SavedAdd your journal

Journals / Niğde Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi / 2026 / Cilt: 17

Training strategies and hyperparameter analysis of U-Net based models for lung segmentation in chest X-ray images

Göğüs X-ışını görüntülerinde akciğer segmentasyonu için U-Net tabanlı modellerin eğitim stratejileri ve hiperparametre analizi

Nihan Kazak Çerçevik(Bilecik Şeyh Edebali University)0000-0001-5339-0055
Sündüs Sena Uysal(BİLECİK ŞEYH EDEBALİ ÜNİVERSİTESİ, MÜHENDİSLİK FAKÜLTESİ)0009-0009-6057-8035
Pages
—
Publication Date
2026-07-09
🏛️Bibliographic metadata and full text access for this article are provided via DergiPark.
DergiPark Record ↗
✨

AI Summary & Key Takeaways

AI Insight
⚡ TL;DR(Too Long; Didn't Read)

The results showed that the Adam optimization algorithm achieved the most stable convergence with a learning rate of 1e-4.

📌 3 Key Takeaways

🎯
Objective & Hypothesis:Accurate identification of lung fields in chest X-ray images is critical for disease detection and reliable clinical analysis.
📊
Key Findings & Data:This study systematically investigates the effects of hyperparameters and architectural preferences on lung segmentation performance in U-Net-based deep learning models.
💡
Conclusion & Impact:The results showed that the Adam optimization algorithm achieved the most stable convergence with a learning rate of 1e-4.
🔬Design:Modeling & Simulation
📍Scope:Comprehensive Sample
⚙️Variables:Akciğer segmentasyonu · U-Net · Dikkat mekanizması

Abstract

Accurate identification of lung fields in chest X-ray images is critical for disease detection and reliable clinical analysis. This study systematically investigates the effects of hyperparameters and architectural preferences on lung segmentation performance in U-Net-based deep learning models. In the experimental process, optimization algorithms, learning rates, stack sizes, backbone structures, and loss functions were evaluated comparatively. The results showed that the Adam optimization algorithm achieved the most stable convergence with a learning rate of 1e-4. In architectural analyses, the integration of the ResNet34 backbone with the attention mechanism improved segmentation success in low-contrast border regions and complex anatomical structures. In loss function analyses, the equally balanced hybrid use of Binary Cross Entropy (BCE) and Dice Loss demonstrated superior performance compared to individual functions in preserving both pixel-based details and global structural integrity. With the best configuration (50 epochs, 8 batch sizes), 0.9736 Dice, 0.9485 IoU, and 0.9845 pixel accuracy were obtained. The model's reliability was validated with 5-fold cross-validation, demonstrating that the hybrid structure offers high generalization capability. These findings highlight the importance of holistic hyperparameter optimization and hybrid loss strategies in medical image segmentation for clinical accuracy.

Özet

Göğüs X-ışını görüntülerinde akciğer alanlarının doğru belirlenmesi, hastalıkların tespiti ve güvenilir klinik analizler için kritik öneme sahiptir. Bu çalışmada, U-Net tabanlı derin öğrenme modellerinde hiperparametrelerin ve mimari tercihlerinin akciğer segmentasyonu performansına etkileri sistematik olarak incelenmiştir. Deneysel süreçte optimizasyon algoritmaları, öğrenme katsayıları, yığın boyutları, omurga yapıları ve kayıp fonksiyonları karşılaştırmalı biçimde değerlendirilmiştir. Sonuçlar, Adam optimizasyon algoritmasının 1e-4 öğrenme katsayısı ile en kararlı yakınsamayı sağladığını göstermiştir. Mimari analizlerde, ResNet34 omurgasının dikkat mekanizması ile entegrasyonu düşük kontrastlı sınır bölgelerinde ve karmaşık anatomik yapılarda segmentasyon başarısını artırmıştır. Kayıp fonksiyonu incelemelerinde, İkili Çapraz Entropi (Binary Cross Entropy – BCE) ve Dice Loss’un eşit oranlı hibrit kullanımı hem piksel bazlı detayların hem de global yapısal bütünlüğün korunmasında tekil fonksiyonlara göre üstün performans sergilemiştir. En iyi konfigürasyon (50 epok, 8 yığın boyutu) ile 0.9736 Dice, 0.9485 IoU ve 0.9845 piksel doğruluğu elde edilmiştir. Modelin güvenilirliği 5-katlı çapraz doğrulama ile doğrulanmış ve hibrit yapının yüksek genelleme yeteneği sunduğu kanıtlanmıştır. Bu bulgular, medikal görüntü segmentasyonunda bütüncül hiperparametre optimizasyonunun ve hibrit kayıp stratejilerinin klinik doğruluk açısından önemini vurgulamaktadır.

