In a laboratory in Elche, Valencia, two researchers have engineered a quiet revolution in how doctors see inside the brain. Silvia De Santis and Maximilian Eggl, working at the Institute for Neurosciences—a joint center of the Spanish National Research Council and Miguel Hernández University—have combined artificial intelligence with physics-based computer simulations to shrink advanced brain MRI scan times by up to 90%, while using just 10% of the data typically needed. The result, published in Communications Medicine, doesn't sacrifice detail for speed. Instead, it promises to transform neuroimaging from a bottleneck in clinical care into a genuinely accessible tool.
The innovation rests on a clever inversion of how AI is usually trained in medicine. Rather than feeding neural networks thousands of real patient scans—a process that depends on patient availability, raises privacy concerns, and can embed clinical biases—the team built their models using physics-based simulations of how water diffuses through brain tissue. These synthetic datasets, generated endlessly and without privacy constraints, proved powerful enough to teach machines to extract rich biological information from sparse, real-world scans. "Using simulations allows us to generate as much data as we need, without depending on patient availability and while avoiding privacy issues," explains Eggl, who leads the AI-inspired Biomarkers of Brain Structure and Function research line at the institute.
The practical impact is immediate and measurable. Diffusion-weighted MRI, a technique that studies water movement in brain tissue to reveal its microscopic structure, typically requires extended scanning sessions. Imagine a patient lying motionless in a scanner for roughly 40 minutes. With this new approach, the same diagnostic information emerges in roughly 8 minutes—a transformation that De Santis notes could be transformative in hospitals drowning under long waiting lists. That efficiency compounds across a health system: more patients scanned in the same time, shorter waits, fewer resources consumed.
The breakthrough hinges on a striking finding: networks trained entirely on simulations achieved high accuracy using only 10% of typical measurement data. This isn't a trade-off between speed and quality—it's a genuine gain in efficiency. "Reducing the acquisition time required makes it possible to incorporate much more advanced MRI techniques, resulting in a greater amount of clinical information available to medical staff," says De Santis, who leads the Translational Imaging Biomarkers Laboratory at the institute.
The implications extend far beyond faster scans. Neurodegenerative diseases like Alzheimer's unfold silently for up to two decades before any visible symptoms emerge, yet diagnosis still relies on techniques developed more than three decades ago. This new approach could enable earlier detection by extracting subtler signals from brain tissue—the molecular whispers that precede clinical decline. Equally promising, the simulation-based method can reanalyze MRI data acquired decades ago, breathing new diagnostic power into old scans that were limited by the technology of their time.
What began as a technical puzzle—how to squeeze rich information from minimal data—has become a pathway toward earlier diagnosis, shorter patient waits, and a bridge between laboratory discovery and clinical practice. In Elche, two researchers have shown that sometimes the fastest way forward isn't simpler scans, but smarter ones.