- Researchers assessed computerized methods to differentiate seven dementia syndromes based on atrophy patterns.
- Researchers hypothesized that the AI binary classifiers and multi-syndrome classifiers could reach high accuracies in differentiating syndromes.
- The multi-syndrome classification showed promise but is not translatable to clinical settings at this time.
- AI machine learning models for binary classification achieved high prediction accuracy ranging from 71 percent to 95 percent.
A new study conducted by scientists affiliated with the Max Planck Institute for Human Cognitive and Brain Sciences in Leipzig, Germany, and the University of Leipzig Medical Center demonstrates that artificial intelligence (AI) machine learning can detect rare types of dementia using medical images.
“Dementia syndromes can be difficult to diagnose,” the researchers wrote. “We aimed at building a classifier for multiple dementia syndromes using magnetic resonance imaging (MRI).”
Worldwide, dementia and average life expectancy are increasing. By 2050, people over the age of 65 will rise to nearly 1.5 billion compared to 524 million in 2010, according to estimates from a report by the World Health Organization and the U.S. National Institute on Aging and U.S. National Institutes of Health. According to the same report, life expectancy at birth is at least 81 years in several countries. Alarmingly, the number of people with dementia is expected to nearly triple worldwide to 152.8 million by 2050 from 27.4 million in 2019, according to The Global Burden of Disease dementia study published last year in The Lancet Public Health.
Dementia is a broad term characterized by changes in the brain with common symptoms that may include psychological changes such as anxiety, depression, personality changes, agitation, paranoia, inappropriate behavior, hallucinations, as well as cognitive challenges such as memory loss, confusion, disorientation, and difficulty with thinking, communication, reasoning, problem-solving, planning, organization, coordination, or motor function.
According to Stanford Medicine, dementia is due to damage or changes in the brain. Several diseases may cause dementia. Common causes include Alzheimer’s disease, vascular dementia, Parkinson’s disease, Pick’s disease, frontotemporal dementia, and dementia with Lewy bodies. Less common causes of AD include Huntington’s disease, Leukoencephalopathies, Creutzfeldt-Jakob disease, amyotrophic lateral sclerosis (ALS), multiple sclerosis (MS), syphilis, and multiple-system atrophy, a group of degenerative brain diseases per Stanford Medicine.
The most common cause of progressive dementia is Alzheimer's disease (AD), which accounts for 60 percent to 70 percent of dementia patients, according to the World Health Organization. By 2050 an estimated 16 million people in the U.S. will have Alzheimer’s disease, according to Harvard Medical School. Roughly 5.8 million people in the U.S. are living with Alzheimer’s disease, of which two-thirds are women, according to a report by AARP and the Women’s Alzheimer’s Movement (WAM).
The neuropsychiatric symptoms associated with Alzheimer’s disease may include depression, social withdrawal, psychosis, wandering, apathy, agitation, distrust in others, disinhibition, and delusions. Short-term memory loss is a common symptom in the early stage of Alzheimer’s disease. In the later stages of the disease, people with Alzheimer's forget how to perform basic daily tasks and are eventually dependent on caregivers for survival.
The scientists used a supervised machine learning AL algorithm with robust prediction methods called a support vector machine (SVM) to classify the data. The AI was used to classify patient group versus controls (binary classification), as well as a multi-syndrome classifier with all seven diagnostic groups against each other (multiclass classification).
“To the best of our knowledge, this is one of the first studies assessing computerized methods to differentiate multiple (here seven) dementia syndromes based on atrophy patterns with MRI-derived volumetric data of the brain and SVM,” the researchers reported.
The study used data from the German Research Consortium of FTLD that included a multicenter cohort using data from 477 subjects consisting of 426 patients and 51 healthy controls. Among the patients, there were 146 with behavioral variant frontotemporal dementia (bvFTD), 72 with Alzheimer’s disease, 58 with progressive nonfluent aphasia (PNFA or nfvPPA–nonfluent/agrammatic variant of PPA), 48 with progressive supranuclear palsy (PSP), 46 with semantic variant primary progressive aphasia (svPPA), 30 with logopenic variant primary progressive aphasia (lvPPA), and 26 with corticobasal syndrome (CBS).
The researchers hypothesized that the AI binary classifiers and the multi-syndrome classifiers could reach high accuracies in differentiating syndromes. The results were promising for the binary classification, less so for the multi-syndrome classifier. The multi-syndrome model performance was over three times higher than chance level, but only at a 47 percent accuracy, according to the researchers. As a proof-of-principle, the researchers concluded that the multi-syndrome classification has promise but is not translatable to clinical settings at this time.
On the other hand, the AI machine learning models for binary classification achieved high prediction accuracy that ranged from 71 percent to 95 percent. The study suggests that using AI to differentiate between dementia syndromes and healthy controls is “ready for translation to clinical routine if validated in external prospective cohorts in the future.”
"Results suggest that automated methods applied to MR imaging data can support physicians in diagnosis of dementia syndromes,” the researchers concluded. “It is particularly relevant for orphan diseases besides frequent syndromes such as Alzheimer’s disease.”
Copyright © 2023 Cami Rosso All rights reserved.