Accurate Diagnosis: How Artificial Intelligence Helps Differentiate Types of Dementia

Dementia includes diseases that affect memory, behavior, and thinking, with Alzheimer's disease and frontotemporal dementia being the most common. Because these diseases present similar symptoms, accurate diagnosis is difficult. This study showed that brain electrical activity, measured by a simple and inexpensive test, can be analyzed with artificial intelligence to identify the type of dementia and its severity with high precision. This could help doctors diagnose earlier and better monitor the disease's progression.

Dementia is a group of neurological diseases characterized by the progressive loss of cognitive, emotional, and social functions, caused by changes in brain function. These changes affect skills such as memory, language, attention, decision-making, behavior, and emotional control, directly interfering with people's autonomy.

It is estimated that more than 50 million individuals currently live with some type of dementia, and this number could almost triple by 2050 due to the aging of the world's population.

The brain areas most frequently affected include the hippocampus, responsible for memory, the frontal cortex, linked to behavior and emotional control, and the temporal cortex, essential for language and recognition. Among the various types of dementia, Alzheimer's disease is the most common, followed by frontotemporal dementia, which usually causes early changes in personality, language, and social behavior.

Alzheimer's disease is primarily characterized by the abnormal accumulation of proteins in the brain, leading to the progressive death of neurons and the loss of connections between them. The disease progresses gradually, beginning with memory lapses and advancing to severe communication difficulties and total dependence. Frontotemporal dementia, on the other hand, mainly affects the frontal and temporal lobes of the brain, regions responsible for regulating behavior, emotions, and language.

Because of the overlapping symptoms between these two diseases, such as cognitive changes and communication difficulties, frontotemporal dementia is frequently confused with Alzheimer's disease, which can delay correct diagnosis and appropriate treatment. Differentiating these conditions is clinically essential, as their progression, management, and impact on the lives of patients and families are distinct.

To aid in the diagnosis of these diseases, various brain imaging techniques are used, such as magnetic resonance imaging (MRI) and positron emission tomography (PET), which allow visualization of structural and metabolic changes in the brain. However, these methods are expensive, complex, and not always accessible.

A promising alternative is electroencephalography (EEG), a non-invasive test that records the brain's electrical activity using sensors placed on the scalp. This test captures electrical oscillations produced by neurons, which reflect the functioning of brain networks. These oscillations are traditionally analyzed in different frequency ranges, each associated with specific mental states and cognitive functions, such as attention, sleep, memory, and emotional processing.

However, the analysis of these brain signals is complex. The brain's electrical signal is noisy, highly variable between individuals, and changes over time. To deal with this complexity, researchers use mathematical methods that transform the raw signal into more organized information, allowing them to observe how different frequencies are distributed over time and between brain regions.

These methods help identify patterns that may indicate neurological changes associated with dementia. Furthermore, statistical characteristics and measures of brain signal complexity can provide clues about the degree of organization or disorganization of neural networks.

Traditionally, artificial intelligence has been used to analyze this brain data. These algorithms learn to differentiate signal patterns associated with healthy individuals and patients with dementia, based on characteristics previously selected by researchers. Although these methods are efficient and relatively easy to interpret, they rely heavily on the manual selection of characteristics and may miss important information present in the raw brain data.

In particular, neural networks specialized in analyzing spatial images and signals can capture patterns related to the organization of brain regions, while others are designed to handle temporal sequences, allowing them to track how brain activity evolves over time.

In this study, researchers proposed an advanced model that combines different types of neural networks to simultaneously analyze the spatial structure and temporal dynamics of brain activity measured by electroencephalography.

The model was designed to perform two tasks at the same time: identify whether the person has Alzheimer's disease, frontotemporal dementia, or normal cognitive functioning, and estimate the severity of cognitive impairment. This integrated approach is important because it allows not only for diagnosing the disease, but also for assessing the stage it is in, which is crucial information for clinical decisions and care planning.

To make the results more reliable and useful for physicians, the researchers also incorporated model interpretation techniques. These techniques allow visualization of which brain regions and which patterns of electrical activity most influenced the algorithm's decisions.

In this way, it is possible to verify if the model is identifying brain signals already known to neurology, such as alterations in certain frequencies in the frontal and central regions of the brain, in addition to revealing possible new biomarkers that are still little explored.

The results showed that the model achieved high accuracy in differentiating individuals with Alzheimer's disease, frontotemporal dementia, and cognitively healthy people. Furthermore, the system was able to predict the severity of the disease with relatively low errors, especially in the case of frontotemporal dementia.

Although distinguishing between Alzheimer's and frontotemporal dementia remains challenging due to similarities in brain patterns, the use of specific feature selection strategies has significantly improved the model's ability to differentiate between these two conditions.

Based on this, the researchers proposed a step-by-step approach, in which the individual is first identified as healthy and then differentiated between the two types of dementia, achieving relevant clinical performance.

These findings indicate that combining electroencephalography with deep learning may represent an accessible, efficient, and interpretable tool for the diagnosis and monitoring of neurodegenerative diseases. By reducing costs and increasing diagnostic accuracy, this type of approach has the potential to be incorporated into clinical practice, especially in settings with limited access to advanced imaging exams.

READ MORE:

Extraction and interpretation of EEG features for diagnosis and severity prediction of Alzheimer’s Disease and Frontotemporal dementia using deep learning

Tuan Vo, Ali K. Ibrahim, Hanqi Zhuang, and Chiron Bang

Biomedical Signal Processing and Control. Volume 112, Part C, February 2026, 108667

https://doi.org/10.1016/j.bspc.2025.108667

Abstract:

Alzheimer’s Disease (AD) is the most common form of dementia, characterized by progressive cognitive decline and memory loss. Frontotemporal dementia (FTD), the second most common form of dementia, affects the frontal and temporal lobes, causing changes in personality, behavior, and language. Due to overlapping symptoms, FTD is often misdiagnosed as AD. Although electroencephalography (EEG) is portable, non-invasive, and cost-effective, its diagnostic potential for AD and FTD is limited by the similarities between the two diseases. To address this, we introduce an EEG-based feature extraction method to identify and predict the severity of AD and FTD using deep learning. Key findings include increased delta band activities in the frontal and central regions as biomarkers. By extracting temporal and spectral features from EEG signals, our model combines a Convolutional Neural Network with an attention-based Long Short-Term Memory (aLSTM) network, achieving over 90% accuracy in distinguishing AD and FTD from cognitively normal (CN) individuals. It also predicts severity with relative errors of less than 35% for AD and approximately 15.5% for FTD. Differentiating FTD from AD remains challenging due to shared characteristics. However, applying a feature selection procedure improves the specificity in separating AD from FTD, increasing it from 26% to 65%. Building on this, we developed a two-stage approach to classify AD, CN, and FTD simultaneously. In this approach, CN is identified first, followed by the differentiation of FTD from AD. This method achieves an overall accuracy of 84% in classifying AD, CN, and FTD.