This research introduces a graph neural network that integrates pathway-based embeddings for identifying rare genetic diseases. The model leverages biomedical graphs to detect phenotypic correlations and improves diagnostic accuracy by learning relational structures across multiple genomic and proteomic datasets.

Christopher Alan Wexler, Olivia Harper Blaine, Sebastian Louis Granger, Isobel Claire Fielding, Frederick Owen Langford

A multi-omics deep learning model is proposed to predict individual drug responses in cancer treatment. Using autoencoder fusion, genomic, transcriptomic, and proteomic features are integrated to personalize therapeutic strategies and enhance outcomes in precision oncology initiatives.

Elena Margaret Crosswell, Tobias Graham Finch, Joanna Isabel Pritchard, Thomas Harvey Blanchard, Charlotte Daisy Fenwick

This paper presents a real-time biometric framework to monitor Parkinson's symptoms using wearable sensors. Time-series analytics and anomaly detection models are implemented to track tremors, gait changes, and motor fluctuations, aiding early diagnosis and personalized intervention strategies for neurodegenerative disorders.

Juliet Annabelle Stone, Hugo Ellis Cartwright, Clara Sophie Maynard, Oscar Harrison Welling, Phoebe Madeleine Eastwood

We propose a hybrid deep neural architecture combining convolutional and transformer layers for segmenting retinal structures in fundus images. The model improves detection accuracy of diabetic retinopathy and macular degeneration while preserving anatomical detail crucial for clinical ophthalmology applications.

Benjamin Isaac Thorne, Amelia Florence Redman, Joseph Edward Langley, Grace Victoria Tennant, Edward Arthur Hollingsworth

An attention-enhanced U-Net architecture using transfer learning is developed for semantic segmentation in cancer histopathology slides. The model assists pathologists in automated grading of malignancies, improving diagnostic throughput and reducing inter-observer variability in digital pathology workflows.

Florence Helena Goodwin, Samuel Oliver Lytton, Madeleine Rose Berrington, George Maxwell Eddington, Isabella Alice Cartledge

This paper presents a residual convolutional neural network designed to analyze multichannel EEG signals for seizure detection. The model captures spatiotemporal dependencies across frequency bands, providing accurate, real-time predictions that aid neurologists in early clinical intervention for epilepsy management.

Edward Nathaniel Briggs, Francesca Louise Copeland, Henry Theodore Clarkson, Matilda Grace Hollister, Jonathan Charles Rowley

An attention-based bidirectional LSTM architecture is developed to predict patient mortality from longitudinal EHR data. This model learns latent temporal patterns and prioritizes clinically significant events, enabling critical care units to make timely decisions with enhanced interpretability and performance.

Harriet Lucille Morton, Felix James Dunmore, Charlotte Evangeline Rowntree, Thomas Julian Hadley, Eleanor Scarlett Bexley