This research applies graph convolutional networks to combine genetic mutations and clinical variables for predicting cancer outcomes. The proposed hybrid model improves prognostic accuracy and provides interpretable biomarkers that support precision oncology and personalized therapeutic decision-making in real-world clinical environments.

Leonard Graham Whitmore, Isabelle Caroline Forbes, William Marcus Dowell, Juliet Naomi Bancroft, Adam Patrick Ellington

This paper presents an unsupervised deep clustering framework for histopathology images, utilizing multi-view representations and contrastive learning. The model groups cellular structures with high fidelity, facilitating scalable annotation and aiding discovery of novel pathological subtypes from whole-slide digital microscopy data.

Oliver Daniel Fairchild, Clara Isabelle Pennington, Hugo Maxwell Bristow, Rosalind Emily Thorne, Samuel Richard Deane

An ensemble of EfficientNet models is developed for automated diabetic retinopathy screening from retinal fundus images. The architecture integrates vascular feature maps to improve lesion detection and classification, offering scalable, cloud-based screening in primary healthcare and underserved clinical environments.

Henry Wallace Grafton, Amelia Rose Stanhope, Rupert Anthony Cotterill, Philippa Louise Winslow, Frederick George Hampton

This study introduces a dual-attention U-Net architecture for segmenting brain tumors from multi-modal MRI scans. The design uses channel and spatial attention with multi-scale skip connections to enhance boundary detection, achieving state-of-the-art performance on benchmark medical imaging datasets.

Julian Edward Lamont, Beatrice Alice Wycliffe, Maxwell Owen Farnsworth, Rebecca Annette Montrose, Cedric Jonathan Pellham

This research proposes a federated learning framework for cardiovascular risk prediction using decentralized hospital EHRs. The system preserves patient privacy while achieving predictive accuracy comparable to centralized models, enabling secure collaboration across institutions for population-scale cardiovascular disease modeling.

Alfred Malcolm Trevors, Madeline Isabelle Forsythe, Graham Thomas Langdon, Eloise Frances Kennington, Vincent Gregory Chilton

This work presents a multimodal transformer architecture combining MRI features and acoustic speech signals for early Alzheimer’s detection. The model captures complementary cues, enhancing sensitivity to cognitive decline and offering a scalable, non-invasive screening method for large-scale clinical and telehealth deployment.

Jonathan Maxwell Bainbridge, Charlotte Eloise Stratton, Edwin Philip Glencross, Diana Scarlett Winsmore, Bernard Louis Camden

This paper introduces an AI diagnostic assistant for rare metabolic diseases, integrating knowledge graphs and Bayesian reasoning to handle data scarcity. It delivers accurate, explainable outputs and facilitates clinical decision-making in complex, poorly understood disease contexts through structured biomedical ontologies.

Francesco Marco Benedetti, Clara Sophia Halberg, George Vincent Royston, Isabelle Madeleine Quenby, Thomas Jeremy Rutledge