AI Transforms Routine Brain MRIs into Potential Stroke Predictors Researchers have developed a promising new method for identifying a common but often elusive precursor of many strokes using artificial intelligence (AI) to analyze routine brain magnetic resonance imaging (MRI) scans. In a study published in the journal Cerebrovascular Diseases, scientists from the Royal Melbourne Hospital detailed a deep learning model capable of detecting subtle signs of atrial fibrillation (AFib) in brain MRIs—signs that are typically invisible to the human eye. "Atrial fibrillation is a significant contributor to the 12 million new strokes that occur globally each year," explained Professor Bernard Yan, a neurologist at the Royal Melbourne Hospital and one of the study's lead authors. "An efficient method to identify AFib could markedly improve our ability to prevent these strokes." The deep learning algorithm used in this study is a 3D convolutional neural network (ConvNeXt). The researchers trained this network on MRI brain scans from 235 patients who had experienced strokes, either due to AFib or other causes. Once trained, the AI system demonstrated an impressive 84% accuracy in distinguishing between strokes caused by AFib and those resulting from other conditions. To develop the neural network, the team utilized NVIDIA A100 TensorCore GPUs along with CUDA 12.1, cuDNN, and NVIDIA Apex for mixed precision training. These advanced tools enabled the researchers to process and analyze large volumes of complex data efficiently. AFib is a widespread condition that leads to ischaemic strokes, which occur when blood vessels in the brain become blocked, depriving the brain of oxygen. This type of stroke accounts for nearly 90% of all stroke cases. Risk factors for AFib include advanced age, obesity, smoking, European ancestry, and high blood pressure. In the United States alone, approximately 5% of the population—around 11 million people—live with AFib. Each year, about 16,000 Americans die from AFib-related strokes. Despite this, AFib often goes undiagnosed in patients who have already undergone brain MRIs, primarily because the signs are so subtle and challenging to detect. Patients with AFib are particularly vulnerable, accounting for up to one-third of all ischaemic strokes. They are also five times more likely to experience a stroke compared to the general population. The study's authors emphasize the need for further validation of their findings. They plan to test the model with a larger dataset and seek external verification to ensure its reliability and generalizability. If these efforts prove successful, the AI-based method could become a valuable tool in predicting and preventing AFib-related strokes. One of the key advantages of this approach is its potential to reduce the invasiveness and cost associated with current diagnostic methods. Traditional methods such as electrocardiograms (ECGs) and cardiac monitoring, while effective, can be time-consuming and resource-intensive. In contrast, point-of-care MRI analysis could offer a simpler, faster, and more affordable alternative. This advancement holds the promise of transforming stroke prevention, making it easier and less costly to identify at-risk individuals. By leveraging the power of AI, healthcare providers could take significant steps toward reducing the incidence and severity of AFib-related strokes. For more details on this groundbreaking research, check out additional coverage and the full research paper.