Background

Although artificial intelligence (AI) tools may aid early diagnosis and prognosis of neurodegenerative disorders, their adoption in memory clinics remains limited. While general guidelines for the development and use of AI in healthcare exist, they are not specific to memory clinics in terms of their goals, users, and challenges, hampering their effective design and implementation. We sought to establish expert consensus for the design, development, and deployment of neuroimaging AI tools for memory clinics using a Delphi process.

Methods

We leveraged the collective knowledge of 16 experts from clinical, academic, and industry backgrounds who were recruited through the Deep Dementia Phenotyping (DEMON) network. During the Delphi process, we conducted three rounds of anonymous questionnaires, with support from a Patient and Public Involvement Representative. In the first round of the Delphi process, the panel answered open-ended questions to identify important concerns on a wide range of topics related to neuroimaging AI tools. In the second round, panelists prioritised concerns on a 7-point Likert scale. In the third round, panelists gained insight into the perspective of their peers through controlled feedback, and either agreed or disagreed with a consensus statement based on the group’s mean Likert score from the previous round.

Results

The panel reached strong (>70%) or moderate (50-69%) agreement on eight statements covering four themes: purpose, quality, interactions with clinicians and memory clinics, and privacy and ethics. Overall, the panel agreed neuroimaging AI tools can contribute to improved patient care at various levels, and identified variable individual-level performance and lack of generalisability as key barriers for implementation in memory clinics. These statements form a first step towards more comprehensive guidelines for neuroimaging AI tools for memory clinics, highlighting experts’ recommendations regarding development, reporting, evaluation, testing, education, and data handling.

Conclusion

The Neuro-AIM expert concensus establishes a comprehensive and novel range of recommendations to further the translation of AI neuroimaging tools from research to practice, improving diagnostic and care pathways for all memory clinic stakeholders.