Texture classification in medical images through generative models and self-supervised learning

• Leonardo C. Gomide PUC Minas
• Alexei M. C. Machado PUC Minas

DOI: https://doi.org/10.5753/sbcas.2025.6934

Abstract

This work proposes a regularized Variational Autoencoder for texture analysis, composed of an encoder, a decoder and a predictor module, with a triple loss function that simultaneously regularizes the image encoding, its reconstruction, and classification. The method presents state-of-the-art classification accuracy for mammographies and generates more separable latent spaces that may contribute to texture analysis.

References

Ha, R. et al. (2018) Convolutional neural network based breast cancer risk stratification using a mammographic dataset. Academic Radiology.

Kingma, D. P. and Welling, M. (2022). Auto-encoding variational bayes. arXiv.

Lehman, C. et al. (2019) Mammographic breast density assessment using deep learning: Clinical implementation. Radiology.

Liu, W. et al. (2020) Towards visually explaining variational autoencoders. In IEEE Computer Vision and Pattern Recognition.

Oliveira, J. . (2008). Toward a standard reference database for computer-aided mammography. In Medical Imaging: Computer-Aided Diagnosis.

Rahaman, N. (2019). On the spectral bias of neural networks. In International Conference on Machine Learning.

Rungue, A. H. A. (2019). Analysis of SVM parametrization in the classification of mammographic texture images. In Brazilian Congress on Automation.

Shehab, M. (2022). Machine learning in medical applications: A review of state-of-the-art methods. Computers in Biology and Medicine.

Song, Y. (2023). Latent traversals in generative models as potential flows. arXiv.

van der Maaten, L. and Hinton, G. (2008). Visualizing high-dimensional data using t-sne. Journal of Machine Learning Research.

Zeiser, F. et al. (2021) Deepbatch: A hybrid deep learning model for interpretable diagnosis of breast cancer in whole-slide images. Expert Systems with Applications.