Project description
Histopathology has contributed significantly to our understanding of biological phenomena and many diseases. It typically involves pathologists visually evaluating a tissue sample under a light microscope to identify structural tissue properties associated with disease. The growing use of Whole-Slide Imaging (WSI) in large-scale, high-throughput digital pathology introduces a number of new scientific challenges, including:
- very large image sizes and volumes of data;
- signal variance — i.e. intra-stain variance — and staining variance — i.e. inter-stain variance; and
- highly heterogeneous images.
These factors make it difficult to apply conventional image processing algorithms, and robustness across analyses from different centres remains an open problem. HistoGraph brings together three computer science laboratories — two specialising in AI, machine learning and medical image analysis, and one specialising in machine learning with graph representations — to work with medical institutes to address this challenge.
The consortium is developing AI-based diagnostic approaches through WSI analysis, aiming to:
- segment multiple anatomical structures and cells in WSIs from multiple sites (hospitals) using deep learning approaches, without requiring additional annotations;
- develop graph-based approaches that naturally capture the spatial context of segmented objects and enable the integration of multi-modal information;
- provide interpretable outputs that yield additional information for both the diagnostic process and its users; and
- rigorously evaluate the approach using standard datasets, with a view to potential clinical application in collaboration with pathologists.
This will be achieved through the following five work packages, each carried out jointly by two laboratories:
- WP1 – Data Collection
- WP2 – Segmentation & Domain Invariance
- WP3 – Graph Modelling
- WP4 – Interpretability
- WP5 – Diagnosis Approach & Evaluation
WP1–3 are led jointly by ICube and IHU, and WP4–5 by l'X and IHU.