Public Funding for Metalynx Ltd

In this collaborative proposal, the Aival Analysis Lab will be used to evaluate commercial radiology AI products on imaging data representative of a broad Scottish population. We will consider two clinical treatment pathways in line with NHS Scotland service and procurement priorities: stroke triage and urgent suspicion of lung cancer triage, assessing at least three commercial products in each case. Furthermore, we will develop a fully integrated platform capable of monitoring the performance of AI products in deployment at clinical sites. The use of AI in healthcare can lead to improved outcomes and increased efficiency in healthcare systems. However, AI models are known to produce variable results when applied to different demographics or when changes are made to a given clinical workflow, causing their accuracy to decrease and reducing operator trust. Our AI model evaluation and monitoring tool provides a way of assessing whether AI models are operating as designed in new deployments. This then accelerates integration into new clinical sites as well as confirming that the desired performance is both initially achieved, and maintained over time. Ensuring that AI models behave as expected means that overall operational costs are reduced for both the vendor and healthcare user as any deviations in performance are found and addressed quickly. Our software provides the ability to assess the performance of different AI models on a given patient population. It can be deployed locally at any clinical site and used to report evidence of performance, fairness and robustness on data from local populations, operators and workflows. We report metrics across subgroups of the population, ensuring that benefits and risks are equally distributed. Our software can explain the reasoning behind the decisions of black-box AI models, providing reports with visual outputs that make this clear and accessible to users. This explainability enables clinical users to gain trust in the outputs of an AI model. Clinical sites also often wish to benchmark different solutions for a given disease indication against one another to determine which is most suitable for their patient population and use case. Our tool provides the functionality to directly generate reports enabling this comparison. The software is light, scalable and easy to integrate. In this project we aim to prove its efficacy in assessing and monitoring AI model performance and its ability to accelerate AI model deployment and reduce costs.

We propose a novel and radical solution that allows healthcare providers and hospitals to rapidly evaluate and test radiology AI products in an independent manner, at a fraction of the resource cost of current evaluation frameworks. An increasing and aging population, with more long-term health conditions, is putting increasing strain on the healthcare system. AI can alleviate this if adopted safely and effectively \[1\]. However, current pathways to examine the trustworthiness and performance of commercial AI products are neither standardised nor scalable. Over 80% of hospital pathways involve imaging and its expert interpretation by radiologists \[11\]. However, England alone had a shortfall of 1,453 clinical radiologist consultants (2021), with clinical radiology directors reporting insufficient numbers for safe and effective patient care. \[11\]. AI and machine learning tools in clinical radiology promise to improve detection rates, streamline clinical workflow and improve patient safety. However, despite early enthusiasm, adoption into routine practice has been slow \[2\]. A key reason for this is that, prior to deployment, each product must be externally and independently validated on local data to ensure reproducibility, generalisability and trustworthiness, along with operational performance and points of failure \[3\]. However, many studies and trials evaluating radiology AI products are hampered by bias, lack of blinding and small datasets and population numbers \[5\]. Furthermore, AI products operate in so-called "black-boxes" whereby the method in which the product operates is unclear to the lay user, creating issues around credibility and trustworthiness. This poses significant cost and potential clinical risk to institutions looking to deploy AI, creating a major barrier to widespread adoption. There is a clear unmet need to develop less resource intensive processes by which organizations can substantiate medical device manufacturer claims and identify products that offer clear clinical benefit to healthcare providers. We propose to create a standardised system to validate commercial healthcare imaging AI products at local clinical sites before procurement. Our solution assesses the performance, fairness, robustness and explainability of a black-box product. This can be used by non-technical clinical managers to quickly evaluate and compare AI products, ensuring suitability for their local populations and work schemes.

**Unlocking the potential of Computer Vision for SMEs - Development of an AI platform enabling non-technical users to build advanced CV applications** There is a significant amount of useful information and value locked in visual images, however the high cost of computer vision expertise and the time it takes to analyse and curate the data, means it's currently impossible for UK SMEs to fully realise its potential. From helping doctors diagnose medical conditions, to detecting crop disease by drones, to ensuring safety protocols on construction sites, computer vision has the potential to unlock new growth and efficiency opportunities across a wide variety of sectors. Unfortunately SMEs rarely have the CV expertise required, and the cost to hire/contract this expertise can be prohibitively expensive. Even then highly-skilled CV scientists are spending ~80% of their time on data curation alone\[3\], with their lack of domain expertise then leading to underperforming, unusable models\[4,5\]. Such challenges result in 55% of projects never making it to production\[3\]. To increase adoption, solutions need to speed-up data curation processes, reduce curation costs and become more accessible to those without technical training. Organisations such as hospitals and manufacturers own both vast amounts of visual data and the domain expertise to interpret it, making them ideally placed to create impactful CV applications. Empowering them to build their own application can help accelerate AI-adoption and the resulting benefits. Collaborating with medical imaging experts at King's College London and retinal imaging SME Optos, Metalynx aims to significantly increase CV adoption amongst UK SMEs by removing the key barriers to adoption -- cost and expertise -- through the development of accessible tools to build advanced CV applications, easily. The ultimate aim of the project is to develop an easy to use, low cost CV technology stack for SMEs that can be used to create advanced applications, without any technical training and can seamlessly integrate into existing works flows to unlock value in already collected data.