The use of artificial intelligence (AI) in dementia diagnostics and treatment is a growing field of interest for both healthcare providers and policymakers. In AI-Mind, two new AI-based medical devices are being developed to rapidly and precisely identify and assess dementia risks in people with mild cognitive impairment (MCI). Through international collaboration, AI-Mind partners develop two AI-based tools: the AI-Mind Connector and the AI-Mind Predictor. These two tools will process routinely collected data innovatively. In AI-Mind, we effectively combine various scientific approaches and product development activities, which involve state-of-the-art research, governance and scoping; implementation of results; innovation and creation of tools; delivery of prototypes; piloting and market outreach.
AI-Mind tools development
AI-based technologies are rapidly developing and have the potential to improve healthcare quality at a reduced cost. However, few examples exist of successfully deployed AI technologies in clinical practice in which effect and value were adequately assessed. To better assess the AI-Mind tools we use the early health technology assessment (HTA) framework for AI that aims to assist decision-makers in determining the value of AI-supported services. HTA provides a multidisciplinary assessment based on scientific methods and results, to promote an equitable, efficient and high-quality health system. Early HTA includes modelling of the potential cost-effectiveness of the implementation of new technologies and can give decision-makers an indication of which potential health gain and/or cost savings are associated with technology adoption.
The developed AI-Mind Connector and Predictor tools will be integrated into a cloud-based diagnostic platform, providing an easy-to-implement service for health professionals. The AI-Mind solution will be tested and validated through a research study involving 1,000 participants with MCI in five European clinical centres. Both tools will be evaluated in a clinical setting considering their software architecture, graphic display, user-friendliness and tools utility.
In AI-Mind we ensure the protection of personal data, as well as the interoperability and sustainability of the AI-Mind data, by complying with various regulations, standards and best practices, including the EU General Protection Data Regulations, supplementary national regulations for data protection and Brain Imaging Data Structure (BIDS) standard. The increasing availability of online resources means that the data need to be created with longevity in mind. Providing other researchers with access to AI-Mind data will facilitate knowledge discovery and improve research transparency. AI-Mind data will be:
- Findable – discoverable with metadata and a standard identification mechanism
- Accessible – accompanied with documentation and tools needed to access the data
- Interoperable – allowing data exchange and re-use between researchers and institutions
- Reusable – released with a copyright licence that will clarify how data can be re-used
The AI-Mind collected data include electroencephalography (EEG) and magnetoencephalography (MEG) data, digital cognitive testing, blood samples, and textual data from questionnaires and neuropsychological testing (NPT). The data is collected from participants enrolled in the AI-Mind study.
AI-Mind model is a program, algorithm or mathematical model derived from AI-Mind data using either classical machine learning (ML) or deep learning techniques (DL). ML is an artificial intelligence technique that can be used to design and train software algorithms to learn from and act on data with minimal human intervention. DL can be described as a machine learning technique that teaches computers to do what comes naturally to humans: learn by example. The AI model allows it to reach a conclusion or make a prediction when provided with sufficient information e.g., AI-Mind Predictor and Connector models fed
For the model development and integration in AI-Mind tools, we explore two AI approaches, probabilistic machine learning using brain network features and deep learning modelling based on non-processed source-reconstructed data. We evaluate these approaches through testing, bias control and assessment of predictive abilities. Thanks to that we will be able to identify the most robust AI-based diagnostic candidates for integration into the AI-Mind digital platform that combines the AI-mind Connector and the AI-Mind Predictor.