New study uses sleep monitoring and AI to detect Alzheimer's

Researchers from Universidad Carlos III de Madrid (UC3M) and Hospital Universitario Severo Ochoa in Leganés, part of the public health network of the Comunidad de Madrid, have developed a methodology that uses artificial intelligence (AI) to analyze brain electrical activity during sleep and facilitate the early diagnosis of Alzheimer's. The study, recently published in the scientific journal GeroScience, demonstrates that analyzing nocturnal brain waves using machine learning techniques makes it possible to non-invasively identify early neural alterations and classify patients into three distinct biological subgroups.

Alzheimer's is a progressive neurodegenerative disease whose clinical symptoms typically manifest 10 to 20 years after the onset of pathological processes in the brain. Currently available drugs are only effective if administered in the early stages of the disease.

Although plasma testing for p-tau217 is beginning to be incorporated into clinical practice in some Spanish hospitals, its availability is not yet uniform across the Sistema Nacional de Salud. In many cases, the clinical assessment of the patient requires supplementing this information with advanced techniques such as positron emission tomography (PET) or invasive procedures, such as a lumbar puncture to analyze the cerebrospinal fluid surrounding the brain and spinal cord. These tests are usually performed in relatively late stages when symptoms are already evident.

Research focused on sleep

This research focuses on sleep during a high-activity period in which the brain carries out cellular clearance of metabolic waste, including the beta-amyloid protein. There is a bidirectional relationship in this process, where Alzheimer's alters sleep architecture, and in turn, sleep disorders favor the mechanisms that accelerate the progression of the disease.

Recording electrical activity during sleep provides us with a window into the biological processes taking place, which, over the years, can lead to forgetfulness and the characteristic symptoms of Alzheimer's."

Arrate Muñoz-Barrutia, study's principal investigator, full professor in the Department of Neuroscience and Biomedical Sciences at UC3M

"With our research, we seek a tool that enables early diagnosis of Alzheimer's, one that is non-invasive, affordable, and capable of being extended to the majority of the population," explains Dr. Anna Michela Gaeta, a specialist in Pulmonology at Hospital Universitario Severo Ochoa.

Artificial Intelligence applied to sleep recording

The research team analyzed a unique database combining nocturnal electrical activity recordings—known as polysomnographies—with data on protein expression obtained from patients' cerebrospinal fluid. This database was created in previous research led by Dr. Gerard Piñol Ripoll, a specialist in Neurology at Hospital Universitario Santa María de Lérida-IRBLLeida, and Dr. Ferrán Barbé, head of the Pulmonology Service at Hospital Universitario Arnau de Vilanova de Lérida and director of the translational research group in respiratory medicine at IRBLleida. The study included 42 patients with mild to moderate Alzheimer's and 58 cognitively healthy controls. Thanks to funding from the Instituto de Salud Carlos III, the team in Lérida compiled all the data and built the repository subsequently analyzed by researchers in this new study.

"Signals were recorded using a series of electrodes placed on the scalp that capture the electrical activity of neurons throughout the night. We used AI to analyze this electrical activity and identify certain changes that may detect the accumulation of proteins in the brain, which will later lead to neurodegenerative diseases," explains another of the study's authors, Lorena Gallego Viñarás, from the Department of Neuroscience and Biomedical Sciences at UC3M. She further clarifies that "AI is not intended to replace any medical tests, but rather to support early diagnosis. This allows information to be analyzed in much greater detail to facilitate earlier diagnoses and, consequently, initiate treatments much sooner."

This AI-based technology has successfully differentiated healthy individuals from patients with Alzheimer's with high accuracy. Furthermore, by cross-referencing these data with key cerebrospinal fluid biomarkers—such as beta-amyloid (Aβ42), phosphorylated tau (p-tau181), total tau (t-tau), and neurofilament light chain (NfL)—the algorithm identified three distinct subgroups or subclusters of Alzheimer's patients. These profiles showed gradual differences in their biomarker levels, proving that the disease presents different biological signatures from its earliest stages.

Results with potential clinical application in the future

In the future, these tools based on nocturnal sleep studies could complement blood tests, such as those for the p-tau217 protein. This approach could provide a cost-effective screening pathway that is feasible from the patient's own home to detect preclinical Alzheimer's while simultaneously treating sleep disorders, thereby helping to slow the progression of cognitive decline.

"Another thing we have learned from this research is that the future of science in this field relies on integrating different disciplines, such as neurology, pulmonology, and engineering. This collaboration is essential because drugs are currently being developed that can only act in the early stages of the disease," concludes Dr. Anna Michela Gaeta.