Our new report From ‘Black Box’ to Trusted Healthcare Tools looks at physiology’s role in unlocking the potential of artificial intelligence for health. The report launched in the Houses of Lords on 27 June 2023.
AI Tools in Healthcare
The UK health system is under ever increasing pressure due to public health challenges such as a growing and ageing population coupled with an overburdened NHS. The ONS estimates that by 2045 4.3% of the country’s population (3.1 million people) will be aged 85 years and over. This significant demographic shift means that the health system needs to adapt to ensure its fit for the future by rapidly diagnosing disease and preventing ill health.
Artificial intelligence and machine-learning technologies are already being used to improve health outcomes. For example, data from patient-monitoring devices collecting physiological measurements such as heart rate, blood pressure and oxygen saturation is being used to personalise care; similarly, data on sleep quality can be used to suggest behavioural changes.
However, work in AI tends to happen in silos. Since healthcare is a vast field, encompassing numerous specialties and sub-specialties, AI developers tend to focus on specific areas of expertise, leading to the creation of specialised tools that only address specific medical conditions or processes. There is a need for cross-disciplinary collaboration to enhance our understanding of, and trust in, the results generated by AI technology which is often seen as a ‘black box’.
Physiological knowledge is essential to improving AI algorithms as it can provide insight into the underlying biological processes and mechanisms that drive various health conditions. This domain knowledge can help inform the development of AI algorithms and ensure that they accurately model the plausible physiological processes and reduce risk of identifying confounding factors. Further, physiologists can interpret and contextualise the data used to train AI models, ensuring that they contain plausible measurements and are representative against known standards for the target end users.
