Neural Diagnostics Portal

Dataset & Source

The model was trained using the Chest X-Ray Images (Pneumonia) dataset from Kaggle, consisting of over 5,800 clinical images. The data is categorized into 'Normal' and 'Pneumonia' (Viral and Bacterial).

Neural Architecture

I utilized MobileNetV2 as the base architecture for its lightweight footprint and high efficiency. A custom "Smart Pipeline" was integrated directly into the model layers:

• Preprocessing Layer Automatic 224x224 resizing & 1/255 rescaling.
• Feature Extraction Global Average Pooling to reduce spatial dimensions.
• Regularization 30% Dropout layer to prevent overfitting.
• Inference Sigmoid activation for binary probability output.

Input Validation Logic

To ensure medical integrity, I implemented a Saturation Gate. Since chest X-rays are grayscale, the system analyzes the HSV saturation mean of every upload. If the mean exceeds a specific threshold (colorful images), the system rejects the input to prevent "False Positive" diagnoses on non-medical images.

Cloud Stack

The backend is hosted on Hugging Face Spaces using Gradio and TensorFlow-CPU. The frontend is a custom Tailwind CSS portal deployed on Netlify, communicating with the AI model via an optimized iframe bridge.