""" OralGuard AI - Hugging Face Spaces Deployment Advanced AI Diagnostic System for Oral Cancer Detection """ import streamlit as st import torch from PIL import Image import pandas as pd import plotly.graph_objects as go import plotly.express as px from datetime import datetime import os import os from pathlib import Path from model_utils import GoogleDriveModelLoader, get_model_display_name, get_magnification_display_name, get_model_size_info from predict_utils import OralGuardPredictor # CRITICAL: Create cache directory cache_dir = Path("./models_cache") cache_dir.mkdir(parents=True, exist_ok=True) os.chmod(cache_dir, 0o755) # Ensure write permissions # Page configuration st.set_page_config( page_title="OralGuard AI", page_icon="đŸĻˇ", layout="wide", initial_sidebar_state="expanded" ) # Custom CSS st.markdown(""" """, unsafe_allow_html=True) # Initialize session state if 'model_loader' not in st.session_state: st.session_state.model_loader = None if 'predictor' not in st.session_state: st.session_state.predictor = None if 'prediction_history' not in st.session_state: st.session_state.prediction_history = [] def check_gdrive_configuration(): """Check if Google Drive file IDs are configured""" try: required_keys = [ 'EFFICIENTNET_100X_ID', 'EFFICIENTNET_400X_ID', 'VGG16_100X_ID', 'VGG16_400X_ID', 'VIT_100X_ID', 'VIT_400X_ID' ] missing_keys = [] for key in required_keys: if not st.secrets.get(key) and not os.getenv(key): missing_keys.append(key) return len(missing_keys) == 0, missing_keys except: return False, required_keys def initialize_models(): """Initialize model loader and predictor""" if st.session_state.model_loader is None: try: st.session_state.model_loader = GoogleDriveModelLoader() st.session_state.predictor = OralGuardPredictor(st.session_state.model_loader) return True except Exception as e: st.error(f"❌ Error initializing models: {str(e)}") return False return True def display_prediction_result(result, image=None): """Display prediction result in a beautiful format""" # Determine styling based on prediction if result['prediction'] == 'Normal': emoji = '✅' color = '#4CAF50' else: emoji = 'âš ī¸' color = '#F44336' # Create columns if image: col1, col2 = st.columns([1, 2]) with col1: st.image(image, use_container_width=True, caption="Input Image") else: col2 = st.container() with col2: st.markdown(f"""

