "Irregular" else 0
+    
+    # Pads mapping
+    pads_map = {"1-2": 1.5, "2-3": 2.5, "3-4": 3.5, "4-5": 4.5, "More than 5": 6, "Not applicable": 2.5}
+    pads_val = pads_map.get(pads_per_day, 2.5)
+    
+    # Obstetric
+    has_obstetric = 1 if gravida > 0 else 0
+    pregnancy_loss = abortions + stillbirths
+    parity_ratio = para / gravida if gravida > 0 else 0
+    
+    family_dm_val = 1 if family_dm == "Yes" else 0
+    family_htn_val = 1 if family_htn == "Yes" else 0
+    
+    # Build feature dict (order must match training!)
+    features = {
+        'age': age,
+        'income': income,
+        'family_size': family_size,
+        'income_per_capita': income_per_capita,
+        'diet_veg': diet_veg,
+        'diet_nonveg': diet_nonveg,
+        'diet_mixed': diet_mixed,
+        'food_whole_grains': 1 if food_grains else 0,
+        'food_green_leafy_vegetables': 1 if food_leafy else 0,
+        'food_cereals': 1 if food_cereals else 0,
+        'food_pulses': 1 if food_pulses else 0,
+        'food_fish': 1 if food_fish else 0,
+        'food_eggs': 1 if food_eggs else 0,
+        'food_meat': 1 if food_meat else 0,
+        'food_jaggery': 1 if food_jaggery else 0,
+        'food_nuts': 1 if food_nuts else 0,
+        'food_iron_count': food_iron_count,
+        'fruit_freq': fruit_freq_val,
+        'veg_freq': veg_freq_val,
+        'tea_coffee_daily': tea_coffee_val,
+        'tea_coffee_daily_missing': 0,
+        'comorb_dm': comorb_dm_val,
+        'comorb_htn': comorb_htn_val,
+        'comorb_thyroid': comorb_thyroid_val,
+        'comorb_none': comorb_none,
+        'comorb_count': comorb_count,
+        'prev_anemia': prev_anemia_val,
+        'prev_anemia_missing': 0,
+        'family_anemia': family_anemia_val,
+        'family_anemia_missing': 0,
+        'hookworm_history': hookworm_val,
+        'hookworm_history_missing': 0,
+        'bleeding_disorder': bleeding_val,
+        'bleeding_disorder_missing': 0,
+        'deworming_done': deworming_val,
+        'deworming_done_missing': 0,
+        'menstrual_regular': menstrual_regular,
+        'menstrual_irregular': menstrual_irregular,
+        'pads_per_day': pads_val,
+        'pads_missing': 0,
+        'gravida': gravida,
+        'para': para,
+        'live_births': live_births,
+        'abortions': abortions,
+        'stillbirths': stillbirths,
+        'has_obstetric_history': has_obstetric,
+        'pregnancy_loss': pregnancy_loss,
+        'parity_ratio': parity_ratio,
+        'family_dm': family_dm_val,
+        'family_dm_missing': 0,
+        'family_htn': family_htn_val,
+        'family_htn_missing': 0,
+    }
+    
+    return features
+
+
+# ============================================================
+# DATA LOGGING
+# ============================================================
+def log_data(session_id, features, image_features, predicted_hb, actual_hb=None):
+    """Log anonymized data for model improvement."""
+    try:
+        record = {
+            'session_id': session_id,
+            'timestamp': datetime.now().isoformat(),
+            'predicted_hb': predicted_hb,
+            'actual_hb': actual_hb,
+            **features,
+            **image_features
+        }
+        
+        df = pd.DataFrame([record])
+        
+        # Append to CSV (or send to Google Sheets in production)
+        if Path(DATA_LOG_PATH).exists():
+            df.to_csv(DATA_LOG_PATH, mode='a', header=False, index=False)
+        else:
+            df.to_csv(DATA_LOG_PATH, index=False)
+        
+        return True
+    except Exception as e:
+        print(f"Logging error: {e}")
+        return False
+
+
+# ============================================================
+# MAIN PREDICTION FUNCTION
+# ============================================================
+def predict_hb(
+    image,
+    age, income_range, family_size, diet,
+    food_grains, food_leafy, food_cereals, food_pulses,
+    food_fish, food_eggs, food_meat, food_jaggery, food_nuts,
+    fruit_freq, veg_freq, tea_coffee,
+    comorb_dm, comorb_htn, comorb_thyroid,
+    prev_anemia, family_anemia, hookworm, bleeding_disorder, deworming,
+    menstrual, pads_per_day,
+    gravida, para, live_births, abortions, stillbirths,
+    family_dm, family_htn,
+    actual_hb_input
+):
+    """Main prediction function."""
