This study investigates the effectiveness of U-Net architecture integrated with various convolutional neural network (CNN) backbones for automated lung cancer detection and segmentation in chest CT images. A balanced dataset of 832 chest CT images (416 malignant and 416 non-malignant) was preprocessed with CLAHE and resized to 128x128 pixels. U-Net models were developed using three CNN backbones - ResNet50, VGG16, and Xception - to segment the lung regions. After segmentation, CNN-based classifiers and hybrid models combining CNN feature extraction with traditional machine learning classifiers (SVM, Random Forest, and Gradient Boosting) were evaluated using 5-fold cross-validation. Accuracy, precision, recall, F1-score, Dice coefficient, and receiver operating system area under the curve (ROC-AUC) were used as evaluation metrics. U-Net with ResNet50 achieved the best performance (Dice: 0.9495, Accuracy: 0.9735) for cancerous lung segmentation, while U-Net with VGG16 achieved the best performance (Dice: 0.9532, Accuracy: 0.9513) for non-cancerous segmentation. For classification, the CNN model with U-Net with Xception achieved 99.1% accuracy, 99.74% recall, and 99.42% F1-score. The hybrid CNN-SVM-Xception model achieved 96.7% accuracy and 97.88% F1-score. Compared with existing methods, our framework consistently outperformed the existing models. In conclusion, combining advanced CNN backbone with U-Net provides a powerful method for segmentation and classification of lung cancer from CT scans, supporting early diagnosis and clinical decision making.
