AI framework for differential diagnosis of dementia etiologies on multimodal data.
Deep learning framework for differential diagnosis of dementia etiologies using multimodal data.
This tool provides a deep learning framework published in Nature Medicine for diagnosing dementia etiologies. It suits clinical researchers and medical AI developers working with multimodal patient data and MRI scans.
It processes individual-level demographics, health history, neurological exams, and multi-sequence MRI scans through a transformer model. The framework classifies individuals into thirteen diagnostic categories across normal cognition, mild cognitive impairment, and ten dementia types.
It suits researchers and data scientists with access to clinical cohorts and GPU infrastructure who want to run advanced dementia classification. It is not suitable for clinicians without technical expertise or those lacking the required Python and PyTorch setup.
Combines demographics, health history, neurological testing, and MRI scans into a single transformer model.
Extracts image embeddings using a fine-tuned Swin UNETR encoder with self-supervised pre-trained weights.
Classifies patients across normal cognition, mild cognitive impairment, and ten distinct dementia types.
Provides an online hosted space to test diagnostic predictions without local installation.
The tool is free under its open-source licence, and an interactive demo is hosted on Huggingface, meaning the only financial cost is running your own hardware and compute infrastructure.
877 open remote AI-training roles, updated November 16, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
These reviewed AI tools do the same job instantly — from our scored directory.
The model supports demographics, health history, neurological testing, physical exams, and multi-sequence MRI scans.
You can use the interactive Huggingface Space demo hosted by the authors to explore the model's predictions in real time.
Yes, you can train and run the model without imaging by using the NonImg flag configuration.
See Zekai first in Google