AI System Detects Alzheimer's Early Without Brain Scans

As Alzheimer's disease affects 55 million people worldwide—a number projected to rise to 139 million by 2050—accurate early detection remains critical, especially in regions with limited access to advanced medical tools. A new AI system called BRAINS offers a promising solution by using artificial intelligence to analyze patient data and identify Alzheimer's risk without requiring expensive brain scans or specialized clinical expertise.

The researchers developed BRAINS to address the global challenge of underdiagnosis in Alzheimer's disease, which accounts for up to 70% of dementia cases. Traditional diagnostic methods like neuropsychological testing and MRI-based assessments are resource-intensive and require specialized interpretation, creating barriers in economically disadvantaged regions. The BRAINS system achieves 77.3% accuracy in classifying Alzheimer's severity, significantly outperforming standard AI models that achieve only 45.4% accuracy.

BRAINS employs a dual-module architecture that combines diagnostic reasoning with case-based retrieval. The system uses large language models (LLMs) pre-trained on Alzheimer's-related medical literature, including clinical guidelines, neurocognitive assessments, and research papers. When analyzing a new patient case, BRAINS retrieves similar historical cases from its database of 1,105 patient records containing demographic information, cognitive scores like MMSE and CDR, and neuroimaging-derived metrics such as normalized whole brain volume and hippocampal volume.

The system's case retrieval module uses cosine similarity to find the most clinically relevant historical cases, which are then integrated into the diagnostic process through a cross-attention mechanism. This allows BRAINS to incorporate population-level trends and biomarker patterns into its assessments. The diagnostic module processes this combined information through prompt-based inference, where the system acts as a professional neurologist specializing in Alzheimer's disease.

Experimental results demonstrate BRAINS' effectiveness across different diagnostic scenarios. In single-clue classification tasks, the system achieves 60.0% accuracy, rising to 71.2% with retrieval augmentation. For multi-label classification involving complex cases with co-occurring conditions, BRAINS reaches 77.3% accuracy with high recall of 98.0%, though it shows lower precision of 59.1% in these complex scenarios. The system particularly excels at early-stage detection, providing explainable outputs that align with clinical insights.

This technology matters because it offers a scalable solution for Alzheimer's screening in both high-resource hospitals and underserved settings. By relying on patient data rather than specialized imaging equipment, BRAINS could help address diagnostic disparities and enable earlier intervention when treatment is most effective. The system's ability to work with heterogeneous data—including speech patterns, daily activity logs, and structural brain summaries—makes it adaptable to various healthcare environments.

The research acknowledges limitations, including the system's conservative decision boundaries in complex neurological cases and challenges with context window sizes when integrating multiple retrieved cases. Future work will need to address these constraints while expanding the system's capabilities across broader neurological domains.