Leveraging Spatial Information in Radiology Reports for Ischemic Stroke Phenotyping

Area of Science:

• Neurology
• Medical Informatics
• Radiology

Background:

• Accurate classification of ischemic stroke phenotypes is crucial for research and treatment.
• Radiology reports contain rich contextual information for phenotype identification.
• Location-specific details like brain region, laterality, and stroke stage are key phenotype characteristics.

Purpose of the Study:

• To develop and evaluate a method for classifying fine-grained ischemic stroke phenotypes using radiology reports.
• To leverage a spatial information extraction system (Rad-SpatialNet) and domain rules for phenotype classification.
• To demonstrate the utility of fine-grained information extraction for complex clinical phenotype determination.

Main Methods:

• Utilized an existing fine-grained spatial information extraction system, Rad-SpatialNet.
• Applied simple domain rules to the extracted information to classify stroke phenotypes.
• Focused on location-specific features: brain region, laterality, stroke stage, and lacunarity.

Main Results:

• Achieved a recall of 89.62% for classifying brain region.
• Attained a recall of 74.11% for classifying brain region, side, and stroke stage together.
• Demonstrated the feasibility of using an information extraction system with domain rules for phenotype classification.

Conclusions:

• An information extraction system with a fine-grained schema and simple domain rules can effectively classify complex ischemic stroke phenotypes.
• The identified phenotypes can aid stroke research, particularly in post-stroke outcomes and personalized treatment planning.
• This approach highlights the value of automated analysis of radiology reports for clinical phenotyping.