Impact of balancing and regularization on semantic segmentation of histopathological images

• Eduardo Henrique Souza Paraíso PUC Minas
• Alexei Manso Correa Machado PUC Minas / UFMG

DOI: https://doi.org/10.5753/sbcas_estendido.2025.6955

Abstract

This study investigates the impact of class balancing and regularization on improving diagnostic agreement in histological images. For example, U-Net models applied to the Prostate Cancer Grade Assessment dataset show that class balancing, combined with traditional loss functions, increases image agreement by up to 6 percentage points. Combining balancing with Focal Loss boosts classification agreement by an average of 13 percentage points compared to using imbalanced datasets with traditional loss functions. A case study on the analysis of prostate Gleason patterns 3 and 4 illustrates the importance of this discussion to clinical decisions and the prognosis of prostate cancer patients.

References

Bulten, W., Kartasalo, K., Chen, P. C., et al. (2022). Artificial intelligence for diagnosis and Gleason grading of prostate cancer: the PANDA challenge. Nat Med, 28(1):154–163.

Chen, Z. (2023). Medical Image Segmentation Based on U-Net. J. Phys.: Conf. Ser., 2547(1):012010.

Falahkheirkhah, K., Tiwari, S., Yeh, K., et al. (2023). Deepfake Histologic Images for Enhancing Digital Pathology. Laboratory Investigation, 103(1):100006.

Gleason, D. F. e Mellinger, G. T. (1974). Prediction of Prognosis for Prostatic Adenocarcinoma by Combined Histological Grading and Clinical Staging. Journal of Urology, 111(1):58–64.

Guerrero, E. D., Lina, R., Lina, R., et al. (2024). A Data Augmentation Methodology to Reduce the Class Imbalance in Histopathology Images. J Digit Imaging. Inform. med.

Haghofer, A., Fuchs-Baumgartinger, A., Lipnik, K., et al. (2023). Histological classification of canine and feline lymphoma using a modular approach based on deep learning and advanced image processing. Sci Rep, 13:19436.

Hancer, E., Traoré, M., Samet, R., et al. (2023). An imbalance-aware nuclei segmentation methodology for H&E stained histopathology images. Biomedical Signal Processing and Control, 83:104720.

Ikromjanov, K., Bhattacharjee, S., Hwang, Y.-B., et al. (2022). Whole Slide Image Analysis and Detection of Prostate Cancer using Vision Transformers. In 2022 ICAIIC, p. 399–402, Jeju Island, Korea, Republic of.

INCA, I. N. D. C. (2023). Estimativa 2023: incidência de câncer no Brasil. Instituto Nacional De Câncer, Rio de Janeiro, RJ.

Lin, T.-Y., Goyal, P., Girshick, R., et al. (2018). Focal Loss for Dense Object Detection.

Meyes, R., Lu, M., de Puiseau, C. W., et al. (2019). Ablation Studies in Artificial Neural Networks.

Ozkan, T. A., Eruyar, A., Cebeci, O., et al. (2016). Interobserver variability in Gleason histological grading of prostate cancer. Scandinavian Journal of Urology, 50(6):420–424.

Raciti, P., Sue, J., e Rodrigo, C. (2020). Novel artificial intelligence system increases the detection of prostate cancer in whole slide images of core needle biopsies. Modern Pathology, 33(10):2058–2066.

Ronneberger, O., Fischer, P., e Brox, T. (2015). U-Net: Convolutional Networks for Biomedical Image Segmentation. In MICCAI 2015, volume 9351, p. 234–241.

Silva-Rodríguez, J., Colomer, A., Sales, M. A., et al. (2020). Going deeper through the Gleason scoring scale: An automatic end-to-end system for histology prostate grading and cribriform pattern detection. Computer Methods and Programs in Biomedicine, 195:105637.