+    
+    # Generate session ID
+    session_id = str(uuid.uuid4())[:8]
+    
+    # Validate image
+    if image is None:
+        return "❌ Please upload an image of the palpebral conjunctiva with color card.", "", ""
+    
+    # Extract image features
+    image_features = extract_image_features(image)
+    if image_features is None:
+        return "❌ Could not process image. Please ensure the palpebral conjunctiva is clearly visible.", "", ""
+    
+    # Convert form to features
+    tabular_features = form_to_features(
+        age, income_range, family_size, diet,
+        food_grains, food_leafy, food_cereals, food_pulses,
+        food_fish, food_eggs, food_meat, food_jaggery, food_nuts,
+        fruit_freq, veg_freq, tea_coffee,
+        comorb_dm, comorb_htn, comorb_thyroid,
+        prev_anemia, family_anemia, hookworm, bleeding_disorder, deworming,
+        menstrual, pads_per_day,
+        gravida, para, live_births, abortions, stillbirths,
+        family_dm, family_htn
+    )
+    
+    # Combine features
+    all_features = {**tabular_features, **image_features}
+    
+    # Create feature vector in correct order
+    X = pd.DataFrame([all_features])[feature_names]
+    
+    # Predict
+    hb_pred = ml_model.predict(X)[0]
+    hb_pred = round(float(hb_pred), 1)
+    
+    # Classify severity
+    if hb_pred >= 12:
+        severity = "Normal"
+        color = "🟢"
+        recommendation = "No signs of anemia detected. Maintain a balanced diet."
+    elif hb_pred >= 11:
+        severity = "Mild Anemia"
+        color = "🟡"
+        recommendation = "Mild anemia indicated. Consider iron-rich foods and consult a healthcare provider."
+    elif hb_pred >= 8:
+        severity = "Moderate Anemia"
+        color = "🟠"
+        recommendation = "Moderate anemia detected. Please consult a healthcare provider for a confirmatory blood test."
+    else:
+        severity = "Severe Anemia"
+        color = "🔴"
+        recommendation = "Severe anemia indicated. Please seek immediate medical attention."
+    
+    # Log data
+    actual_hb = float(actual_hb_input) if actual_hb_input and actual_hb_input > 0 else None
+    log_data(session_id, tabular_features, image_features, hb_pred, actual_hb)
+    
+    # Format result
+    result = f"""
+    ## {color} Estimated Hemoglobin: {hb_pred} g/dL
+    
+    ### Status: {severity}
+    
+    {recommendation}
+    
+    ---
+    *Session ID: {session_id}*  
+    *Save this ID if you want to provide actual Hb value later.*
+    """
+    
+    return result, session_id, f"{hb_pred} g/dL"
+
+
+# ============================================================
+# GRADIO INTERFACE
+# ============================================================
+def create_interface():
+    """Build the Gradio UI."""
+    
+    with gr.Blocks(title="Anemia Screening Tool", theme=gr.themes.Soft()) as app:
+        
+        gr.Markdown("""
+        # 🩺 Non-Invasive Anemia Screening Tool
+        
+        Estimate hemoglobin levels from a photo of the palpebral conjunctiva (inner eyelid).
