colab link - https://colab.research.google.com/drive/1KIHaHtwpfgfa68eYx6qRypJkMsiy0ZGR?usp=sharing

This project looks at different methods to detect lung cancer from chest X-rays. I began with a simple CNN, then used transfer learning with frozen and fine tuned models, and finally tested a feature-based method with Isolation Forest. The f ine-tuned model gave the best accuracy, while the feature-based approach made the results easier to explain. The work also included preprocessing (grayscale, denoising, CLAHE), performance metrics, and confusion-matrix analysis. Overall, the study shows the trade-off between accuracy and explainability, and suggests that combining both can be useful for research and education.

Dataset The dataset followed a simple folder structure with two classes: • Normal– chest X-rays without signs of cancer.(from Kaggle) • Cancer– chest X-rays with abnormal regions.(from Kaggle)