- Researchers have developed a machine-learning model that analyzes MRI brain scans to detect Alzheimer’s disease, achieving 92.87% accuracy in distinguishing mild cognitive impairment or Alzheimer’s.
- The model identified structural patterns associated with cognitive decline, with volume loss in specific brain regions emerging as a possible early biomarker of the disease.
- Researchers also found sex-related differences in brain changes, suggesting that biological factors, such as hormonal changes, may influence how Alzheimer’s develops.
Alzheimer’s disease is a progressive condition that can cause memory loss and cognitive decline. Detecting Alzheimer’s typically requires a comprehensive medical evaluation, which a person may only receive after presenting symptoms that could suggest a decline in brain functioning.
Alzheimer’s disease slowly worsens over time, and an early, accurate diagnosis can be beneficial for treating the progression of the disease. However, early diagnosis is often challenging because initial symptoms may resemble typical age-related changes in memory or thinking
As such, diagnostic methods to diagnose Alzheimer’s disease early, or even predict the onset of symptoms, could be critical for maximizing the effectiveness of emerging, disease-modifying treatments and manage the condition.
A new study, published in Neuroscience, highlights an artificial intelligence (AI) tool that analyzes MRI scans and identifies patterns of brain volume loss associated with Alzheimer’s disease.
Findings indicate that the model could accurately predict the condition, suggesting that machine-learning techniques may help detect the disease earlier than traditional diagnostic approaches.
To develop the predictive model, researchers from Worcester Polytechnic Institute analyzed 815 MRI scans from participants aged 69 to 84.
The scans came from the Alzheimer’s Disease Neuroimaging Initiative, a large research project that collects imaging data from people with normal cognition, mild cognitive impairment, and Alzheimer’s disease.
As Alzheimer’s disease injures neurons and leads to a loss of brain tissue, these scans could contain subtle changes that may indicate early disease development.
The team used a machine learning model to measure brain volume across 95 different regions. An algorithm then compared these measurements to identify patterns distinguishing healthy brains from those affected by cognitive impairment or Alzheimer’s disease.
From analyzing the large dataset of brain scans, the researchers also identified several brain regions where structural changes were strongly associated with the disease.
Notably, volume loss in the hippocampus, amygdala, and entorhinal cortex were among the strongest indicators of Alzheimer’s disease across age and sex groups.
The hippocampus plays a key role in memory and learning, the amygdala regulates emotions, and the entorhinal cortex is involved in memory, navigation, and perception, and among the first parts of the brain to be affected by Alzheimer’s disease.
Interestingly, researchers also found that individuals aged 69 to 76, the youngest group studied, commonly showed volume loss in the right hippocampus, suggesting this region may serve as an early biomarker for the disease.
Medical News Today spoke with Dung Trinh, MD, internist for the MemorialCare Medical Group and chief medical officer of the Healthy Brain Clinic in Irvine, CA, about the possible role of the right hippocampus.
“The paper points to the hippocampus as one of the earliest and most consistently structures in Alzheimer’s affecting memory, with rapid tissue loss occurring early in the disease process,” Trinh told us.
“In this dataset, the 69 to 76 age group showed substantial right hippocampal volume decreases, which likely means that this region was sensitive to subtle early-stage neuro degeneration before more widespread cortical changes became dominant,” he detailed.
“I would frame it as a promising signal rather than a definitive standalone biomarker because the study is still based on one cohort and internal validation only,” noted Trinh.
Additionally, the study also uncovered some differences in how Alzheimer’s may affect male and female brains.
In female brain scans, volume loss was more prominent in the left middle temporal cortex. This is a region associated with language and visual processing.
However, in male brain scans, changed were more pronounced in the right entorhinal cortex.
The researchers propose that these differences may relate to hormonal changes linked with aging, such as declines in estrogen and testosterone, which have been previously associated with Alzheimer’s risk in females and males.
Trinh noted that he found the sex-specific asymmetries interesting and agrees with the authors assumption as a possible reason:
“The authors discuss a biologically credible framework involving hormonal change, especially reduced estradiol after menopause, genetic risk such as [the] APOE-e4 [genetic variant], and neuroinflammatory processes interacting with amyloid and tau pathology. Those factors though were not directly measured in this study, so they should be viewed as possible explanations rather than proven causes.”
The research team plans to continue refining their predictive models using more advanced deep-learning approaches.
Trinh cautioned that while the study shows promise, further validation is still necessary:
“AI-based imaging can detect multiregional structural patterns that may be hard to appreciate by eye, and this study suggests those patterns may emerge across the transition from cognitively normal to mild cognitive impairment to Alzheimer’s. If future validation occurs, it could help clinicians identify higher-risk patients earlier, monitor progression more closely, and eventually tailor treatment plans around an individual’s neuroanatomical profile.”
“In practice, that could mean earlier intervention, better patient selection for disease-modifying therapies, and closer monitoring of those most likely to decline. But I would stress that this paper shows promise, not clinical readiness,” he noted.
“It would help to combine MRI with other biomarkers — for example amyloid, tau, blood-based biomarkers, genetics, and longitudinal follow-up — to show whether the model predicts real-world progression, not just classification within one dataset,” added Trinh.
The research team also aim to investigate other factors that may influence Alzheimer’s development, including conditions such as diabetes.
If validated in larger populations, the research team suggests that AI-based tools could eventually help clinicians identify individuals at risk for Alzheimer’s disease earlier, improving both diagnosis and the ability to test new therapies.