🔖

Cite This Article

Çerçevik, N. K., Uysal, S. S. (2026). Training strategies and hyperparameter analysis of U-Net based models for lung segmentation in chest X-ray images. Niğde Ömer Halisdemir Üniversitesi Mühendislik Bilimleri Dergisi, Cilt: 17. https://doi.org/10.28948/ngumuh.1900094

📄 Full Text PDF Viewer

📄

PDF görüntüleyici hazırlanıyor...

References (26)

  1. S. Candemir, S. Jaeger, K. Palaniappan, J. P. Musco, R. K. Singh, Z. Xue, A. Karargyris, S. Antani, g. Thoma ve C.J. McDonald, “Lung segmentation in chest radiographs using anatomical atlases with nonrigid registration,” IEEE Transactions on Medical Imaging, 33(2), 577-590, 2013. https://doi.org/10.1109/TMI.2013.2290491
  2. O. Ronneberger, P. Fischer ve T. Brox, “U-Net: Convolutional networks for biomedical image segmentation,” Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015, 9351, sayfa 234-241, 2015. MICCAI 2015. Lecture Notes in Computer Science, vol. 9351. Springer, Cham. https://doi.org/10.1007/978-3-319-24574-4_28.
  3. Z. Zhou, M.M. Rahman Siddiquee, N. Tajbakhsh ve J. Liang, “UNet++: A nested U-Net architecture for medical image segmentation,” Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, 11045, sayfa 3-11, 2018. DLMIA ML-CDS 2018 2018. Lecture Notes in Computer Science, vol 11045. Springer, Cham. https://doi.org/10.1007/978-3-030-00889-5_1
  4. O. Oktay, J. Schlemper, L.L. Folgoc, M. Lee, M. Heinrich, K. Misawa, K. Mori, S. McDonagh, N. Y. Hammerla, B. Kainz, B. Glocker ve D Rueckert, “Attention U-Net: learning where to look for the pancreas,” arXiv preprint arXiv:1804.03999, 2018. https://doi.org/10.48550/arXiv.1804.03999
  5. A. Das, “Adaptive UNet-based Lung Segmentation and Ensemble Learning with CNN-based deep features for automated COVID-19 diagnosis,” Multimed Tools Appl, 81, 5407–5441, 2022. https://doi.org/10.1007/s11042-021-11787-y
  6. J. Hofmanninger, F. Prayer, J. Pan, S. Röhrich, H. Prosch ve G. Langs, “Automatic lung segmentation in routine imaging is primarily a data diversity problem, not a methodology problem,” European Radiology Experimental, 4(1), 1-13, 2020. https://doi.org/10.1186/s41747-020-00173-2
  7. S. Gite, A. Mishra ve K. Kotecha, “Enhanced lung image segmentation using deep learning,” Neural Comput & Applic, 35, 22839–22853, 2023. https://doi.org/10.1007/s00521-021-06719-8
  8. S. Din, M. Shoaib ve E. Serpedin, “CXR-seg: A novel deep learning network for lung segmentation from chest X-ray images,” Bioengineering, 12(2), 167, 2025. https://doi.org/10.3390/bioengineering12020167
  9. F. Turk ve M. Kılıçaslan, “Lung image segmentation with improved U-Net, V-Net and Seg-Net techniques,” PeerJ Computer Science, 11, e2700, 2025. https://doi.org/10.7717/peerj-cs.2700
  10. M. Murugappan, A.K. Bourisly, N.B. Prakash, M. G. Smithra ve U. R. Acharya, “Automated semantic lung segmentation in chest CT images using deep neural network,” Neural Comput & Applic, 35, 15343–15364, 2023. https://doi.org/10.1007/s00521-023-08407-1
  11. C.H. Sudre, W. Li, T. Vercauteren, S. Ourselin ve M.J. Cardoso, “Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations,” Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, 10553, 240-248, 2017. https://doi.org/10.1007/978-3-319-67558-9_28
