Researchers from the University of Edinburgh, including Shuyang Wu, Yifu Qiu, Ines P. Nearchou, Sandrine Prost, Jonathan A. Fallowfield, Hakan Bilen, and Timothy J. Kendall, have developed a new approach to improve the analysis of whole-slide images (WSIs) in computational pathology. Their work, published in the journal Nature Communications, focuses on enhancing the efficiency and accuracy of multiple-instance learning (MIL) techniques used in this field.
Computational pathology involves the use of digital images to analyze tissue samples, often at multiple magnifications to capture different features. Traditional methods rely on multiple inputs across these magnifications, which can be inflexible and computationally expensive. The researchers propose a new framework called the Multi-scale Pyramidal Network (MSPN) that addresses these challenges.
The MSPN is designed to be a plug-and-play module that can be added to existing attention-based MIL frameworks. It consists of two main components: grid-based remapping and the coarse guidance network (CGN). Grid-based remapping uses high magnification features to derive coarse features, while the CGN learns the context of these coarse features. This approach allows the network to retain the link between features across different scales, making it more efficient and flexible.
The researchers benchmarked the MSPN against four attention-based frameworks and a pre-trained MIL framework across four clinically relevant tasks and three types of foundation models. They found that the MSPN consistently improved the performance of MIL across all configurations and tasks. The MSPN is also lightweight and easy to use, making it a practical solution for enhancing the analysis of whole-slide images in computational pathology.
While this research is primarily focused on the medical field, the principles of multi-scale analysis and efficient feature learning could have broader applications in the energy sector. For example, similar techniques could be used to analyze satellite images for renewable energy site selection, where features at different scales (e.g., landscape, local terrain, and vegetation) need to be considered. Additionally, the lightweight and plug-and-play nature of the MSPN could make it a valuable tool for integrating advanced analytics into existing energy management systems.
Source: Wu, S., Qiu, Y., Nearchou, I.P. et al. Enabling progressive whole-slide image analysis with multi-scale pyramidal network. Nat Commun 14, 1234 (2023). https://doi.org/10.1038/s41467-023-36961-1
This article is based on research available at arXiv.