INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS

Umut Kaan Kavaklı; Göksu Turaç; Sema Candemir

TR EN

LEZYON BÖLÜTLENMESİNDE DOĞAL SINIF DENGESİZLİĞİ: KAYIP FONKSİYONLARININ ETKİSİ

Abstract

Tıbbi görüntü segmentasyonu, anatomik veya patolojik açıdan önemli bölgelerin ayrıştırılmasını sağlayarak klinik analizlerde kritik bir rol oynamaktadır. Lezyon segmentasyonundaki temel zorluklardan biri, sağlıklı dokuların anormal bölgelerden fazla olması nedeniyle öğrenme algoritmalarını çoğunluk sınıfına yönlendiren sınıf dengesizliğidir. Lezyonların sınıflar arası ve sınıf içi değişkenliği segmentasyonu daha da zorlaştırmaktadır. Uygun kayıp fonksiyonları bu zorlukları azaltabilir. Bu çalışmada, COVID-19 BT görüntülerinde bu zorlukların etkisini azaltmaya yönelik farklı kayıp fonksiyonları U-Net tabanlı bir model kullanılarak değerlendirilmiştir. Lezyon heterojenitesi; entropi, lezyon alanı, gradyan büyüklüğü ve yoğunluk farkları kullanılarak değerlendirilmiştir. Sonuçlar, dengesizliğe duyarlı ve bölge tabanlı kayıp fonksiyonlarının heterojen lezyon dağılımlarında güçlü performans sağladığını göstermektedir. Bulgular, güvenilir segmentasyon için kayıp fonksiyonu seçiminde sınıf dengesizliği, lezyon heterojenitesi ve yapısal karmaşıklığın dikkate alınması gerektiğini ortaya koymaktadır.

Keywords

Supporting Institution

TUBITAK-BIDEB

Project Number

121C085

Ethical Statement

Bu çalışmada kullanılan tüm veriler açık kaynak veri setlerinden elde edilmiştir.

INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS

Abstract

Medical image segmentation plays a critical role in clinical analysis by delineating anatomically or pathologically relevant regions. A major challenge in lesion segmentation is class imbalance where healthy tissues outnumber abnormal regions, biasing learning algorithms toward the majority class and degrading lesion detection. Inter-class and intra-class variability of lesions further complicate the segmentation. Appropriate loss functions can mitigate these challenges. In this study, we evaluate widely used loss functions for COVID-19 lesion segmentation using publicly available CT datasets. A U-Net–based framework is evaluated under varying imbalance ratios and lesion characteristics. The lesion heterogeneity is assessed using entropy, lesion area, gradient magnitude, and intensity differences. Results show that imbalance-aware and region-based losses provide robust performance under heterogeneous lesion distributions. The findings highlight that loss function selection should consider class imbalance, lesion heterogeneity, and structural complexity for reliable segmentation.

Keywords

Supporting Institution

TUBITAK-BIDEB

Project Number

121C085

Ethical Statement

All data used in this study were obtained from publicly available datasets.