  12. S.S.M. Salehi, D. Erdogmus ve A. Gholipour, “Tversky loss function for ımage segmentation using 3D fully convolutional deep networks,” Machine Learning in Medical Imaging, 10541, 379-387, 2017. https://doi.org/10.1007/978-3-319-67389-9_44
  13. S. Jadon, “A survey of loss functions for semantic segmentation,” 2020 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (CIBCB), sayfa 1-7, Via del Mar, Chile,2020. https://doi.org/10.1109/CIBCB48159.2020.9277638
  14. L.N. Smith, “Cyclical learning rates for training neural networks,” 2017 IEEE Winter Conference on Applications of Computer Vision (WACV), sayfa 464-472, Santa Rosa, CA, USA, 2017. https://doi.org/10.1109/WACV.2017.58
  15. N.S. Keskar, D. Mudigere, J. Nocedal, M. Smelyanskiy ve P.T.P. Tang, “On large-batch training for deep learning: Generalization gap and sharp minima,” arXiv preprint arXiv:1609.04836, 2016. https://doi.org/10.48550/arXiv.1609.04836
  16. G. Litjens, T. Kooi, B.E. Bejnordi, A.A.A. Setio, F. Ciompi, M. Ghafoorian, J.A.W.M. van der Laak, B. Van Ginneken ve C.I. Sánchez, “A survey on deep learning in medical image analysis,” Medical Image Analysis, 42, 60-88, 2017. https://doi.org/10.1016/j.media.2017.07.005
  17. M.H. Hesamian, W. Jia, X. He ve P. Kennedy, “Deep learning techniques for medical image segmentation: achievements and challenges,” Journal of Digital Imaging, 32(4), 582-596, 2019. https://doi.org/10.1007/s10278-019-00227-x
  18. Y. Zhang, “Lung segmentation with nasnet-large-decoder net,” arXiv preprint, arXiv:2303.10315, 2023. https://doi.org/10.48550/arXiv.2303.10315
  19. T. Karmakar, “Chest X-ray lungs segmentation,” Kaggle, 2020. Available: https://www.kaggle.com/datasets/iamtapendu/chest-x-ray-lungs-segmentation, [Accessed Nov. 9, 2025].
  20. S. Jaeger, S. Candemir, S. Antani, Y.X.J. Wáng, P.X. Lu ve G. Thoma, “Two public chest X-ray datasets for computer-aided screening of pulmonary diseases,” Quantitative Imaging in Medicine and Surgery, 4(6), 475-477, 2014. https://doi.org/10.3978/j.issn.2223-4292.2014.11.20
  21. O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg ve L. Fei-Fei, “ImageNet large scale visual recognition challenge,” International Journal of Computer Vision, 115(3), 211–252, 2015. https://doi.org/10.1007/s11263-015-0816-y
  22. M. Tan ve Q. V. Le, “EfficientNet: Rethinking model scaling for convolutional neural networks,” Proceedings of the 36th International Conference on Machine Learning (ICML), 97, sayfa 6105-6114, 2019. https://doi.org/10.48550/arXiv.1905.11946
  23. K. Simonyan ve A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint, arXiv:1409.1556, 2015. https://doi.org/10.48550/arXiv.1409.1556
  24. K. He, X. Zhang, S. Ren ve J. Sun, “Deep residual learning for image recognition,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), sayfa 770–778, 2016. https://doi.org/10.1109/CVPR.2016.90
  25. J.C. Souza, J.O.B. Diniz, J.L. Ferreira, G.L.F. Da Silva, A.C. Silva ve A.C. De Paiva, “An automatic method for lung segmentation and reconstruction in chest X-ray using deep neural networks,” Computer Methods and Programs in Biomedicine, 177, 285-296, 2019. https://doi.org/10.1016/j.cmpb.2019.06.005
  26. S.A. Salkade ve S.V. Rathi, “An adaptive convolution neural network model for tuberculosis detection and diagnosis using semantic segmentation,” Polish Journal of Radiology, 90, e124-e135, 2025. https://doi.org/10.5114/pjr/200628