By [MIT News Staff]
Stroke is one of the leading causes of disability worldwide, and a rapid and precise assessment of brain damage is the key to improving recovery outcomes. A new study published in IEEE describes an artificial intelligence (AI) approach that could give clinicians a more complete picture of brain changes after a stroke and pinpoint critical areas for interventions. Despite the clinical implications, it is not expected to be deployed tomorrow. This work is interesting because it merges computational neuroscience approaches and artificial intelligence.
The researchers Wojciech Ciezobka, Joan Falcó-Roget, Cemal Koba, lead by Dr. Alessandro Crimi have developed a fully automated system that processes brain scans, maps how different regions influence each other – a property known as effective connectivity – and then uses interpretable AI models to predict outcomes. The researchers are from Poland, but the data were acquired at Saint Louis Hospital (Washington University).
From scans to networksThe system starts with MRI scans of stroke patients and estimates how activity in one brain area drives activity in another. These causal relationships between regions, the researchers say, provide richer diagnostic information than static images alone. Here is the first thing that captured our attention: causality was discovered by a very interesting approach called reservoir computing, a computational framework that utilizes a fixed, randomly connected network (the “reservoir”) to process temporal data.
Transparent AI for critical decisionsMedical AI tools have often faced skepticism because many operate as ‘black boxes’, making it difficult for clinicians to trust them in critical care settings. The team approach builds transparency into the pipeline, offering both predictive power and a clear explanation of why the model reached its conclusion, this tool is called LIME (Local Interpretable Model-agnostic Explanations).
Looking aheadWe foresee in the future more and more approaches that merge computational neuroscience and artificial intelligence.
For more information, see the original article: “End-to-End Stroke Imaging Analysis Using Effective Connectivity and Interpretable Artificial Intelligence” https://ieeexplore.ieee.org/document/10839398
Releted topics:
- Computer science and technology
- Artificial intelligence
- Machine learning
- Algorithms
- Data
- Electrical engineering and computer science (EECS)
- School of Engineering
- Innovation and Entrepreneurship (I&E)