Segmentation of Lung Cancer in CT Images using Modified U-Net Architecture

This paper introduces a deep learning method for lung cancer segmentation through Computed Tomography (CT) image analysis, with the goal of aiding radiologists in the early detection of lung cancer. For this purpose, a new dataset with manually annotated segmentation masks of cancer was created and used for testing purposes. The proposed segmentation network is based on the modified U-Net architecture tailored to the specific challenges of lung cancer detection. The network's design optimizes computational efficiency by reducing the number of channels in convolutional blocks and replacing pooling layers with convolutional layers in the decoder path, thus preserving spatial information. The modified U-Net achieved a Dice coefficient of 0.7998, demonstrating competitive performance compared to state-of-the-art models.