Thomas Allcock

Research Topic: Explainable AI in Digital Pathology: The Role of Visualising Deep Neural Networks in Supporting the Augmented Pathologist

Supervisors: Dr Rebecca Brannan, Dr Andy Hanby, Professor Andy Bulpitt

About Tom: Tom completed undergraduate and masters degrees in Physics at Nottingham.

Project Description: Explainable Artificial Intelligence for Diagnostic Digital Pathology is a project that addresses the opacity of current deep-learning cancer-diagnosis systems by designing and evaluating interpretable algorithms that match or surpass the accuracy of black-box models; building on the digitisation of whole-slide histopathology, the work first adapts a prototype-based network—originally created for fine-grained natural-image recognition—to breast-cancer sub-typing and grading, then develops an attention mechanism that highlights diagnostically salient regions and predicts Nottingham Prognostic Index groups by fusing primary-tumour and lymph-node slides, and finally integrates the prototype and attention approaches into a hybrid framework that provides multilevel explanations while extending to lung-cancer sub-type prediction; large annotated datasets are curated and processed, models are implemented in PyTorch and validated through cross-validation, interpretability is quantified via localisation fidelity and pathologist agreement alongside conventional accuracy metrics, and ablation studies isolate each explanatory component’s contribution; expected outputs include three validated XAI models, a rigorous evidence base showing that transparency need not compromise performance, and a clinically relevant foundation for decision-support tools that bolster pathologist trust, ease regulatory approval, and ultimately improve patient outcomes.