Artificial Intelligence in Practice
28 and 29 January 2021
This was the third AI meeting jointly organised by the BIR and RCR in the last two years, this year held online due to the COVID-19 pandemic. The meeting was jointly chaired by Drs. Hugh Harvey and Amrita Kumar, both of whom are consultant radiologists and members of the BIR Clinical Intelligence and Informatics special interest group .
From a position of uncertainty two years ago as to whether AI was a threat or opportunity to a position where the role of AI was being more clearly defined in clinical applications in the 2020 meeting, the current meeting covered a multitude of areas including clinical applications, workflow management, stakeholder engagement, addressing concerns about privacy, obtaining regulatory approval and technology assessment.
The virtual format of the current meeting provided an opportunity for the sponsors (OSL, Zebra Vision, Siemens Healthineers, Aidence and qure.ai) to showcase their products via “breakout” sessions which proved to be very informative to the participants.
The speakers included academic and clinical radiologists, a breast surgeon, industry representatives, clinical and non-clinical specialists from bodies such as NHSX and the Department of Trade involved in technology development, deployment and assessment.
Patient engagement in the development and deployment of AI was a theme emphasised by all speakers. This was particularly important for data collection and sharing, where current regulation including the GDPR has to be rigorously followed.
New insights into the causes of heart disease and risk stratification for survival prediction were notably highlighted by two cardiac radiologists, particularly for survival prediction by cardiac motion analysis in cardiomyopathies using deep learning and assessing coronary artery calcification on non-contrast chest CT.
Classification of chest radiographs as normal and abnormal using AI and prioritising the abnormal radiographs for further review including categorisation of these radiographs, detection and follow-up of lung nodules and standardising protocols for reporting these nodules were two other thoracic applications emphasising collaboration between AI developers and imaging departments. The role of AI in detecting and characterising COVID-19 related lung abnormalities was also highlighted by a consultant radiologist and industry representative.
Real-world deployment of AI in breast screening including challenges pertaining to data collection, sharing and analysis, besides ensuring that the process remains safe and trustworthy to patients was highlighted by a consultant radiologist actively involved in breast screening in Scotland and a breast surgeon who was herself diagnosed with and treated for breast cancer.
The role of AI in dementia imaging and in optimisation of a stroke treatment pathway were two areas where collaboration between clinicians and various stakeholders was well demonstrated. This was notable in the stroke treatment pathway, where the “door in door out (DIDO)” time was significantly reduced resulting in less patient morbidity and significant cost savings.
Factors influencing the adoption of AI including safety by minimising radiation dose and errors in interpretation, cost minimisation by reducing waste and increasing efficiency and legislation were highlighted. Concerns about data privacy and security and navigating the complex regulatory framework both in the UK and Europe for deployment and evaluation of AI were well discussed – in particular, terminology such as a clinical effectiveness report, periodic safety update report and unique device identifier were explained in a simple manner.
The panel discussions at all AI meetings have been very stimulating and the current meeting was no exception. The role of the AI buyer’s guide developed by NHSX, the challenges of information governance and information technology encountered by clinicians, the lack of knowledge about AI by procurement teams and the workload pressures in the NHS on a day- to-day basis making it difficult to facilitate innovation were all areas that provided a healthy and stimulating discussion.
The role of NHS AI-lab, part of NHSX in providing AI awards to developers to facilitate adoption of AI in the NHS, the role of health economics in the development and deployment of AI and the role of NICE in Health Technology Assessment (HTA) provided the delegates with insights into the depth of AI development, deployment and adoption in the UK.
The meeting ended with optimism that AI is here to stay. Rather than replace the clinician including the radiologist, AI will augment the role of a second clinician/second reader in many situations and provide the trust and reassurance that patients deserve, knowing that their concerns about regulation and privacy have been appropriately addressed.
Review by Dr Vijay Jayaram, Chair Clinical Intelligence and Informatics special interest group, BIR and Consultant Radiologist at Princess Alexandra Hospital, Harlow.