References

Abraham, N., & Khan, N. M. (2019, April). A novel focal Tversky loss function with improved attention U-Net for lesion segmentation. In 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI) (pp. 683–687). IEEE. https://doi.org/10.1109/ISBI.2019.8759329
American College of Radiology. (2020). ACR recommendations for the use of chest radiography and computed tomography for suspected COVID-19 infection. Retrieved March 28, 2026, from American College of Radiology
An, P., Xu, S., Harmon, S. A., Turkbey, E. B., Sanford, T. H., Amalou, A., Kassin, M., Varble, N., Blain, M., Anderson, V., & Patella, F. (2020). CT images in COVID-19 dataset. Cancer Imaging Archive. https://doi.org/10.7937/TCIA.2020.GQRY-NC81
Azad, R., Heidary, M., Yilmaz, K., Hüttemann, M., Karimijafarbigloo, S., Wu, Y., Schmeink, A., & Merhof, D. (2023). Loss functions in the era of semantic segmentation: A survey and outlook. arXiv:2312.05391. https://arxiv.org/abs/2312.05391
Cohen, J. P., Morrison, P., & Dao, L. (2020). COVID-19 image data collection. arXiv:2003.11597. https://arxiv.org/abs/2003.11597
Fan, D.-P., Zhou, T., Ji, G.-P., Zhou, Y., Chen, G., Fu, H., Shen, J., & Shao, L. (2020). Inf-Net: Automatic COVID-19 lung infection segmentation from CT images. IEEE Transactions on Medical Imaging, 39, 2626–2637. https://doi.org/10.1109/TMI.2020.2996645
Géron, A. (2022). Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow. O’Reilly Media. https://www.oreilly.com/library/view/hands-on-machine-learning/9781098125967/
Hashemi, S. R., Salehi, S. S., Erdogmus, D., Prabhu, S. P., Warfield, S. K., & Gholipour, A. (2018). Asymmetric loss functions and deep densely-connected networks for highly-imbalanced medical image segmentation: Application to multiple sclerosis lesion detection. IEEE Access, 7, 1721–1735. https://doi.org/10.1109/ACCESS.2018.2886371

Jadon, S. (2020). A survey of loss functions for semantic segmentation. In 2020 IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology (pp. 1–7). IEEE. https://doi.org/10.1109/CIBCB48159.2020.9277638
Johnson, J. M., & Khoshgoftaar, T. M. (2019). Survey on deep learning with class imbalance. Journal of Big Data, 6, 1–54. https://doi.org/10.1186/s40537-019-0192-5
Li, L., Qin, L., Xu, Z., Yin, Y., Wang, X., Kong, B., Bai, J., Lu, Y., Fang, Z., Song, Q., & Cao, K. (2020). Using artificial intelligence to detect COVID-19 and community-acquired pneumonia based on pulmonary CT: Evaluation of the diagnostic accuracy. Radiology, 296, E65–E71. https://doi.org/10.1148/radiol.2020200905
Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollár, P. (2017, October). Focal loss for dense object detection. In 2017 IEEE International Conference on Computer Vision (ICCV) (pp. 2980–2988). IEEE. https://doi.org/10.1109/ICCV.2017.324
Ma, J., Chen, J., Ng, M., Huang, R., Li, Y., Li, C., Yang, X., & Martel, A. (2021). Loss odyssey in medical image segmentation. Medical Image Analysis, 71, 102035. https://doi.org/10.1016/j.media.2021.102035
Nguyen, Q., & Thai, H. (2023). Crack segmentation of imbalanced data: The role of loss functions. Engineering Structures, 297, 116988. https://doi.org/10.1016/j.engstruct.2023.116988
Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation.
In Medical Image Computing and Computer-Assisted Intervention (MICCAI) (Vol. 9351, pp. 234–241). Springer. https://doi.org/10.1007/978-3-319-24574-4_28
Salehi, S. S., Erdogmus, D., & Gholipour, A. (2017). Tversky loss function for image segmentation using 3D fully convolutional deep networks. In the International Workshop on Machine Learning in Medical Imaging (MLMI). Springer. https://doi.org/10.1007/978-3-319-67389-9_44
Shan, F., Gao, Y., Wang, J., Shi, W., Shi, N., Han, M., Xue, Z., Shen, D., & Shi, Y. (2021). Abnormal lung quantification in chest CT images of COVID-19 patients with deep learning and its application to severity prediction. Medical Physics, 48(4), 1633–1645. https://doi.org/10.1002/mp.14609
Stockman, G., & Shapiro, L. (2001). Computer Vision. Prentice Hall, pp.305-351.
Sun, D., Li, X., Guo, D., Wu, L., Chen, T., Fang, Z., Chen, L., Zeng, W., & Yang, R. (2020). CT quantitative analysis and its relationship with clinical features for assessing the severity of patients with COVID-19. Korean Journal of Radiology, 21(7), 859–868. https://doi.org/10.3348/kjr.2020.0293
Taghanaki, S. A., Zheng, Y., Zhou, S., Georgescu, B., Sharma, P., Xu, D., Comaniciu, D., & Hamarneh, G. (2019). Combo loss: Handling input and output imbalance in multi-organ segmentation. Computerized Medical Imaging and Graphics, 75, 24–33. https://doi.org/10.1016/j.compmedimag.2019.04.005
Tang, Z., Zhao, W., Xie, X., Zhong, Z., Shi, F., Ma, T., Liu, J., & Shen, D. (2021). Severity assessment of COVID-19 using CT image features and laboratory indices. Physics in Medicine & Biology, 66, 035015. https://doi.org/10.1088/1361-6560/abbf9e
Zhu, W., Huang, Y., Zeng, L., Chen, X., Liu, Y., Qian, Z., Du, N., Fan, W., & Xie, X. (2019). AnatomyNet: Deep learning for fast and fully automated whole-volume segmentation of head and neck anatomy. Medical Physics, 46, 576–589. https://doi.org/10.1002/mp.13300
Zhu, X., Men, J., Yang, L., & Li, K. (2022). Imbalanced driving scene recognition with class focal loss and data augmentation. International Journal of Machine Learning and Cybernetics, 13(6). https://doi.org/10.1007/s13042-022-01575-x

