Research Outputs

No peer-reviewed publications yet. What follows is the research work that exists — dissertation, software and instruments in development — stated at its actual stage rather than dressed up. First submissions are planned alongside the GenECG-Diagnostic evaluation.

Software and models

GenECG-Diagnostic — generative ECG diagnostic model

In development, 2026. PyTorch · YOLOv8/YOLO11 · Vision Transformer (google/vit-base-patch16-224) · SNOMED-CT. Repository

A two-stage pipeline automating how a clinician reads a 12-lead ECG image. Stage one: a YOLO object detector locates and crops the twelve lead panels from a raw 3×4 grid, trained on a custom heuristically-labelled dataset. Stage two: a Vision Transformer fine-tuned with BCEWithLogitsLoss predicts multiple SNOMED-CT diagnostic codes per ECG, so more than one diagnosis is permitted per image. Evaluation uses mAP for lead detection and per-label F1/AUC for diagnosis; formal reporting of these metrics is the current work in progress.

PROBAST+AI scoring tool

In development. A structured instrument for appraising risk of bias and applicability in clinical prediction models and medical foundation models, addressing a gap in how such models are currently critiqued.

Dissertation

Ethical Challenges Facing the Implementation of Gamification in Healthcare: A Systematic Literature Review & Thematic Framework

MSc dissertation, Swansea University Medical School, September 2022. 60 credits. Unpublished.

PRISMA-guided systematic review with a SPIDER search strategy and CASP quality appraisal, analysed by thematic analysis. Identified seven interacting ethical themes — harm, privacy, manipulation and exploitation, inequality and stereotype reinforcement, data ownership, the merging of real and virtual worlds, and shifting power dynamics — and mapped their cause-and-effect relationships into an interactive framework for clinicians, digital health developers and health IT policymakers.

Research proposals

Multimodal Diagnostic Decision Support System for Cardiovascular Disease Classification: An Experimental Study

Research proposal, University of Edinburgh, 2025.

Retrospective experimental design using the DataLoch Heart Disease Registry within the NHS Lothian Trusted Research Environment, under UK GDPR, the Data Protection Act 2018 and the 5 Safes framework. Fuses a visual modality (12-lead ECG images) with tabular demographics and vital signs through intermediate tensor-level fusion, recalibrated toward sensitivity because a missed cardiac event costs more than a false alarm. Reporting to TRIPOD+AI; risk of bias scored with PROBAST+AI.

Understanding the Role of Wearable Devices in Supporting Cardiovascular Disease Patients

Research proposal, Swansea University, July 2022. Morriston Hospital, Swansea Bay University Health Board.

Qualitative phenomenological design using one-to-one semi-structured interviews with purposive sampling, investigating how cardiovascular patients experience wearable biometric monitoring, and the usability, fatigue and privacy barriers that shape adoption.

Critical appraisal

Critical Appraisal of "Leveraging Electronic Health Records and Knowledge Networks for Alzheimer's Disease Prediction and Sex-Specific Biological Insights"

University of Edinburgh, 2025.

Methodological audit of a UCSF EHR cohort study combining Random Forest and XGBoost models with the SPOKE knowledge graph — examining train-test protocol, cross-validation, AUROC/AUPRC discrimination, selection bias, missing-data handling and data leakage across longitudinal visits, alongside the sex-stratified hazard ratios the paper reports.

Applied analyses

Seven further data-science projects with documented results and code — Apache Spark readmission prediction across ~100,000 encounters, a nine-year SQL analysis of NHS Scotland prescribing, OECD policy analytics and others — are listed under the folios, with repositories on GitHub.