An approach to pulmonary nodule detection based on U-Net and 3D-CNN

This study explores the effectiveness of combining U-Net and 3D-CNN models for Pulmonary nodule detection and diagnosis in medical images. Data preprocessing involved converting files to DICOM format, standardizing pixel values, and enhancing image quality through normalization and zeroing techniques. Key preprocessing steps included scaling pixel values, saving images in PNG format, and employing morphological operations and binarization for effective image extraction and segmentation. Data augmentation using OpenCV enhanced model generalization. The U-Net model performed pixel-level segmentation with a symmetric encoder-decoder architecture, while the 3D-CNNmodel processed volumetric data for feature extraction and classification. Integrating these models resulted in an effective lung nodule detection and classification system.