Deep Learning for Histological & Radiological Analysis
Neural network architectures for tumour characterisation from whole-slide images and radiology scans.
Overview
Modern deep learning architectures (CNNs, Vision Transformers, graph networks) enable extraction of high-dimensional features from histopathological slides and radiological images far beyond what classical radiomics can capture.
Applications
- Sarcoma grading — FNCLCC grade prediction from H&E whole-slide images
- Renal tumour subtyping — automated pathology classification
- Glioma characterisation — IDH mutation and MGMT methylation status prediction from pre-operative MRI
Key methodological contributions
- Multiple-instance learning (MIL) frameworks for slide-level labels
- Multi-scale feature aggregation across magnification levels
- Uncertainty quantification and explainability (GradCAM, attention maps)
- Prospective multi-centric validation