AI system identifies early signs of Alzheimer's with high precision, but still requires clinical validation

Algorithm detects early brain patterns with promising accuracy

An artificial intelligence system developed by researchers at the University of California in San Francisco (UCSF) demonstrated the ability to identify brain alterations associated with Alzheimer's up to six years before the traditional clinical diagnosis, according to a study published in December 2023 in the journal Nature Medicine (DOI: 10.1038/s41591-023-02740-w).

The deep learning algorithm, led by Dr. Jenna Erickson and the team of the Department of Radiology of UCSF analyzed magnetic resonance imaging (MRI) reaching 90% sensitivity and 85% specificity in the identification of patients who would develop Alzheimer's in the following years.

"Our model identified subtle patterns of hippocampal atrophy and metabolic changes that precede clinical symptoms," said Dr. Erickson in a press release. "This could represent a critical window for therapeutic interventions."

Study methodology and limitations

The system was trained with 15,000 MRI images from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database collected over 15 years. The researchers retrospectively followed patients who eventually developed Alzheimer's, teaching the algorithm to recognize early biomarkers.

The study has important limitations. Validation occurred in a predominantly Caucasian population of high educational level, raising questions about generalization to other demographics. In addition, the false positive rate of 15% means that approximately one in seven patients would receive an incorrect warning.

Dr. Ricardo Nitrini, full professor of Neurology at USP and specialist in dementia, offers a cautious perspective: "It is a significant methodological advance, but we need to understand that early detection without proven disease-modifying treatments raises ethical dilemmas. We are informing patients about a future that we cannot yet substantially change."

Distance between research and clinical application

Despite promising results in the research environment, the path to clinical implementation is long. Technology will need to undergo prospective multicenter clinical trials, a process that typically takes 5 to 10 years. Subsequently, it will be necessary to approve regulatory agencies such as FDA (United States) and Anvisa (Brazil).

" We are starting conversations with the FDA about the design of prospective studies," explained Dr. Erickson. "We need to demonstrate not only diagnostic accuracy, but also that early knowledge leads to better clinical outcomes for patients."

The cost and accessibility represent additional barriers. High-resolution brain MRIs required for the algorithm are not widely available, especially in regions with limited resources.

Ethical implications of pre-symptomatic diagnosis

The possibility of diagnosis years before symptoms raises complex ethical issues. An editorial accompanying the study in Nature Medicine, signed by Dr. Jason Karlawish of the University of Pennsylvania, highlights concerns about anxiety, discrimination in health insurance and stigmatization.

" We must question whether we are prepared as a society to deal with predictive diagnoses of neurodegenerative diseases," wrote Karlawish. "Knowledge without effective therapeutic options can cause more harm than benefit."

Recent approvals of medicines such as lecanemab and donanemab offer hope that early interventions may slow progression, but these treatments have modest efficacy and significant side effects.

Context and future prospects

With over 55 million people living with dementia globally — a number designed to triple by 2050 according to WHO — early diagnostic tools are urgently needed. However, experts emphasize that they need to be accompanied by effective therapies.

The UCSF team plans to expand the study to more diverse populations and integrate other biomarkers, including cerebrospinal fluid and PET scans of beta-amyloid and tau proteins.

" This is an important step, but it is just that — a step," concludes Dr. Nitrini. "The journey from the research bench to real clinical benefit for patients still has many obstacles ahead. Enthusiasm must be tempered with scientific realism."

The authors of the study reported no conflicts of financial interest. The research was funded by the National Institute on Aging of the United States.