A study published in the medical journal The Lancet assessed the use of machine learning and clinically gathered data (such as cognitive and behavioral screening scores) to classify and discriminate FTD and Alzheimer’s disease (AD) in underrepresented groups.

Diagnosing FTD and other dementias is complicated, with FTD taking 3.6 years on average to diagnose. As the study highlights, those seeking a diagnosis in Latin American countries face additional obstacles, such as the lack of standardized diagnostic practices across countries and the diversity of the various instruments used to evaluate cognitive status. In addition, technologies and methods effective in diagnosis, such as amyloid/tau PET imaging or fluid-based biomarker assessment tools, are less accessible to practitioners in Latin America.

These obstacles to diagnosis also impede research. Differences in the sociocultural and genetic makeup of countries and variable levels of clinical expertise in FTD and AD make it challenging to collect data that can be generalized and compared to other countries.

To address the obstacles faced by practitioners in Latin America, the authors of the study present a computational framework for classifying FTD and AD that can effectively utilize diverse samples. The framework combines traditional statistical analysis with newer machine-learning models. Using this framework, the authors identified the most generalizable features of FTD and AD that can be used to discriminate between the diseases, and between people who are and aren’t living with a form of dementia.

The researchers recruited 1,792 participants from 11 centers that are members of the Multi-Partner Consortium to Expand Dementia Research in Latin America (three in both Argentina and Colombia, two in both Chile and Mexico, and one in Peru). All participants were diagnosed with FTD or AD following the standard procedures at each facility.

The authors found that Random Forest (RF), a machine-learning model that works well with complex data, was the most effective in classifying FTD and AD. Despite the demographic diversity of the participants and the samples they provided, RF was still highly effective at classifying FTD and AD, with the authors noting an efficiency rating of 93%.

Using the RF model, the authors found that data from cognitive screenings, social cognition tests, executive functioning measures, and neuropsychiatric exams showed the greatest capability to classify FTD and AD. For FTD specifically, neuropsychiatric and social cognition measures were the most useful for classifying the disease.

Machine learning is an increasingly promising tool for diagnosing and researching FTD. Click here to read about a German study that also evaluated the effectiveness of machine learning in FTD research.