Acceptance of AI-based diagnostic tools in neuroimaging

Abstract

Artificially intelligent (AI) methods are employed in many areas of medical sciences, while they are still under-utilized in functional neuroimaging methods like magnetoencephalography (MEG). This study was performed to obtain prerequisite information of the acceptance and potential adoption of AI-based analysis methods among prominent MEG clinicians and research scientists.

The study was conducted as a technology case study focusing on MEGIN Oy (Espoo, Finland) who is the global leader for MEG technology. In the future products, AI-based methods and cloud computing are strategically important for widening the clinical applications of MEG in larger patient populations and expanding the MEG market.

Semi-structured interviews were conducted with MEG users in three different hospitals in the US and three in Europe. The interviews directly probed the opinions and perspectives of the clinicians and researchers of the AI methods on MEG. The study utilized the Technology Acceptance Model and Unified Theory of Acceptance and Use of Technology frameworks, both to formulate interview questions and to account for the factors that emerged from the interview data.

All interviewees showed very positive attitude towards automated and AI-based data processing methods in MEG. They also want to widen the applicability of MEG in larger patient populations. Time-efficiency of AI methods was considered the biggest advantage, along with multi-dimensionality of the interpretation. Therefore, the AI tools could advance learning of the development phases of brain disorders.

Opinions on the transparency of algorithms varied, but all interviewees agreed that the validation process should have maximal transparency. New AI tools should be developed considering multiple empirical evidence and AI model training with data specific to the brain disorders. The clinical experts need to know what data is put in the tool, what data has been used in the tool validation, and how the tool’s accuracy and reliability has been proven. They also want visualization methods to assess the quality of the data and the results.

Besides user acceptance, factors were discussed that in general explain the lagging adoption of AI. They include limited access to patient data, questions of data security and ownership, and regulatory barriers.