A4 Article in conference proceedings
Self-supervised Contrastive Pre-training Integrated with Multi-level Co-attention for Survival Prognosis from Whole Slide Images (2024)

Gao, J., Hao, X., Jin, S., & Xu, H. (2024). Self-supervised Contrastive Pre-training Integrated with Multi-level Co-attention for Survival Prognosis from Whole Slide Images. In P. You, S. Liu, & J. Wang (Eds.), Proceedings of International Conference on Image, Vision and Intelligent Systems 2023 (ICIVIS 2023) (pp. 650-658). Springer. Lecture Notes in Electrical Engineering, 1163. https://doi.org/10.1007/978-981-97-0855-0_62

JYU authors or editors

Publication details

All authors or editors: Gao, Junxiu; Hao, Xinyu; Jin, Shan; Xu, Hongming

Parent publication: Proceedings of International Conference on Image, Vision and Intelligent Systems 2023 (ICIVIS 2023)

Parent publication editors: You, Peng; Liu, Shuaiqi; Wang, Jun

Conference:

Place and date of conference: Baoding, China, 16.-18.8.2023

ISBN: 978-981-97-0854-3

eISBN: 978-981-97-0855-0

Journal or series: Lecture Notes in Electrical Engineering

ISSN: 1876-1100

eISSN: 1876-1119

Publication year: 2024

Publication date: 25/02/2024

Number in series: 1163

Pages range: 650-658

Number of pages in the book: 770

Publisher: Springer

Publication country: Singapore

Publication language: English

DOI: https://doi.org/10.1007/978-981-97-0855-0_62

Publication open access: Not open

Publication channel open access: 

Abstract

Survival analysis is of paramount importance in guiding the development of optimal treatment strategies for cancer patients. Because of the rich prognostic information contained in whole slide images (WSIs), multiple instance learning (MIL) approaches integrated with WSI analysis have been widely used in survival risk prediction. However, existing MIL methods often fail to encompass the complete range of histological image features, including critical information from local tissue regions, which limits their performance in survival prognosis. To address this limitation, we employ a self-supervised learning mechanism to train a feature extractor which can capture the intricate characteristics of WSIs. Furthermore, we introduce an attention mechanism that incorporates local patches and clusters to guide the fusion of multi-level features for survival outcome prediction. The proposed method demonstrates excellent performance on the widely recognized TCGA-COAD dataset. Experimental findings indicate that the integration of pre-trained feature extractors with MIL method and the fusion of multi-level histological features yield notable advancements in survival risk predictions.

Keywords: cancerous diseases; forecasts; histology; computer vision; machine learning

Contributing organizations

Ministry reporting: Yes

Reporting Year: 2024

Preliminary JUFO rating: 1