Researchers Unveil New AI and Brain Imaging Insights into Neurodegenerative Diseases

In a groundbreaking study, researchers from Massachusetts General Hospital have employed advanced artificial intelligence (AI) techniques to analyze two decades' worth of retrospective 3D brain imaging data, encompassing roughly 308,000 images from 17,000 patients. This extensive dataset includes various neuroimaging modalities such as MRI, PET, and CT, reflecting heterogeneity common in real-world clinical settings. The goal was to develop an AI model capable of detecting multiple neurodegenerative disorders—such as dementia, Alzheimer's disease, Lewy body dementia, and mild cognitive impairment—with high accuracy.

The key challenge addressed by the team was the complexity and inconsistency of real-world medical imaging data, often plagued by biases related to age, scanner type, and other confounding factors. Inspired by language models, the researchers created a neural network that could accept variable numbers of images and was designed to focus on structural brain features indicative of disease, rather than extraneous patient or technical factors. The model was trained to identify disease markers based on brain structure size and asymmetry, especially in subcortical regions, across data from multiple hospitals, demonstrating robust performance even when tested outside its training environment.

The findings showed that their AI model achieved an area under the curve (AUC) exceeding 0.84 for several conditions, indicating strong discriminative ability. While it was less effective for certain disorders like multiple sclerosis and epilepsy, its performance in detecting dementia-related disorders across diverse hospital sources suggests significant potential for clinical integration. This advancement points toward AI tools that could help clinicians diagnose neurodegenerative diseases earlier and more accurately, especially in heterogeneous community healthcare settings.

The study emphasizes the importance of developing explainable AI models and expanding datasets to improve diagnostic reliability. Future research aims to enhance transparency and apply these models for prognosis and treatment outcome prediction, moving closer to precision medicine for neurodegenerative conditions.

Published in Alzheimer’s & Dementia, the research was led by Matthew Leming, Ph.D., and Hyungsoon Im, Ph.D., aiming to transform how we detect and monitor neurological disorders in clinical practice. This work could pave the way for wider adoption of AI-driven diagnostics in healthcare environments where data diversity and complexity are major hurdles.

Source: https://medicalxpress.com/news/2025-07-qa-discuss-insights-neurodegeneration-ai.html