Region segmentation for lung cancer CT image using 3D U- Net model

Background Lung cancer detection through medical imaging is critical for early diagnosis and effective treatment planning. This study proposes a deep learning-based approach for automated lung segmentation in computed tomography (CT) scans, utilizing the Task06Lung dataset from the Medical Segmentation Decathlon (MSD) Challenge. Methods The dataset underwent preprocessing steps including resampling, normalization, and data augmentation to ensure consistency and diversity. Two U-Net-based architectures Simple U-Net and UNetM were implemented for segmentation. The models employed an encoder–decoder framework with skip connections to facilitate accurate feature extraction and reconstruction of lung regions. Training was performed using the Dice Loss function to address class imbalance, and a sliding window inference technique was applied to optimize memory usage during validation. Results Performance evaluation was conducted using segmentation metrics and confusion matrix analysis. The best model achieved a Dice score of 0. 67 at epoch 59. Additionally, the model demonstrated high classification performance, with a precision, recall, and F1-score of 0. 99, indicating strong accuracy in segmenting lung regions. Visualizations comparing predicted segmentations with ground truth masks supported the model’s effectiveness, while the confusion matrix highlighted areas requiring further improvement. Conclusion The proposed models showed strong performance in segmenting lung tissue in CT images. However, challenges remain in handling complex cancerous structures and fine anatomical boundaries. Future improvements may involve advanced data augmentation strategies and the integration of more sophisticated architectures, such as Attention U-Net, to enhance overall segmentation accuracy.