Romanian researchers use math and AI to spot aneurysms, diabetes and Alzheimer’s markers

A European project led from Sibiu says it can combine approximation theory and artificial intelligence to analyze CT, MRI and OCTA images without sacrificing interpretability. The work was tested with local hospitals and published in international journals, with potential impact on faster diagnosis and better clinical support.

Why it matters: - The Ma.Me.Mo.Bi.A. project aims to turn advanced mathematics into clinical tools that help doctors detect disease earlier and interpret imaging results more clearly. - The work targets three conditions with major public-health impact: abdominal aortic aneurysms, diabetic eye disease and Alzheimer’s-related brain changes. - The project emphasizes controllable models, which could reduce the risks of opaque AI systems in medical decision-making.

What happened: - An interdisciplinary research team in Romania said it developed methods that combine fundamental equations and artificial intelligence to detect medical anomalies directly from CT, MRI and OCTA images. - The European project Ma.Me.Mo.Bi.A. — Mathematical Methods and Models for Biomedical Applications — operates under financing contract no. 760076/23.05.2023. - Prof. Dr. Gianluca Vinti coordinates the project, and Prof. Dr. Ana-Maria Acu manages execution. - The team worked with Polisano Clinic Sibiu and the Sibiu County Clinical Emergency Hospital to test the algorithms on real cases.

The details: - For abdominal aortic aneurysms, the researchers developed an almost fully automated procedure that removes calcium plaques from CT images and segments blood vessels. - The aneurysm method combines U-Net-type neural networks with Sampling Kantorovich mathematical operators. - The project says the aneurysm system performs comparably to advanced deep-learning methods while remaining transparent and controllable throughout the process. - For diabetes screening, the team used OCTA retinal images to reduce background noise and reconstruct the blood vessel network. - The diabetes algorithms measure vascular connectivity and distinguish between images from healthy people and diabetic patients with high accuracy. - For Alzheimer’s disease, the methods were applied to brain MRI to automate segmentation and volume measurement of white matter, gray matter and cerebrospinal fluid. - Those brain-image biomarkers help evaluate structural degradation linked to Alzheimer’s and other neurodegenerative diseases. - The project says the methods can support physicians without losing data control or interpretability.

Between the lines: - The project reflects a broader effort to make AI in medicine more explainable, not just more powerful. - The combination of approximation theory and neural networks suggests a hybrid approach that aims to balance precision with transparency. - The collaboration with hospitals shows the work is moving beyond theory and into applied testing. - The project also functioned as training ground for PhD students and postdoctoral researchers involved in software development and advanced research.

What’s next: - The project’s published results may support broader academic and clinical adoption of the methods. - The team’s international collaborations and conference activity suggest more cross-border research could follow. - Continued testing could determine whether the algorithms can move closer to routine clinical use.

The bottom line: - Ma.Me.Mo.Bi.A. presents a model for medical AI that pairs mathematical rigor with practical imaging analysis, aiming to help doctors detect disease earlier and more transparently.