Multimodal Fine-Grained Annotated Dataset for Ischemic Stroke Across the Full Cycle

Regarding the characteristic of multiple lesions in ischemic stroke, cranial CT or MRI imaging is used in clinical diagnosis to reveal the location and scope of the lesions, which plays a key role in subsequent symptomatic therapy and prevention of recurrence. However, precise localization of ischemic lesions requires professional neurosurgeons to detect the lesion area in a patient’s CT/MRI scans, which is time-consuming and difficult. In order to achieve automatic detection of lesions, we introduce the dataset of ischemic stroke across the full cycle containing CT and MRI image sequences, with structured annotation by means of imaging reports. The imaging report, issued and reviewed by two neurosurgeons, facilitates fine-grained annotations of multiple ischemic lesions from diagnostic descriptions and conclusions. This finely annotated dataset plays an important role in the development of cross modal deep learning models based on AI for multi-object detection. Our full cycle fine-grained annotated dataset aims to promote refined diagnosis of ischemic stroke in clinical multimodal image analysis, which is helpful for symptomatic therapy and prognosis of patients. It is worth mentioning that this dataset is the most comprehensive resource for ischemic stroke to date, and its utility is demonstrated through a baseline model.