stevia-s/multiclass-lungdisease-detection-using-xai

This project is a deep learning system that analyzes CT scan images of lungs and classifies them into one of three categories: COVID, pneumonia, or normal. The core problem it addresses is that manually reading CT scans is time consuming, depends on having a specialist available, and can be error prone in subtle or complex cases. An AI system that flags a likely diagnosis is meant to help radiologists work faster and catch patterns that might otherwise be missed. What makes this project distinctive is its focus on explainability. Standard deep learning models are often described as black boxes because they produce a result without showing how they reached it, which is a real problem in medical settings where clinicians need to understand and trust a decision before acting on it. This project uses a technique called Grad-CAM, short for Gradient-weighted Class Activation Mapping, to generate a heatmap laid over the original CT scan that visually highlights the regions of the lung image that most influenced the model's classification. The model architecture combines features extracted by two separate neural networks, ResNet50 and VGG16, then fuses those features together before making the final classification. The pipeline runs a CT scan through image preprocessing, feature extraction, a classification layer, and then the Grad-CAM explainability step, producing both a predicted disease and a visual explanation for it. The system is built in Python using TensorFlow and Keras, with a Streamlit based web interface for running predictions. Reported accuracy ranges from about 90% to 97% depending on the dataset and training setup, according to the README. Listed future plans include a web based upload system, real time hospital integration, and mobile deployment, none of which are built yet. The project states plainly that it is intended for academic and research purposes only.