Optimizing Vision Transformers for Histopathology: Pretraining and Normalization in Breast Cancer Classification
Journal of Imaging · 30.04.2024 · 10 (5) · 108
DOI: 10.3390/jimaging10050108 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Imaging, online erschienen am 30.04.2024. Verfasst von Giulia Lucrezia Baroni und 5 weiteren Autor:innen. Beteiligte Einrichtung: Department of Mathematics, Computer Science and Physics, University of Udine, 33100 Udine, Italy. Die Arbeit wurde bisher 45-mal zitiert (Crossref) und nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: HORIZON EUROPE Marie Sklodowska-Curie Actions, BosomShield project, Next-Generation EU.
Abstract (Originalsprache)
This paper introduces a self-attention Vision Transformer model specifically developed for classifying breast cancer in histology images. We examine various training strategies and configurations, including pretraining, dimension resizing, data augmentation and color normalization strategies, patch overlap, and patch size configurations, in order to evaluate their impact on the effectiveness of the histology image classification. Additionally, we provide evidence for the increase in effectiveness gathered through geometric and color data augmentation techniques. We primarily utilize the BACH dataset to train and validate our methods and models, but we also test them on two additional datasets, BRACS and AIDPATH, to verify their generalization capabilities. Our model, developed from a transformer pretrained on ImageNet, achieves an accuracy rate of 0.91 on the BACH dataset, 0.74 on the BRACS dataset, and 0.92 on the AIDPATH dataset. Using a model based on the prostate small and prostate medium HistoEncoder models, we achieve accuracy rates of 0.89 and 0.86, respectively. Our results suggest that pretraining on large-scale general datasets like ImageNet is advantageous. We also show the potential benefits of using domain-specific pretraining datasets, such as extensive histopathological image collections as in HistoEncoder, though not yet with clear advantages.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Mathematics, Computer Science and Physics, University of Udine, 33100 Udine, Italy.
- Istituto di Anatomia Patologica, Azienda Sanitaria Universitaria Friuli Centrale, 33100 Udine, Italy.
ORCID-Profile: Giulia Lucrezia Baroni, Laura Rasotto, Kevin Roitero, Vincenzo Della Mea
Förderung
- HORIZON EUROPE Marie Sklodowska-Curie Actions, BosomShield project
- Next-Generation EU
Zitieren
APA
Baroni, G. L., Rasotto, L., Roitero, K., Tulisso, A., Di Loreto, C., & Della Mea, V. (2024). Optimizing Vision Transformers for Histopathology: Pretraining and Normalization in Breast Cancer Classification. Journal of Imaging, 10(5), 108. https://doi.org/10.3390/jimaging10050108
Vancouver
Baroni GL, Rasotto L, Roitero K, Tulisso A, Di Loreto C, Della Mea V. Optimizing Vision Transformers for Histopathology: Pretraining and Normalization in Breast Cancer Classification. Journal of Imaging. 2024;10(5):108. doi:10.3390/jimaging10050108
BibTeX
@article{Baroni2024,
author = {Baroni, Giulia Lucrezia and Rasotto, Laura and Roitero, Kevin and Tulisso, Angelica and Di Loreto, Carla and Della Mea, Vincenzo},
title = {Optimizing Vision Transformers for Histopathology: Pretraining and Normalization in Breast Cancer Classification},
journal = {Journal of Imaging},
year = {2024},
volume = {10},
number = {5},
pages = {108},
doi = {10.3390/jimaging10050108}
}
EUVIMED