https://2a27790ada4470e62b.gradio.live (To use it live)
Medical Imaging AI: Cancer Detection System(USING SYNTHETIC DATA)
Overview This is a deep learning-based medical imaging analysis system designed to detect potential cancerous abnormalities in medical scans. The system uses synthetic data for training and provides a user-friendly interface for analyzing uploaded medical images.
How to Use the System
1.Upload an Image: Drag and drop or click to upload a medical scan (X-ray, MRI, CT, etc.) 2.Adjust Sensitivity: Use the slider to set detection sensitivity (higher values catch more potential cases but may increase false positives) 3.Click Analyze: The system will process the image and generate a diagnostic report 4.Review Results: Examine the risk assessment, visualization, and recommendations
Key Features
- Advanced Image Processing:
Adaptive histogram equalization CLAHE (Contrast Limited Adaptive Histogram Equalization) Noise reduction Automatic format conversion (handles grayscale, RGB, RGBA)
- Deep Learning Model:
Based on MobileNetV3Small architecture for efficiency Transfer learning from ImageNet weights Custom classification head with regularization Trained with class weighting for imbalanced data
3.Comprehensive Reporting:
Risk score (0-1 probability) Risk level classification (Low/Moderate/High) Confidence estimation Personalized recommendations based on sensitivity
- Visualization:
Side-by-side comparison of original and processed images Enhanced view highlights potential abnormalities
5.Customization:
Adjustable sensitivity threshold Detailed model information available
How It Was Built
1.Synthetic Data Generation:
Created realistic synthetic medical images with abnormalities Simulated different tissue textures and lesion characteristics
2.Model Development:
Used transfer learning with MobileNetV3Small Added custom dense layers with regularization Implemented careful class weighting for the imbalanced dataset Used multiple metrics (AUC, precision, recall) for evaluation
3.Training Process:
Learning rate scheduling Early stopping based on validation AUC Learning rate reduction on plateau 30 epochs of training with batch size 32
4.Deployment
Built with Gradio for easy web interface Includes comprehensive error handling All values converted to native Python types for compatibility
Technical Specifications
Framework: TensorFlow 2.12.0 Base Model: MobileNetV3Small Input Size: 224×224 pixels Processing: CLAHE + Adaptive Histogram Equalization Output: Probability score (0-1) with risk assessment
⚠️ Disclaimer: This system is for educational and demonstration purposes only. It uses synthetic data and is not intended for clinical use.
Welcome to the Medical Imaging AI platform — a smart, deep learning-powered tool for identifying potential cancerous abnormalities in medical scans. Designed with an intuitive interface and a robust backend, it empowers users to visualize, analyze, and understand risk assessments — all in real time.
- ✅ Adaptive Histogram Equalization
- ✅ CLAHE (Contrast Limited Adaptive Histogram Equalization)
- ✅ Noise Reduction for Clarity
- ✅ Auto Support for RGB, RGBA, and Grayscale Formats
- ⚙️ MobileNetV3Small backbone (efficient and accurate)
- 🏋️ Transfer learning from ImageNet
- 🔁 Custom classification head with L2 regularization & dropout
- ⚖️ Class weighting for imbalanced data
- 📊 Risk Score: (0 - 1 probability)
- 🟢🟡🔴 Risk Level: Low, Moderate, or High
- ✅ Confidence Estimation
- 💡 Personalized recommendations
- 🎛️ Sensitivity Control Slider
- 🖼️ Original vs Processed Image Comparison
- 📈 Model performance metrics (AUC, precision, recall)
🧩 System Architecture
graph TD A[User Uploads Image] --> B[Image Preprocessing] B --> C[CLAHE + Histogram Equalization + Noise Reduction] C --> D[Model Inference (MobileNetV3Small)] D --> E[Risk Score + Classification] E --> F[Visualization & Recommendations]
📌 Uses synthetic training data only ❌ Not FDA/clinically approved 🚫 Not a substitute for professional diagnosis 🧪 Intended for education/demo use