Lung Cancer Segmentation and Identification System Based on Improved U-Net Model

Deep convolutional neural networks have made continuous progress in the task of medical CT image segmentation and recognition. With the development of computer-aided diagnosis technology, researchers have proposed many automated medical image segmentation and recognition techniques in recent years. However, traditional lung cancer recognition methods face challenges such as limited data, sample imbalance, limited feature extraction of network models, and fixed receptive fields. To address these issues, we propose a lung cancer segmentation and recognition method based on an improved U-Net model. Our approach enhances data diversity through a custom 3D data generator. Additionally, we replace the feature extraction part of the U-Net model with the ResNet model, leveraging ResNet's deep feature learning capability and the U-Net model's excellent segmentation capability. We also improved a 3D-VGG16 model and combined it with the enhanced U-Net model to further boost performance. Finally, we employed model fine-tuning to optimize the model for specific lung cancer recognition tasks. Experimental results show that our proposed method achieves superior performance in lung cancer region segmentation and recognition. Most importantly, we successfully applied the enhanced model to a lung cancer diagnosis system and were able to provide strong support for real-life lung cancer region segmentation and recognition.