Researchers at MBZUAI in Abu Dhabi have created an artificial intelligence system called MAGNET-AD that can predict Alzheimer's disease onset up to 20 years before a patient shows any symptoms, using advanced neural networks to identify biological patterns invisible to conventional clinical assessment. The breakthrough arrives as dementia cases are projected to reach 152 million globally by 2050, according to a landmark study published in The Lancet Public Health .

On World Health Day 2026, under the global theme "Together for health. Stand with science," the researchers at MBZUAI, the world's first university dedicated entirely to AI research, are demonstrating what AI-powered medicine looks like in practice. According to the World Health Organization (WHO), a new case of dementia arises somewhere in the world every three seconds, yet there is still no cure. Early warning, however, may be nearly as valuable as prevention itself .

What Makes Early Alzheimer's Detection So Difficult?

The challenge with Alzheimer's disease is that by the time symptoms appear, significant brain damage has already occurred. MAGNET-AD uses a spatiotemporal graph neural network, a type of machine learning model that tracks patterns across both space and time, to identify biological markers that are nearly impossible to detect through standard clinical assessment. Alongside this system, researchers developed ClinGRAD, a companion AI tool that analyzes brain MRIs, genomic data, and clinical records simultaneously to classify dementia subtypes with 98.75% accuracy .

"Early detection is everything in a disease with no cure," said Salma Hassan, a Ph.D. researcher whose team developed ClinGRAD.

Salma Hassan, Ph.D. Researcher at MBZUAI

The ClinGRAD system was peer-reviewed and published at MICCAI 2025, a top-tier medical imaging conference, and was evaluated on the ANMerge dataset, a multicentric, multimodal collection of patient data designed to demonstrate robustness across diverse populations. In a field where a dementia diagnosis can reduce life expectancy by anywhere from 3 to 30 years depending on age of onset, according to a January 2025 systematic review of more than five million patients published in the BMJ, that level of precision matters enormously .

How Are Researchers Using AI to Detect Multiple Diseases Through Eye Scans?

One of the most counterintuitive findings in modern medicine is that some of the body's most revealing signals are visible through the retina. MBZUAI researchers demonstrated at Cleveland Clinic Abu Dhabi that a simple eye scan can flag early signs of diabetes, hypertension, Alzheimer's disease, and heart disease, all non-invasively and before a patient feels anything is wrong .

In the United Arab Emirates, where the International Diabetes Federation estimates diabetes affects approximately 16% of the adult population, among the highest rates in the world, the implications for population health screening are significant. The team is also developing AI systems that combine retinal vascular imaging with ECG (electrocardiogram) data to detect early heart failure, acting as a digital second opinion that catches what might otherwise be missed .

Ways AI Models Are Breaking Language Barriers in Healthcare

• Multilingual Medical AI: BiMediX, an Arabic-English medical large language model (LLM), a type of AI trained on vast amounts of text to understand and generate human language, enables reliable medical understanding and communication across languages and has been downloaded over 140,000 times on Hugging Face, the open-source AI model repository .
• Multimodal Medical Understanding: BiMediX2 extends the system's capabilities to understand medical images such as X-rays, MRIs, and CT scans, alongside Arabic and English language support, making it possible for the AI to interpret both text and visual medical data .
• Expanded Language Coverage: The model's linguistic capabilities have been further extended to support Hindi, a language spoken by over 600 million people worldwide, supported by the MBZUAI-IIT joint research seed grant, ensuring that medical AI reaches underserved populations globally .
• Accessible Delivery Methods: By integrating with platforms such as Telegram and mobile apps, and supporting both text and voice-based interactions, the system is designed to reach users with limited health literacy in remote and underserved communities, delivering preliminary medical guidance in their own language around the clock .

The AI Arabic Doctor project, powered by the BiMediX family of medical AI models developed by Dr. Hisham Cholakkal and his team, has received multiple international recognitions, including the Meta Llama Impact Innovation Award 2024 and the NVIDIA Academic Grant 2025. The team has published its research at leading AI and medical conferences, including EMNLP and MICCAI, and has open-sourced its models, data, and code, aligning with MBZUAI's commitment to advancing AI research within the Middle East region and across the globe .

Can AI Improve Fetal Screening in Low-Resource Settings?

According to the WHO, congenital anomalies affect approximately one in every 33 babies born worldwide, an estimated six million births each year. Associate Professor Mohammad Yaqub has spent his career narrowing that gap. His ScanNav technology, the world's first regulated AI fetal anomaly scan assessment system, went from an Oxford laboratory to FDA approval to deployment across GE Healthcare's global network, and now supports the care of millions of women each year .

At MBZUAI, that work has expanded into FetalCLIP, an AI model trained on more than 210,000 ultrasound images, the largest dataset of its kind, capable of detecting fetal heart defects and delivering precise anatomical measurements faster and more accurately than was previously possible. The team has since developed MobileFetalCLIP, which delivers the same capabilities as FetalCLIP in a lightweight model designed to run on edge devices, extending its reach to low-resource settings where reliable fetal screening is most critically needed .

What Other Medical Breakthroughs Are Emerging From MBZUAI?

Beyond Alzheimer's detection and fetal screening, MBZUAI researchers are working on additional medical frontiers. In November 2025, MBZUAI and GenBio AI won the UAE Artificial Intelligence Award in the Scientific Research category for their work on an AI-Driven Digital Organism (AIDO), a large-scale simulation of human biology that spans from gene activity and protein behavior to cellular function and organ systems. The project's DNA, RNA, Protein, and Cell foundation models can predict the properties of molecules and cells, while the General Expression Transformer (GET) can predict how genes behave under specific conditions before a single laboratory experiment is conducted .

These five areas of medical AI research represent a striking illustration of how quickly artificial intelligence is moving from research papers to real-world impact. By combining open-source models available on platforms like Hugging Face with specialized medical datasets and domain expertise, MBZUAI researchers are demonstrating that the future of medicine may depend less on waiting for symptoms to appear and more on catching disease before it takes hold .