To derive meaningful and reproducible models of brain functionality from stroke images, catering to both clinical and research objectives, presents a formidable challenge due to the substantial variation in lesion frequency and patterns.
Consequently, the necessity for extensive datasets becomes paramount, along with the imperative of employing fully automated image post-processing tools for comprehensive analysis. The progression of such tools, particularly leveraging artificial intelligence, crucially hinges on the availability of expansive datasets for effective model training and testing. In this context, we introduce a publicly accessible dataset encompassing 2,888 multimodal clinical MRIs of patients dealing with acute and early subacute strokes. This dataset includes meticulous manual lesion segmentation as well as accompanying metadata. Offering a wealth of high-quality, extensively supervised human knowledge, this dataset becomes a pivotal resource for fueling artificial intelligence models and driving the ongoing evolution of tools aimed at automating diverse tasks that currently rely on human intervention. These tasks encompass lesion segmentation, labeling, computation of disease-relevant scores, and exploration of lesion-based studies that establish correlations between functionality and frequency-based lesion mapping.
Source Nature