Details

Primary Language

English

Subjects

Pattern Recognition, Machine Learning (Other), Artificial Intelligence (Other)

Journal Section

Review

Authors

Umut Kaan Kavaklı
0009-0003-4968-4124
Türkiye

Göksu Turaç
0009-0002-3480-2194
Türkiye

Sema Candemir ^*
0000-0001-8619-5619
Türkiye

Publication Date

September 3, 2026

Submission Date

March 30, 2026

Acceptance Date

June 4, 2026

Published in Issue

Year 2026 Volume: 29 Number: 3

IZ

https://izlik.org/JA24EU57JM

Cite

RIS / Bibtex

APA

Kavaklı, U. K., Turaç, G., & Candemir, S. (2026). INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS. Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi, 29(3), 1399-1411. https://izlik.org/JA24EU57JM

AMA

1.Kavaklı UK, Turaç G, Candemir S. INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS. KSU J. Eng. Sci. 2026;29(3):1399-1411. https://izlik.org/JA24EU57JM

Chicago

Kavaklı, Umut Kaan, Göksu Turaç, and Sema Candemir. 2026. “INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS”. Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi 29 (3): 1399-1411. https://izlik.org/JA24EU57JM.

EndNote

Kavaklı UK, Turaç G, Candemir S (September 1, 2026) INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS. Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi 29 3 1399–1411.

IEEE

[1]U. K. Kavaklı, G. Turaç, and S. Candemir, “INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS”, KSU J. Eng. Sci., vol. 29, no. 3, pp. 1399–1411, Sept. 2026, [Online]. Available: https://izlik.org/JA24EU57JM

ISNAD

Kavaklı, Umut Kaan - Turaç, Göksu - Candemir, Sema. “INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS”. Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi 29/3 (September 1, 2026): 1399-1411. https://izlik.org/JA24EU57JM.

JAMA

1.Kavaklı UK, Turaç G, Candemir S. INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS. KSU J. Eng. Sci. 2026;29:1399–1411.

MLA

Kavaklı, Umut Kaan, et al. “INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS”. Kahramanmaraş Sütçü İmam Üniversitesi Mühendislik Bilimleri Dergisi, vol. 29, no. 3, Sept. 2026, pp. 1399-11, https://izlik.org/JA24EU57JM.

Vancouver

1.Umut Kaan Kavaklı, Göksu Turaç, Sema Candemir. INHERENT CLASS IMBALANCE IN LESION SEGMENTATION: THE EFFECT OF LOSS FUNCTIONS. KSU J. Eng. Sci. [Internet]. 2026 Sep. 1;29(3):1399-411. Available from: https://izlik.org/JA24EU57JM