Harvard Medical School has developed the BrainIAC AI model, which can analyze brain MRI scans across several clinical areas at once.
The model assesses brain age, predicts dementia risk, identifies tumor mutations and predicts cancer survival. Unlike most existing AI systems, which are trained for a single specific task, BrainIAC is universal and maintains high accuracy even with a limited amount of data.
The algorithm was trained on nearly 49,000 MRI scans using self-supervised learning, an approach in which the model independently identifies patterns in unlabeled data. This makes it possible to use large volumes of medical information efficiently without labor-intensive annotation.
In tests on images covering seven clinical tasks, BrainIAC outperformed three specialized AI systems. BrainIAC reflects an important shift in medical AI. In the past, each task required a separate algorithm; now a universal tool is emerging that can 'translate' data into several diagnostic conclusions at once.
