Researchers developed an AI model that accurately distinguishes between multiple dementia types using extensive, heterogeneous brain imaging data.
Key Details
- 1The model was trained and tested on 308,000 3D brain images from 17,000 patients collected over two decades.
- 2It detects vascular dementia, Alzheimer's, Lewy body dementia, Parkinson's, and mild cognitive impairment, with AUC >0.84 for these conditions.
- 3The dataset included multiple modalities (T1 MRI, T2 MRI, CT, PET), reflecting real-world clinical complexity and variation.
- 4The neural network is structured to handle varying numbers and types of images per patient (1–14), mitigating confounding variables like scanning site and age.
- 5Testing across multiple hospital sites demonstrated the model’s robustness to real-world heterogeneity.
- 6Future directions include larger datasets and development of explainable AI for neuroimaging disease detection.
Why It Matters
This study addresses a major barrier for clinical translation of imaging AI—performance on real-world, heterogeneous datasets—demonstrating a feasible path for deploying robust differential diagnostic tools across healthcare settings.
Source
EurekAlert
Related News
•EurekAlert
AI-Powered OCT Enables Rapid 'Optical Biopsy' for Early Endometrial Cancer Detection
A team at Washington University has developed a catheter-based 3D OCT system with AI to quickly and noninvasively detect early endometrial cancers.
•EurekAlert
AI Clinical Reasoning in Diagnostics and Digital Fatigue in Healthcare
Recent JMIR features explore large language models in clinical diagnostics and digital fatigue among healthcare professionals.
•EurekAlert
KAIST, MIT, Microsoft Develop Efficient AI Image Upsampling for Robotics
KAIST, MIT, and Microsoft have created 'Upsample Anything,' a training-free AI method to restore high-resolution visual data from compressed images with up to 16x improved GPU memory efficiency.