A generalisable pixel classifier for cellular phenotyping in multiplex immunofluorescence images

Osvaldo Ulises Garay, Louisa Elena Ambühl, Thomas G Bird, Eleanor Barnes, William L. Irving, Ryan Walkley, Ian A Rowe

Research output: Contribution to journalArticlepeer-review

Abstract

Objectives: This study aimed to evaluate the cost-effectiveness (CE) of four hepatocellular carcinoma (HCC) in the UK, the GAAD algorithm, which combines patient gender and age with Elecsys® alpha-fetoprotein (AFP) and Elecsys protein induced by vitamin K absence-II (PIVKA-II) biomarker assays, ultrasound (US), US+AFP and GAAD+US.
Methods: A de novo micro-simulation model was developed in Microsoft Excel® from the perspective of the UK National Health Service to calculate life years, quality-adjusted life-years (QALYs), costs, incremental CE ratios, and net monetary benefits. Parameters were sourced from peer-reviewed published literature, national guidelines, and public cost databases. Sensitivity and scenario analyses were performed to evaluate the impact of parameter and structural uncertainty on the results.
Results: In a simulated cohort of 100,000 patients, discounted costs and QALYs per patient were £8,663 and 6·066 for US, £9,095 and 6·076 for US+AFP, £8,719 and 6·078 for GAAD alone, and £9,114 and 6·086 for GAAD+US. At a CE threshold of £20,000/QALY, GAAD was the most cost-effective strategy; however, although most costly, GAAD+US was the most clinically effective. Sensitivity and scenario analyses indicated that HCC incidence, along with the costs associated with diagnostic performance influence CE.
Conclusion: Considering the cost of US and the low incidence of HCC in the UK, this study suggests that GAAD alone or in combination with US are cost-effective surveillance strategies compared with US and US+AFP. Whilst GAAD+US showed the highest QALY increase, GAAD alone is considered preferable with regard to CE; however, better performance estimates for GAAD+US are needed to confirm.
Original languageEnglish
JournalPLoS ONE
DOIs
Publication statusPublished - 3 Dec 2025

Fingerprint

Dive into the research topics of 'A generalisable pixel classifier for cellular phenotyping in multiplex immunofluorescence images'. Together they form a unique fingerprint.

Cite this