[论文解读] Advanced U-Net Architectures with CNN Backbones for Automated Lung Cancer Detection and Segmentation in Chest CT Images

Alireza Golkarieh, Kiana Kiashemshaki+2 more|ArXiv.org|Jul 14, 2025

Radiomics and Machine Learning in Medical Imaging被引用 3

一句话总结

论文评估了在胸部CT图像上使用 U-Net 分割并以 VGG16、ResNet50、Xception 作为骨干网络的性能，并分析了基于 CNN 和混合 CNN+ML 的肺癌检测分类器，在使用 U-Net 分割的情况下实现了高达 99.1% 的分类准确率。

ABSTRACT

This study investigates the effectiveness of U-Net architectures integrated with various convolutional neural network (CNN) backbones for automated lung cancer detection and segmentation in chest CT images, addressing the critical need for accurate diagnostic tools in clinical settings. A balanced dataset of 832 chest CT images (416 cancerous and 416 non-cancerous) was preprocessed using Contrast Limited Adaptive Histogram Equalization (CLAHE) and resized to 128x128 pixels. U-Net models were developed with three CNN backbones: ResNet50, VGG16, and Xception, to segment lung regions. After segmentation, CNN-based classifiers and hybrid models combining CNN feature extraction with traditional machine learning classifiers (Support Vector Machine, Random Forest, and Gradient Boosting) were evaluated using 5-fold cross-validation. Metrics included accuracy, precision, recall, F1-score, Dice coefficient, and ROC-AUC. U-Net with ResNet50 achieved the best performance for cancerous lungs (Dice: 0.9495, Accuracy: 0.9735), while U-Net with VGG16 performed best for non-cancerous segmentation (Dice: 0.9532, Accuracy: 0.9513). For classification, the CNN model using U-Net with Xception achieved 99.1 percent accuracy, 99.74 percent recall, and 99.42 percent F1-score. The hybrid CNN-SVM-Xception model achieved 96.7 percent accuracy and 97.88 percent F1-score. Compared to prior methods, our framework consistently outperformed existing models. In conclusion, combining U-Net with advanced CNN backbones provides a powerful method for both segmentation and classification of lung cancer in CT scans, supporting early diagnosis and clinical decision-making.

研究动机与目标

• Motivate improved accuracy in lung cancer detection and segmentation from chest CT scans.
• Evaluate U-Net segmentation performance with three pretrained CNN backbones (VGG16, ResNet50, Xception).
• Assess classification performance using CNNs trained on U-Net segmentations and hybrid CNN+ML classifiers (SVM, RF, GB).
• Apply robust cross-validation and multiple metrics to compare segmentation and classification outcomes.
• Position the proposed framework against existing studies to highlight gains in segmentation and detection accuracy.

提出的方法

• Preprocess CT images with CLAHE, resize to 128×128, and normalize pixel values.
• Segment lungs using U-Net with VGG16, ResNet50, or Xception backbones (encoder from pretrained CNNs).
• Train independent CNN classifiers on U-Net segmentations for cancer vs non-cancer detection.
• Develop hybrid models by feeding CNN-extracted features into SVM, Random Forest, and Gradient Boosting classifiers.
• Evaluate with 5-fold cross-validation using Dice, IoU, accuracy, precision, recall, F1, ROC-AUC; report means and SDs.

实验结果

研究问题

• RQ1 Do U-Net segmentations with different pretrained backbones (VGG16, ResNet50, Xception) yield superior lung segmentation for cancer vs non-cancer CT images?
• RQ2 What is the classification performance of CNN models trained on U-Net segmentations, and how do hybrid CNN+ML approaches compare?
• RQ3 Which backbone and classifier combination offers the best balance of precision, recall, and ROC-AUC for lung cancer detection on chest CT data?

主要发现

• U-Net with ResNet50 achieved cancerous segmentation Dice of 0.9495 and accuracy 0.9735 (±0.1015).
• U-Net with VGG16 best segmented non-cancerous lungs with Dice 0.9532 and accuracy 0.9513 (±0.1006).
• U-Net with Xception provided consistent results (cancerous Dice 0.9383, non-cancerous Dice 0.9521).
• CNN model trained on U-Net segmentations with Xception backbone achieved 99.10% accuracy, 99.74% recall, and 99.42% F1-score (ROC-AUC 0.9911).
• Hybrid CNN+ML models achieved competitive performance, with CNN-SVM-Xception reaching strong accuracy and F1, and CNN-GB-VGG16 attaining high ROC-AUC.
• Pre-trained backbones showed strong classification with Xception providing balanced results (accuracy 0.9580, recall 0.9950, F1 0.9739, ROC-AUC 0.9193).