+        
+        **Instructions:**
+        1. Upload a clear photo of the lower inner eyelid WITH the color calibration card visible
+        2. Fill in the health information form
+        3. Click "Estimate Hemoglobin"
+        
+        ⚠️ *This is a screening tool, not a diagnosis. Please consult a healthcare provider for confirmation.*
+        
+        ---
+        """)
+        
+        with gr.Row():
+            # Left column - Image upload
+            with gr.Column(scale=1):
+                gr.Markdown("### 📷 Image Upload")
+                image_input = gr.Image(
+                    label="Palpebral Conjunctiva Image (with color card)",
+                    type="numpy"
+                )
+                gr.Markdown("*Ensure the grey/white color card is visible in the image for accurate calibration.*")
+            
+            # Right column - Form
+            with gr.Column(scale=2):
+                gr.Markdown("### 📋 Health Information")
+                
+                with gr.Accordion("Demographics", open=True):
+                    with gr.Row():
+                        age = gr.Slider(minimum=10, maximum=80, value=25, step=1, label="Age (years)")
+                        family_size = gr.Slider(minimum=1, maximum=15, value=4, step=1, label="Family Size")
+                    income_range = gr.Dropdown(
+                        choices=["Below ₹5,000", "₹5,000 - ₹10,000", "₹10,000 - ₹15,000", 
+                                "₹15,000 - ₹25,000", "₹25,000 - ₹50,000", "Above ₹50,000"],
+                        value="₹10,000 - ₹15,000",
+                        label="Monthly Family Income"
+                    )
+                
+                with gr.Accordion("Diet & Nutrition", open=True):
+                    diet = gr.Radio(
+                        choices=["Vegetarian", "Non-vegetarian", "Mixed"],
+                        value="Mixed",
+                        label="Dietary Habit"
+                    )
+                    gr.Markdown("**Iron-rich foods consumed regularly:**")
+                    with gr.Row():
+                        food_grains = gr.Checkbox(label="Whole Grains", value=True)
+                        food_leafy = gr.Checkbox(label="Leafy Vegetables", value=True)
+                        food_cereals = gr.Checkbox(label="Cereals", value=True)
+                    with gr.Row():
+                        food_pulses = gr.Checkbox(label="Pulses/Lentils", value=True)
+                        food_fish = gr.Checkbox(label="Fish", value=False)
+                        food_eggs = gr.Checkbox(label="Eggs", value=False)
+                    with gr.Row():
+                        food_meat = gr.Checkbox(label="Meat", value=False)
+                        food_jaggery = gr.Checkbox(label="Jaggery", value=False)
+                        food_nuts = gr.Checkbox(label="Nuts", value=False)
+                    with gr.Row():
+                        fruit_freq = gr.Dropdown(
+                            choices=["Daily", "Twice a week", "Once a week", "Rarely", "Never"],
+                            value="Twice a week",
+                            label="Fruit Consumption"
+                        )
+                        veg_freq = gr.Dropdown(
+                            choices=["Daily", "Twice a week", "Once a week", "Rarely", "Never"],
+                            value="Daily",
+                            label="Vegetable Consumption"
+                        )
+                    tea_coffee = gr.Radio(
+                        choices=["Yes", "No"],
+                        value="Yes",
+                        label="Daily Tea/Coffee Consumption"
+                    )
+                
+                with gr.Accordion("Medical History", open=False):
+                    gr.Markdown("**Current Conditions:**")
+                    with gr.Row():
+                        comorb_dm = gr.Checkbox(label="Diabetes", value=False)
+                        comorb_htn = gr.Checkbox(label="Hypertension", value=False)
+                        comorb_thyroid = gr.Checkbox(label="Thyroid Disorder", value=False)
+                    with gr.Row():
+                        prev_anemia = gr.Radio(choices=["Yes", "No"], value="No", label="Anemia in past 12 months")
+                        family_anemia = gr.Radio(choices=["Yes", "No"], value="No", label="Family history of anemia")
+                    with gr.Row():
+                        hookworm = gr.Radio(choices=["Yes", "No"], value="No", label="History of hookworm")
+                        bleeding_disorder = gr.Radio(choices=["Yes", "No"], value="No", label="Bleeding disorder")
+                    with gr.Row():
