Lung cancer remains a significant global health challenge, emphasizing the critical need for early detection methods. This project proposes a deep-learning solution for classifying lung nodules found in chest CT scans, potentially revolutionizing early cancer diagnosis.
Our approach comprises three key stages:
Firstly, we employ various image processing techniques, including normalization, filtering, and thresholding, to segment the lung region from CT scans. This preprocessing step aims to isolate the lung area for subsequent analysis.
Next, we introduce a modified U-Net architecture, dubbed Conv-Unet, for segmenting individual lung nodules within the preprocessed lung region. Conv-Unet is designed to extract more intricate features compared to a standard U-Net, which could lead to superior nodule segmentation results.
Finally, Convolutional Neural Networks (CNNs) are trained to classify the segmented nodules based on the LIDC-IDRI dataset. The objective here is to categorize the nodules as either benign or malignant, thereby aiding in clinical decision-making.
This project relies on the following Python libraries:
- TensorFlow/Keras
- scikit-image
- SimpleITK (optional, for medical image processing)
Installation:
pip install tensorflow keras scikit-image [optional] SimpleITKNote: Specific versions of the libraries may be required depending on your project setup.
Use the ipynb notebook uploaded .