This study presents a real-time ensemble model integrating multisensor wearable data to predict cardiovascular events. The framework enables early alerts by analyzing physiological markers such as HRV, respiration, and movement using decision trees and boosting algorithms on embedded platforms.

Oscar Nathaniel Ford, Petra Elisabeth Hansen, Lucas Gregory Barrett, Isabelle Sophie Lang, George Patrick McAllister

This paper proposes a vision transformer-based system to classify diabetic retinopathy and macular degeneration. By incorporating self-attention, our model outperforms CNN baselines in accuracy and interpretability while highlighting critical retinal lesions for ophthalmologist review and training.

Chloe Francesca Bell, Xiaojun Li, Henrik Carl Norberg, Isabella Grace Flynn, Matthew David Thatcher

This study utilizes a time-aware deep learning architecture that combines clinical, surgical, and biometric data streams to predict adverse postoperative outcomes. Our method captures complex dependencies across multiple timelines and medical parameters, enhancing risk scoring accuracy.

Daniel Christopher Walsh, Li Na Zhang, William Eric Houghton, Sophie Margaret Clarke, Jan Philipp Schneider

A hybrid convolutional-recurrent framework is introduced for classifying seizure activity in continuous EEG streams. The system identifies early ictal patterns with high sensitivity, enabling real-time alerts and improving automated monitoring in remote neurology clinics and ICU settings.

Victor Emmanuel Rhodes, Hao Wen Liu, Clara Madeleine Bouchard, Thomas Alan Goodman, Emily Rose Peterson

This research applies Bayesian networks to study causal relationships between biomarkers in Alzheimer’s patients. Our approach models temporal progression using longitudinal data, offering interpretable inferences for identifying key early indicators of cognitive decline in neurodegenerative conditions.

George Alexander Novak, Julia Theresa Weber, Benjamin Scott Fields, Natalie Christine Rowe, Erik David Johansson

This work explores a federated learning framework that enables decentralized diabetic retinopathy diagnosis across multiple hospitals. Without sharing sensitive data, the model aggregates insights from isolated sources while ensuring privacy compliance and enhancing diagnostic performance in real-world ophthalmic imaging deployments.

Harvey Leon Chapman, Felix André Duval, Amelia Charlotte Briggs, Maximilian Hugo Voss, Grace Eleanor Thornton

A self-supervised deep clustering model is introduced to delineate tumor boundaries in radiomics data. By leveraging contrastive representation learning, the method boosts segmentation precision, particularly in low-contrast regions, offering improvements in surgical planning and treatment outcome assessments for oncology patients.

Isla Catherine Boyd, Roman Eduard Klein, Daphne Louise Mercer, Julian Oscar McPherson, Charlotte May Whittaker