Overview
Lung cancer remains one of the deadliest diseases globally, and early detection is critical in improving survival rates. In this project, I worked on building a convolutional neural network (CNN) to classify CT-scan images of lungs as either normal or cancerous. Implementing such techniques in the healthcare industry has immense potential to assist doctors in saving lives by streamlining diagnosis.
The datasets for this project are publicly available on kaggle. Below are two example images from the dataset. The image on the left shows a normal set of lungs, and the one on the right cancerous lungs.
The primary objective of this project was to classify lung CT-scans into two categories: normal and cancerous lungs. Additionally, I also experimented with a more challenging dataset containing four classes, which introduced additional variations and complexity.
I implemented a CNN model with the following architecture:
- Three convolutional layers: Responsible for extracting critical features like patterns and edges from the images.
- One max pooling layer: Used to reduce dimensionality while preserving important features.
- Two dense layers: These layers perform the final classification based on the features extracted.
For the two-class dataset (normal vs. cancerous), the model achieved a perfect 100% accuracy during validation. To make this result more practical, I developed a GUI where users can input a CT-scan image, and the trained model predicts whether the lungs are normal or cancerous.
The four-class dataset was more challenging due to the increased complexity. The model achieved an accuracy of 84% during validation. This opens avenues for future research and optimization.
GitHub
The full code is available on my GitHub.
Contact
For more information or inquiries about this project, please contact
me at:
westling01@gmail.com.