Ongoing
Preprint forthcoming
PMMI · Imperial College London · since 2025
Prescription‑Mediated Multi‑Morbidity Index
Multimorbidity is usually measured by counting diagnoses, as established indices such as Charlson and Elixhauser do. PMMI is built from GP prescription records instead. The question is whether the medicines a person takes carry information about their mortality risk that their diagnoses miss.
- PrescriptionsGP prescription records in UK Biobank, grouped into sex-specific British National Formulary categories
- SelectionElastic-net Cox regression and stability selection calibrated with SHARP, to find a small set that holds up when the data change
- CheckRandom survival forests, to see what a non-linear model would pick
- ScoreA short integer score you can read off without a model
Developed from my bachelor thesis. Supervised by Dr Dragana Vuckovic and Prof. Marc Chadeau-Hyam, with Dr Rin Wada. EBS Omics group, Department of Epidemiology & Biostatistics, School of Public Health.
Finished
King’s College London, BMEIS · summer 2025
Checking X-ray image quality automatically
Hospitals check their fluoroscopy systems by imaging a test object, a phantom, and finding the finest pattern of lines the system can still resolve. This is usually done by eye, which is slow and varies between readers.
At the research summer school at St Thomas’ Hospital I wrote a MATLAB pipeline that automates this reading for the Leeds TOR 18FG phantom. It takes about a second per image and resolves detail roughly 50% finer than human readers. Results were validated against ImageJ.
- PrepareClean up the DICOM image
- LocateFind the phantom’s rim with Canny edges and a Hough transform
- AlignStraighten it with Powell optimisation and a Radon transform
- MeasureContrast in each of the 21 bar groups, cut off at MTF10
- ReportThe limiting resolution, in line pairs per mm
Supervised by Prof. Kawal Rhode, School of Biomedical Engineering & Imaging Sciences, King’s College London. With thanks to Carlo Saija for his support with MATLAB.