ayushirathour
/

chest-xray-pneumonia-detection

@@ -1,248 +0,0 @@
-# PneumoDetect AI - Externally Validated Pneumonia Detection System
-**🏥 96.4% sensitivity • 86% accuracy • Externally validated on 485 independent chest X-rays**
----
-[![🚀 Try Live Demo](https://img.shields.io/badge/🚀-Try%20Live%20Demo-brightgreen?style=for-the-badge&logo=streamlit)](https://pneumodetectai.streamlit.app/) [![📁 GitHub](https://img.shields.io/badge/📁-Full%20Project-blue?style=for-the-badge&logo=github)](https://github.com/ayushirathour/chest-xray-pneumonia-detection) [![🤗 Model](https://img.shields.io/badge/🤗-Hugging%20Face-yellow?style=for-the-badge)](https://huggingface.co/ayushirathour/chest-xray-pneumonia-detection) [![License: MIT](https://img.shields.io/badge/License-MIT-orange.svg?style=for-the-badge)](https://opensource.org/licenses/MIT)
-## ✨ What This Does
-Transform chest X-ray images into pneumonia screening results in seconds. This AI model serves as a **research-validated screening support tool** to assist in pneumonia detection tasks and support educational applications.
-**Suitable for:** Research, education, screening support evaluation, and medical AI development projects.
-## 🚀 Get Started in 30 Seconds
-### Installation
-```bash
-pip install tensorflow huggingface-hub pillow numpy requests
-```
-### Complete Working Example
-```python
-from huggingface_hub import hf_hub_download
-import tensorflow as tf
-import numpy as np
-from PIL import Image
-# Download and load model
-model_path = hf_hub_download(
-    repo_id="ayushirathour/chest-xray-pneumonia-detection",
-    filename="best_chest_xray_model.h5"
-)
-model = tf.keras.models.load_model(model_path)
-def predict_pneumonia(image_path):
-    """
-    Predict pneumonia from chest X-ray image.
-    Args:
-        image_path (str): Path to chest X-ray image
-    Returns:
-        tuple: (diagnosis, confidence) - diagnosis string and confidence score
-    """
-    img = Image.open(image_path).convert('RGB')
-    img = img.resize((224, 224))
-    img_array = np.array(img) / 255.0
-    img_array = np.expand_dims(img_array, axis=0)
-    prediction = model.predict(img_array)[0][0]
-    result = "Pneumonia" if prediction > 0.5 else "Normal"
-    confidence = prediction if prediction > 0.5 else 1 - prediction
-    return result, confidence
-# Example usage
-result, confidence = predict_pneumonia("chest_xray.jpg")
-print(f"Diagnosis: {result} (Confidence: {confidence:.2%})")
-```
-> **Input Requirements:** 224×224 RGB chest X-ray images (PNG, JPEG, or DICOM formats supported)
-**🎯 That's it!** Upload any chest X-ray image and get instant screening results.
-## 📊 Performance at a Glance
-![External Validation Results](https://via.placeholder.com/600x300/2563eb/ffffff?text=External+Validation+Accuracy%3A+86%25)
-| Metric | Score | What This Means |
-|--------|-------|-----------------|
-| **Accuracy** | 86.0% | Correctly identifies 86 out of 100 cases |
-| **Sensitivity** | 96.4% | Catches 96 out of 100 pneumonia cases |
-| **Specificity** | 74.8% | Correctly identifies 75 out of 100 normal cases |
-| **F1-Score** | 87.7% | Balanced overall performance |
-**📈 Validation:** Tested on 485 independent chest X-rays (234 normal, 251 pneumonia) from external datasets to ensure real-world reliability.
-![Performance Comparison Chart](https://via.placeholder.com/600x250/10b981/ffffff?text=Internal+vs+External+Validation+Performance)
-## 🎯 Why Choose PneumoDetect AI?
-### ✅ **Externally Validated**
-Unlike many AI models, we tested this on completely independent data (485 samples) to prove robust generalization across different data sources.
-### ✅ **Screening Optimized**
-96.4% sensitivity means it rarely misses pneumonia cases - critical for research applications focused on high-sensitivity detection.
-### ✅ **Research Ready**
-Built on MobileNetV2 for fast inference, deployed with Streamlit demo, FastAPI REST API capabilities, and comprehensive documentation for research applications.
-### ✅ **Transparent Performance**
-Only 8.8% accuracy drop from training to external validation shows robust generalization across different datasets and imaging protocols.
