On 20 May, researchers published an article about a tool to improve the diagnosis of Alzheimer’s disease (AD) in the journal PLOS ONE. While new guidelines recommend the use of a brain imaging technique called amyloid-PET to help diagnose AD, patients find these scans burdensome, they are expensive and not widely available. Identifying patients that would benefit from a scan is therefore valuable. Bringing together expertise from patient care as well as Artificial Intelligence (AI), the team developed a new method to improve the diagnosis of AD. The computerised decision support approach used machine learning to classify 286 people based on neuropsychology, APOE, and MRI. These were participants from the Amsterdam Dementia Cohort, consisting of 286 participants of comprises of 135 healthy participants, 108 people living with AD due to dementia, 33 people with frontotemporal dementia as well as 10 people living with vascular dementia.
For uncertain cases, hypothetical amyloid-PET results were added to increase diagnostic certainty. If certainty improved, the actual PET result was used. Results showed that the computerised method recommended amyloid-PET for 21% of patients, achieving diagnostic certainty for 66%. This approach was better than three comparison scenarios: no amyloid-PET (54%), amyloid-PET according to appropriate use criteria (55%), and amyloid-PET for all (54%). This method selected 21% of memory clinic patients for amyloid-PET without compromising diagnostic performance, and promoted a cost-effective implementation. In addition, it aids clinicians in decision-making during evaluations.