{emoji} {result['prediction']}

Confidence: {result['confidence']*100:.2f}%

""", unsafe_allow_html=True) # Probability chart fig = go.Figure(data=[ go.Bar( x=['Normal', 'OSCC'], y=[result['probabilities']['Normal']*100, result['probabilities']['OSCC']*100], marker_color=['#4CAF50', '#F44336'], text=[f"{result['probabilities']['Normal']*100:.1f}%", f"{result['probabilities']['OSCC']*100:.1f}%"], textposition='auto', ) ]) fig.update_layout( title="Class Probabilities", yaxis_title="Probability (%)", height=300, showlegend=False ) st.plotly_chart(fig, use_container_width=True) # Detailed metrics st.markdown("### 📊 Detailed Results") col1, col2, col3 = st.columns(3) with col1: st.metric("Prediction", result['prediction']) st.metric("Confidence", f"{result['confidence']*100:.2f}%") with col2: st.metric("Normal Probability", f"{result['probabilities']['Normal']*100:.2f}%") st.metric("OSCC Probability", f"{result['probabilities']['OSCC']*100:.2f}%") with col3: st.metric("Inference Time", f"{result['inference_time']:.3f}s") st.metric("Model Used", result['model']) # Individual model results for ensemble if 'individual_results' in result: st.markdown("### 🤖 Individual Model Predictions") individual_data = [] for model_name, res in result['individual_results'].items(): individual_data.append({ 'Model': get_model_display_name(model_name), 'Prediction': res['prediction'], 'Confidence': f"{res['confidence']*100:.2f}%", 'Normal Prob': f"{res['probabilities']['Normal']*100:.2f}%", 'OSCC Prob': f"{res['probabilities']['OSCC']*100:.2f}%" }) st.dataframe(pd.DataFrame(individual_data), use_container_width=True) def single_image_prediction(): """Single image prediction interface""" st.markdown("## đŸ–ŧī¸ Single Image Analysis") col1, col2 = st.columns([2, 1]) with col1: uploaded_file = st.file_uploader( "Upload Histopathology Image", type=['png', 'jpg', 'jpeg'], help="Upload a histopathology image of oral tissue" ) with col2: st.markdown("### âš™ī¸ Configuration") model_type = st.selectbox( "Select Model", ['ensemble', 'efficientnet', 'vgg16', 'vit'], format_func=get_model_display_name, help="Choose which model to use for prediction" ) magnification = st.selectbox( "Magnification Level", ['100x', '400x'], format_func=get_magnification_display_name, help="Select the magnification level of your image" ) # Show expected download size if model_type != 'ensemble': sizes = get_model_size_info() st.info(f"đŸ“Ļ Download size: {sizes[model_type]}") else: st.info(f"đŸ“Ļ Download size: ~2.94 GB (all 3 models)") if uploaded_file is not None: # Display image FIRST image = Image.open(uploaded_file) st.image(image, caption="Uploaded Image", use_container_width=True) # Predict button if st.button("đŸ”Ŧ Analyze Image", key="single_predict", type="primary"): with st.spinner("🔄 Analyzing image..."): try: # Make prediction if model_type == 'ensemble': result = st.session_state.predictor.predict_ensemble(image, magnification) else: result = st.session_state.predictor.predict_single_model( image, model_type, magnification ) # Display result (pass image for display) display_prediction_result(result, image) # Add to history st.session_state.prediction_history.append({ 'timestamp': datetime.now().strftime('%Y-%m-%d %H:%M:%S'), 'prediction': result['prediction'], 'confidence': result['confidence'], 'model': result['model'] }) st.success("✅ Analysis complete!") except Exception as e: st.error(f"❌ Error during prediction: {str(e)}") import traceback st.code(traceback.format_exc()) def batch_image_prediction(): """Batch image prediction interface""" st.markdown("## 📁 Batch Image Analysis") col1, col2 = st.columns([2, 1]) with col1: uploaded_files = st.file_uploader( "Upload Multiple Histopathology Images", type=['png', 'jpg', 'jpeg'], accept_multiple_files=True, help="Upload multiple histopathology images for batch processing" ) with col2: st.markdown("### âš™ī¸ Configuration") model_type = st.selectbox( "Select Model", ['ensemble', 'efficientnet', 'vgg16', 'vit'], format_func=get_model_display_name, key="batch_model" ) magnification = st.selectbox( "Magnification Level", ['100x', '400x'], format_func=get_magnification_display_name, key="batch_mag" ) if uploaded_files: st.info(f"📊 {len(uploaded_files)} images uploaded") # Display thumbnails cols = st.columns(min(len(uploaded_files), 5)) for idx, file in enumerate(uploaded_files[:5]): with cols[idx]: img = Image.open(file) st.image(img, use_container_width=True) if len(uploaded_files) > 5: st.info(f"... and {len(uploaded_files) - 5} more images") if st.button("đŸ”Ŧ Analyze All Images", key="batch_predict", type="primary"): # Load images images = [Image.open(file) for file in uploaded_files] with st.spinner("🔄 Analyzing images..."): try: # Batch prediction results = st.session_state.predictor.predict_batch( images, model_type, magnification ) # Display summary st.success(f"✅ Analysis complete! Processed {len(results)} images") # Summary statistics successful = [r for r in results if r.get('status') == 'success'] normal_count = sum(1 for r in successful if r['prediction'] == 'Normal') oscc_count = sum(1 for r in successful if r['prediction'] == 'OSCC') avg_confidence = sum(r['confidence'] for r in successful) / len(successful) if successful else 0 col1, col2, col3, col4 = st.columns(4) with col1: st.metric("Total Images", len(results)) with col2: st.metric("Normal", normal_count) with col3: st.metric("OSCC", oscc_count) with col4: st.metric("Avg Confidence", f"{avg_confidence*100:.1f}%") # Results table st.markdown("### 📋 Detailed Results") results_data = [] for idx, result in enumerate(successful): results_data.append({ 'Image': f"Image_{idx+1}", 'Prediction': result['prediction'], 'Confidence': f"{result['confidence']*100:.2f}%", 'Normal Prob': f"{result['probabilities']['Normal']*100:.2f}%", 'OSCC Prob': f"{result['probabilities']['OSCC']*100:.2f}%", 'Time (s)': f"{result['inference_time']:.3f}" }) df = pd.DataFrame(results_data) st.dataframe(df, use_container_width=True) # Download results csv = df.to_csv(index=False) st.download_button( "đŸ“Ĩ Download Results (CSV)", csv, "oralguard_batch_results.csv", "text/csv", key='download-csv' ) # Visualization fig = px.pie( values=[normal_count, oscc_count], names=['Normal', 'OSCC'], title="Prediction Distribution", color_discrete_sequence=['#4CAF50', '#F44336'] ) st.plotly_chart(fig, use_container_width=True) except Exception as e: st.error(f"❌ Error during batch prediction: {str(e)}") def about_page(): """About page""" st.markdown(""" ## About OralGuard AI **OralGuard AI** is a production-ready AI diagnostic system designed to address Pakistan's oral cancer crisis. ### đŸŽ¯ Key Features - **90%+ Accuracy** on international benchmarks (Mendeley oral pathology dataset) - **On-Demand Model Loading** - Models download only when selected - **Ensemble Architecture** combining EfficientNet-B5, VGG16, and ViT-B/16 - **Dual Magnification** support (100× and 400×) - **Batch Processing** for multiple images ### đŸĨ Clinical Impact - Reduces diagnostic turnaround from **5-14 days to same-day** preliminary screening - Addresses Pakistan's **severe pathologist shortage** - Enables **specialist-level diagnostics** in remote areas - **Cost-effective**: Significantly lower per-case cost ### đŸ”Ŧ Technical Specifications - **Models**: EfficientNet-B5 (340 MB), VGG16-BN (1.6 GB), ViT-B/16 (1.0 GB) - **Image Size**: 224×224 pixels - **Classes**: Normal, OSCC (Oral Squamous Cell Carcinoma) - **Deployment**: Hugging Face Spaces with Google Drive integration ### 📊 Performance Metrics - **100× Magnification**: 84.91% accuracy (ViT-B/16) - **400× Magnification**: 81.43% accuracy (ViT-B/16) - **Ensemble**: Further improved performance ### âš–ī¸ Disclaimer This system is designed for **preliminary screening and clinical decision support**. All diagnoses should be confirmed by qualified pathologists. Not a replacement for professional medical judgment. --- **Version**: 1.0.0 | **Last Updated**: October 2024 """) def main(): """Main application""" # Header st.markdown("

