Lung Nodule Characterization in CT Scans Using Hybrid 3D Attention U-Net Segmentation and Transfer Learning-Based Classification Approach
DOI:
https://doi.org/10.65405/.v10i37.649الكلمات المفتاحية:
Lung cancer diagnosis, segmentation, U-Net, transfer learning, EfficientNet, CT image analysis, IQ-OTH/NCCD dataset.الملخص
Distinguishing early between benign and malignant pulmonary nodules is essential for
timely and effective lung cancer care. In this study, we introduce a two-stage deep learning system
designed to automatically analyze lung nodules in CT scans. The first stage utilizes a 3D U-Net
model enhanced with soft attention gates to accurately isolate nodule regions from the surrounding
lung tissue. This approach takes full advantage of the 3D spatial context while minimizing
interference from unrelated anatomical features. To handle the challenge of imbalanced data where
nodules occupy only a small fraction of the scan train the segmentation model utilizing a combined
loss function that blends Dice loss with binary cross-entropy. In the second stage, the segmented
nodules are fed into a classification pipeline built on EfficientNet-B7, adapted through transfer
learning. The model analyzes 2D patches cropped from multiple axial slices of the segmented
volume to categorize each nodule as normal, benign, or malignant. We evaluated our method on the
IQ-OTH/NCCD lung cancer dataset and achieved a test accuracy of 95.2%, with both sensitivity
and specificity surpassing 94% for detecting malignant cases. Our comparative evaluation shows a
clear advantage over traditional methods that rely on manually engineered features, for instance, the
Gabor-GLCM combined with SVM, which achieved only 71.8% accuracy in our experiments. By
combining attention-guided nodule segmentation with a transfer learning–based classifier, our
approach captures richer, more discriminative features, leading to consistently reliable diagnostic
outcomes. Beyond performance gains, this study offers a practical, end-to-end pipeline that is both
reproducible and well-suited for clinical utilize particularly in settings with limited resources. The
design also reflects a growing emphasis in medical AI on systems that are not only accurate but also
efficient and interpretable, supporting real-world clinical decision-making
التنزيلات
المراجع
[1]. Sun, K. X., Liang, X., Zhu, Q., Wu, H. L., Zhang, G. Y., Yao, Y. F., ... & Wei,
W. Q. (2025). Global patterns and trends in cancer-related premature death and their
impact on life expectancy across 185 countries: a population-based analysis. Military
Medical Research, 12(1), 1-17.
[2]. Szumiejko, A., Ptak, M., & Dołęga-Kozierowski, B. (2025). Computational
Methods in Breast Cancer Diagnostics and Surgery Planning: A Review. Archives of
Computational Methods in Engineering, 1-33.
[3]. Al-Zoghby, A. M., Ismail Ebada, A., Saleh, A. S., Abdelhay, M., & Awad, W. A.
(2025). A Comprehensive Review of Multimodal Deep Learning for Enhanced
Medical Diagnostics. Computers, Materials & Continua, 84(3).
[4]. Verma, S., & Kumar, M. (2025). A hybrid machine learning model for skin
disease classification using discrete wavelet transform and gray level co-occurrence
matrix (GLCM). Multimedia Tools and Applications, 84(14), 12835-12853.
[5] Attallah, O. (2025). Multi-Domain Feature Incorporation of Lightweight
Convolutional Neural Networks and Handcrafted Features for Lung and Colon
Cancer Diagnosis. Technologies, 13(5), 173.
[6]. Kazemi, A., & Esmaeili, M. (2025). Reservoir Surrogate Modeling Using U-Net
with Vision Transformer and Time Embedding. Processes, 13(4), 958.
[7] Tan, S. L., Selvachandran, G., Paramesran, R., & Ding, W. (2025). Lung cancer
detection systems applied to medical images: a state-of-the-art survey. Archives of
Computational Methods in Engineering, 32(1), 343-380.
[8]. Banerjee, T., Singh, D. P., Kour, P., Swain, D., Mahajan, S., Kadry, S., & Kim, J.
(2025). A novel unified Inception-U-Net hybrid gravitational optimization model
(UIGO) incorporating automated medical image segmentation and feature selection
for liver tumor detection. Scientific Reports, 15(1), 29908.
[9]. Ochoa-Ornelas, R., Gudiño-Ochoa, A., García-Rodríguez, J. A., & UribeToscano, S. (2025). A robust transfer learning approach with histopathological
images for lung and colon cancer detection using EfficientNetB3. Healthcare
Analytics, 7, 100391.
[10]. Laughton, C. (2025). Optimized Transfer Learning Strategies for Accurate and
Robust Medical Image Analysis. Transactions on Computational and Scientific
Methods, 5(9).
[11]. Mahmoud, M. A., Wen, Y., Liufu, Y., Pan, X., Su, R., & Guan, Y. (2025). A
Lightweight Dual-Output Vision Transformer for Enhanced Lung Nodule
Classification Using CT Images. Technology in Cancer Research & Treatment, 24,
15330338251370439.
التنزيلات
منشور
إصدار
القسم
الرخصة
هذا العمل مرخص بموجب Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
