Dynamic Alteration of the Loss Function in Convolutional Neural Networks for Medical Image Segmentation

• Vinicius Ferreira de Souza PUC Minas
• Alexei M. C. Machado PUC Minas / UFMG

DOI: https://doi.org/10.5753/sbcas.2026.21408

Resumo

This work presents a method for training convolutional neural networks (CNNs) based on the dynamic alteration of the loss function that guides model optimization. First, a comparative analysis of six different loss functions applied to CNNs for medical image segmentation is conducted. Through experiments with the colon cancer dataset from the Medical Segmentation Decathlon, distribution-based functions (Focal Loss, TopK Loss, and Binary Cross-Entropy) and region-based functions (Dice Loss, IoU Loss, and Tversky Loss) are evaluated. The results demonstrate that distribution-based functions consistently outperform region-based functions. Furthermore, the dynamic alteration of the functions achieved an increase in accuracy of up to 12% when compared to composite approaches. Qualitative analysis revealed that this approach not only improves quantitative metrics but also enhances the visual quality of segmentation. These findings provide valuable insights for optimizing CNN performance in medical image segmentation and contribute to the development of more accurate solutions in AI-assisted diagnosis.

Referências

Antonelli, M., Reinke, A., Bakas, S., Farahani, K., Kopp-Schneider, A., Landman, B. A., Litjens, G., Menze, B., Ronneberger, O., Heller, N., et al. (2022). The medical segmentation decathlon. Nature Communications, 13(1):4128.

Drozdzal, M., Vorontsov, E., Chartrand, G., Kadoury, S., and Pal, C. (2016). The importance of skip connections in biomedical image segmentation. In Deep Learning and Data Labeling for Medical Applications, pages 179–87. Springer.

Lin, T.-Y., Goyal, P., Girshick, R., He, K., and Dollár, P. (2020). Focal loss for dense object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(2):318–27.

Ma, J., Chen, J., Ng, M., Huang, R., Li, Y., Li, C., Yang, X., and Martel, A. L. (2021). Loss odyssey in medical image segmentation. Medical Image Analysis, 71:102035.

Milletari, F., Navab, N., and Ahmadi, S.-A. (2016). V-net: Fully convolutional neural networks for volumetric medical image segmentation.

Rahman, M. A. and Wang, Y. (2016). Optimizing intersection-over-union in deep neural networks for image segmentation. In Advances in Visual Computing, pages 234–44. Springer.

Ronneberger, O., Fischer, P., and Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention, pages 234–41.

Salehi, S. S. M., Erdogmus, D., and Gholipour, A. (2017). Tversky loss function for image segmentation using 3d fully convolutional deep networks. In Machine Learning in Medical Imaging, pages 379–87. Springer.

Tan, M. and Le, Q. (2019). Efficientnet: Rethinking model scaling for convolutional neural networks. In International Conference on Machine Learning, volume 97, pages 6105–14.

Taubert, O., Götz, M., Schug, A., and Streit, A. (2020). Loss scheduling for class-imbalanced image segmentation problems. In IEEE International Conference on Machine Learning and Applications, pages 426–31.

Tian, Y., Su, D., Lauria, S., and Liu, X. (2022). Recent advances on loss functions in deep learning for computer vision. Neurocomputing, 497:129–58.

Wang, Q., Ma, Y., Zhao, K., and Tian, Y. (2020). A comprehensive survey of loss functions in machine learning. Annals of Data Science, pages 1–26.

Wu, Z., Shen, C., and van den Hengel, A. (2016). Bridging category-level and instance-level semantic image segmentation.

Xu, H., He, H., Zhang, Y., Ma, L., and Li, J. (2023). A comparative study of loss functions for road segmentation in remotely sensed road datasets. International Journal of Applied Earth Observation and Geoinformation, 116:103159.