đŸĻˇ OralGuard AI

", unsafe_allow_html=True) st.markdown("

Advanced AI Diagnostic System for Oral Cancer Detection

", unsafe_allow_html=True) # Check Google Drive configuration configured, missing = check_gdrive_configuration() if not configured: st.error("❌ Google Drive file IDs not configured!") st.markdown(""" ### 🔧 Configuration Required Please configure Google Drive file IDs in Hugging Face Secrets: 1. Go to your Space Settings → Repository secrets 2. Add the following secrets with your Google Drive file IDs: """) for key in missing: st.code(f"{key} = your_google_drive_file_id") st.info("📖 See README.md for detailed setup instructions") st.stop() # Initialize models if not initialize_models(): st.stop() # Sidebar with st.sidebar: st.markdown("### đŸŽ¯ Navigation") page = st.radio( "Select Mode", ["đŸ–ŧī¸ Single Image", "📁 Batch Analysis", "â„šī¸ About"], label_visibility="collapsed" ) st.markdown("---") st.markdown("### âš™ī¸ System Status") if torch.cuda.is_available(): st.success("✅ GPU Available") st.info(f"đŸ–Ĩī¸ {torch.cuda.get_device_name(0)}") else: st.warning("âš ī¸ Running on CPU") loaded_models = len(st.session_state.model_loader.loaded_models) st.metric("Loaded Models", loaded_models) if loaded_models > 0: if st.button("đŸ—‘ī¸ Clear Model Cache"): st.session_state.model_loader.clear_cache() st.rerun() st.markdown("---") # Model size info with st.expander("đŸ“Ļ Model Sizes"): sizes = get_model_size_info() st.write(f"**EfficientNet-B5:** {sizes['efficientnet']}") st.write(f"**VGG16-BN:** {sizes['vgg16']}") st.write(f"**ViT-B/16:** {sizes['vit']}") st.write(f"**Total (one magnification):** {sizes['total_per_magnification']}") # Main content if "đŸ–ŧī¸ Single Image" in page: single_image_prediction() elif "📁 Batch Analysis" in page: batch_image_prediction() elif "â„šī¸ About" in page: about_page() if __name__ == "__main__": main()