Harnessing Self-Supervised Features for Histopathological Image Retrieval and Classification Through Efficient Fine-Tuning

• José Solenir L. Figuerêdo UEFS
• Luciano Araújo D. Filho UEFS
• Rodrigo Tripodi Calumby UEFS

DOI: https://doi.org/10.5753/sbcas.2025.7677

Resumo

This study evaluates machine learning models pre-trained using the self-supervised architectures iBOT and DINOv2 for disease/lesion classification and histopathological image retrieval. Experiments were conducted using five datasets and explored three fine-tuning strategies: Full, Low-Rank Adaptation (LoRa), and Linear Probe. The LoRa technique yielded significant effectiveness gains, improving model effectiveness by over 40% in some scenarios (LoRa vs Full). In image retrieval, the DINO-Hist model outperformed iBOT-Hist, demonstrating statistically significant superiority in MAP@10 and MAP@40. These findings underscore the adaptability of self-supervised architectures and the critical role of fine-tuning in enhancing model effectiveness.

Referências

Abels, E. et al. (2019). Computational pathology definitions, best practices, and recommendations for regulatory guidance: a white paper from the digital pathology association. J Pathol, 249(3):286–294.

Baharoon, M. et al. (2024). Evaluating general purpose vision foundation models for medical image analysis: An experimental study of dinov2 on radiology benchmarks.

Balestriero, R. et al. (2023). A cookbook of self-supervised learning.

Bueno, G. et al. (2020). Data for glomeruli characterization in histopathological images. Data in Brief, 29:105314.

Cerqueira, S. et al. (2021). Pathospotter classifier: Uma serviço web para auxílio à classificação de lesões em glomérulos renais. In Anais do XXI SBCAS, pages 60–70, Porto Alegre, RS, Brasil. SBC.

Chagas, P. et al. (2020). Classification of glomerular hypercellularity using convolutional features and support vector machine. Artif Intell Med, 103:101808.

Chen, L. et al. (2019). Self-supervised learning for mia using image context restoration. Med Image Anal, 58:101539.

Filiot, A. et al. (2023). Scaling self-supervised learning for histopathology with masked image modeling. medRxiv.

Kataria, T. et al. (2023). To pretrain or not to pretrain? a case study of domain-specific pretraining for semantic segmentation in histopathology. In Medical Image Learning with Limited and Noisy Data, pages 246–256, Cham. Springer Nature Switzerland.

L’Imperio, V. et al. (2021). Digital pathology for the routine diagnosis of renal diseases: a standard model. Journal of Nephrology, 34(3):681–688.

Majumdar, S. et al. (2023). Gamma function based ensemble of cnn models for breast cancer detection in histopathology images. Expert Systems with Applications, 213:119022.

Mohammad Alizadeh, S. et al. (2023). A novel siamese deep hashing model for histopathology image retrieval. Expert Systems with Applications, 225:120169.

Oquab, M. et al. (2023). Dinov2: Learning robust visual features without supervision.

Ozen, Y. et al. (2021). Self-supervised learning with graph neural networks for region of interest retrieval in histopathology. In ICPR, pages 6329–6334.

Wickstrøm, K. K. et al. (2023). A clinically motivated self-supervised approach for content-based image retrieval of ct liver images. CMIG, 107:102239.

Yamaguchi, R. et al. (2021). Glomerular classification using convolutional neural networks based on defined annotation criteria and concordance evaluation among clinicians. Kidney International Reports, 6(3):716–726.

Yang, P. et al. (2020). A deep metric learning approach for histopathological image retrieval. Methods, 179:14–25. Interpretable machine learning in bioinformatics.

Zheng, Y. et al. (2022). Encoding histopathology whole slide images with location-aware graphs for diagnostically relevant regions retrieval. Med Image Anal, 76:102308.