NeuroAPS-Net: Lightweight AI Model for Alzheimer's Classification
Researchers have developed NeuroAPS-Net, a lightweight geometric deep learning model for Alzheimer's disease classification using structural MRI. The model converts T1-weighted MRI into anatomically informed 2D point clouds via Anatomical Priority Sampling (APS), creating the ADNI-2DPC dataset—the first neuroanatomically labeled MRI-derived point cloud dataset. NeuroAPS-Net incorporates anatomical priors through region-aware feature encoding and ROI token aggregation. Experiments show competitive classification accuracy with significantly reduced inference latency and GPU memory compared to traditional 3D CNNs, enabling deployment in resource-constrained settings.
Key facts
- NeuroAPS-Net is a lightweight geometric deep learning model for Alzheimer's classification.
- It uses Anatomical Priority Sampling (APS) to convert T1-weighted MRI into 2D point clouds.
- The ADNI-2DPC dataset is the first neuroanatomically labeled MRI-derived point cloud dataset.
- NeuroAPS-Net incorporates region-aware feature encoding and ROI token aggregation.
- The model achieves competitive accuracy with lower inference latency and GPU memory usage.
- The work addresses limitations of computationally expensive 3D CNNs.
- Alzheimer's disease is a progressive neurodegenerative disorder and major cause of dementia.
- Structural MRI is widely used to analyze AD-related brain atrophy.
Entities
—