+                        deworming = gr.Radio(choices=["Yes", "No"], value="Yes", label="Deworming done")
+                    with gr.Row():
+                        family_dm = gr.Radio(choices=["Yes", "No"], value="No", label="Family history: Diabetes")
+                        family_htn = gr.Radio(choices=["Yes", "No"], value="No", label="Family history: Hypertension")
+                
+                with gr.Accordion("Menstrual & Obstetric History", open=False):
+                    with gr.Row():
+                        menstrual = gr.Radio(
+                            choices=["Regular", "Irregular", "Not applicable"],
+                            value="Regular",
+                            label="Menstrual Cycle"
+                        )
+                        pads_per_day = gr.Dropdown(
+                            choices=["1-2", "2-3", "3-4", "4-5", "More than 5", "Not applicable"],
+                            value="2-3",
+                            label="Pads per day (first 2-3 days)"
+                        )
+                    gr.Markdown("**Obstetric History (enter 0 if not applicable):**")
+                    with gr.Row():
+                        gravida = gr.Slider(minimum=0, maximum=10, value=0, step=1, label="Gravida (pregnancies)")
+                        para = gr.Slider(minimum=0, maximum=10, value=0, step=1, label="Para (deliveries)")
+                        live_births = gr.Slider(minimum=0, maximum=10, value=0, step=1, label="Live Births")
+                    with gr.Row():
+                        abortions = gr.Slider(minimum=0, maximum=5, value=0, step=1, label="Abortions")
+                        stillbirths = gr.Slider(minimum=0, maximum=5, value=0, step=1, label="Stillbirths")
+        
+        gr.Markdown("---")
+        
+        with gr.Row():
+            with gr.Column():
+                actual_hb_input = gr.Number(
+                    label="Already have a blood test result? Enter actual Hb (g/dL) — Optional",
+                    value=0,
+                    minimum=0,
+                    maximum=25
+                )
+            with gr.Column():
+                submit_btn = gr.Button("🔬 Estimate Hemoglobin", variant="primary", size="lg")
+        
+        gr.Markdown("---")
+        
+        # Results
+        with gr.Row():
+            result_output = gr.Markdown(label="Result")
+        with gr.Row():
+            session_output = gr.Textbox(label="Session ID", interactive=False)
+            hb_output = gr.Textbox(label="Estimated Hb", interactive=False)
+        
+        # Privacy notice
+        gr.Markdown("""
+        ---
+        ### 🔒 Privacy Notice
+        
+        Your data is collected **anonymously** to improve this screening tool. We do NOT collect:
+        - Name, address, phone number, or email
+        - Raw images (only extracted color features are stored)
+        - Any personally identifiable information
+        
+        By using this tool, you consent to anonymous data collection for research purposes.
+        
+        ---
+        *Developed by JIPMER Anemia Screening Study | For research use only*
+        """)
+        
+        # Connect button
+        submit_btn.click(
+            fn=predict_hb,
+            inputs=[
+                image_input,
+                age, income_range, family_size, diet,
+                food_grains, food_leafy, food_cereals, food_pulses,
+                food_fish, food_eggs, food_meat, food_jaggery, food_nuts,
+                fruit_freq, veg_freq, tea_coffee,
+                comorb_dm, comorb_htn, comorb_thyroid,
+                prev_anemia, family_anemia, hookworm, bleeding_disorder, deworming,
+                menstrual, pads_per_day,
+                gravida, para, live_births, abortions, stillbirths,
+                family_dm, family_htn,
+                actual_hb_input
+            ],
+            outputs=[result_output, session_output, hb_output]
+        )
+    
+    return app
+
+
+# ============================================================
+# MAIN
+# ============================================================
+if __name__ == "__main__":
+    print("Loading models...")
+    load_models()
+    
+    print("Starting app...")
+    app = create_interface()
+    app.launch()

best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:855d25f15fdb3c145219cdb418c6686aa74a43aa85b493f2d1b5c98bb909bc17
+size 6779380

model.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09c7c101f7a5788273dcde5063d5eddda26c80e76142bd5b3935c824d7165634
+size 1351178

requirements.txt

@@ -0,0 +1,7 @@
+gradio>=4.0.0
+numpy>=1.24.0
+pandas>=2.0.0
+opencv-python-headless>=4.8.0
+ultralytics>=8.0.0
+scikit-learn>=1.3.0
+xgboost>=2.0.0