-![Confusion Matrix](https://via.placeholder.com/400x400/dc2626/ffffff?text=Confusion+Matrix%0ADetailed+Classification+Results)
-<details>
-<summary><b>🔧 Advanced Integration Examples</b></summary>
-### Batch Processing Multiple X-rays
-```python
-from pathlib import Path
-def batch_screening(folder_path):
-    results = []
-    for img_file in Path(folder_path).glob("*.{jpg,jpeg,png}"):
-        result, confidence = predict_pneumonia(str(img_file))
-        results.append({
-            'image_id': img_file.stem,
-            'diagnosis': result,
-            'confidence': f"{confidence:.1%}",
-            'priority': 'HIGH' if 'Pneumonia' in result and confidence > 0.8 else 'NORMAL'
-        })
-    return results
-screening_results = batch_screening("research_xrays/")
-```
-### REST API Integration
-```python
-import requests
-def api_prediction(image_path, api_endpoint):
-    with open(image_path, 'rb') as f:
-        response = requests.post(f"{api_endpoint}/predict", files={"file": f})
-    return response.json()
-# For research deployments
-# result = api_prediction("xray.jpg", "https://your-research-api.com")
-```
-### Integration with DICOM Medical Images
-```python
-import pydicom
-from PIL import Image
-def predict_from_dicom(dicom_path):
-    # Load DICOM file
-    dcm = pydicom.dcmread(dicom_path)
-    img_array = dcm.pixel_array
-    # Convert to PIL Image and process
-    img = Image.fromarray(img_array).convert('RGB')
-    # Use standard prediction pipeline
-    return predict_pneumonia_from_pil(img)
-```
-</details>
-## 🏥 Research & Educational Applications
-### **Primary Use Cases**
-- **🔍 Pneumonia Detection Research:** High-sensitivity screening algorithm development
-- **⚡ Educational Applications:** AI-assisted learning for medical students and researchers
-- **👨‍⚕️ Algorithm Validation:** Benchmarking and comparison studies
-- **📚 Medical AI Development:** Foundation for advanced pneumonia detection systems
-### **Deployment Options**
-This model is designed to support radiology workflows through multiple deployment options:
-- **Web Interface:** Research demonstration via Streamlit demo
-- **REST API:** Programmatic integration for research applications
-- **Batch Processing:** High-volume analysis capabilities for research datasets
-- **DICOM Support:** Compatible with medical imaging research standards
-![Clinical Workflow Integration](https://via.placeholder.com/600x300/7c3aed/ffffff?text=Research+Workflow+Integration%0AValidated+Algorithm+for+Studies)
-## 🔬 Technical Architecture
-![Model Architecture](https://via.placeholder.com/600x250/059669/ffffff?text=MobileNetV2+Architecture%0AOptimized+for+Research+Deployment)
-<details>
-<summary><b>🏗️ Technical Details</b></summary>
-### Model Specifications
-- **Architecture:** MobileNetV2 with transfer learning
-- **Input:** 224×224 RGB chest X-ray images
-- **Output:** Binary classification (Normal/Pneumonia) with confidence scores
-- **Framework:** TensorFlow 2.x / Keras
-- **Optimization:** Balanced for accuracy and inference speed
-### Training & Validation Pipeline
-- **Training Data:** Curated subset of Kaggle Chest X-ray Dataset
-- **External Validation:** 485 independent samples from separate data sources
-- **Preprocessing:** Standardized resizing, normalization, and data augmentation
-- **Quality Control:** Systematic filtering for optimal training data quality
-### Performance Visualizations
-- **ROC Curves:** Area under curve analysis
-- **Precision-Recall Curves:** Optimized for medical screening research
-- **Class Distribution:** Balanced validation dataset
-- **Error Analysis:** Detailed false positive/negative breakdown
-</details>
-## 📈 Validation Results
-![ROC Curve](https://via.placeholder.com/400x300/1d4ed8/ffffff?text=ROC+Curve%0AAUC%3A+0.89)
-![Precision-Recall Curve](https://via.placeholder.com/400x300/db2777/ffffff?text=Precision-Recall+Curve%0AOptimized+for+Screening)
-The model demonstrates strong performance across multiple evaluation metrics with particular strength in sensitivity (pneumonia detection) which is critical for screening research applications.
-## ⚠️ Medical Considerations & Limitations
-### 🚨 **Important Usage Guidelines**
-**This is a research prototype and educational tool, NOT a diagnostic device.**
-- **Research Use Only:** Designed for academic research, education, and algorithm development
-- **Not for Clinical Diagnosis:** Not approved for clinical use or patient care decisions
-- **Professional Oversight Required:** Any research involving medical data should follow appropriate protocols
-- **False Positive Rate:** 25.2% - important consideration for research design and interpretation
-### **Technical Limitations**
-- **Training Scope:** Optimized for standard posteroanterior chest X-rays
-- **Population Variance:** Performance may differ across demographic groups and datasets
-- **Equipment Dependency:** Best results with standard medical imaging protocols
-- **Not FDA Approved:** For research and educational purposes only
-### **Recommended Research Workflow**
-1. **Algorithm Evaluation:** Use model for pneumonia detection research
-2. **Comparative Studies:** Benchmark against other detection algorithms
-3. **Educational Applications:** Demonstrate AI capabilities in medical imaging
-4. **Further Development:** Foundation for enhanced pneumonia detection systems
-## 📚 Research & Citation
-If you use PneumoDetect AI in your research or educational work:
-```bibtex
-@misc{rathour2025pneumodetect,
-  title={PneumoDetect AI: Externally Validated Pneumonia Detection System},
-  author={Rathour, Ayushi},
-  year={2025},
-  note={Externally validated deep learning system achieving 96.4% sensitivity on 485 independent samples},
-  url={https://huggingface.co/ayushirathour/chest-xray-pneumonia-detection}
-}
-```
-## 👩‍💻 Creator
-**Ayushi Rathour** - Biotechnology Graduate specializing in AI for Healthcare
-[![GitHub](https://img.shields.io/badge/GitHub-@ayushirathour-24292f?style=flat&logo=github)](https://github.com/ayushirathour) [![LinkedIn](https://img.shields.io/badge/LinkedIn-Connect-0a66c2?style=flat&logo=linkedin)](https://www.linkedin.com/in/ayushi-rathour/) [![Email](https://img.shields.io/badge/Email-Contact-ea4335?style=flat&logo=gmail)](mailto:ayushirathour1804@gmail.com)
----
-**🔬 Advancing Medical AI Through Rigorous Validation** | Built with ❤️ for the research community