Examining Neuroradiologists’ Task-Specific Diagnostic Confidence in Reporting Acute and Chronic Stroke for Retrospectively Accelerated 0.5 T MR Images

Abstract

Accelerated MRI is key for emergency medicine situations like acute ischemic stroke (AIS). Though acceleration alters image quality, neuroradiologists may remain able to report stroke pathology. This thesis examines trade-offs between acceleration factor (R) and neuroradiologists' task-specific diagnostic confidence in reporting AIS and chronic stroke, and assesses correlations between confidence and image quality metrics (IQMs). 18 participants were scanned using 0.5 Tesla MRI. Image data were retrospectively undersampled (R = 1-7X) and reconstructed via compressed sensing. Diagnostic confidence was ranked (1-5 Likert scale) and was correlated to IQMs via non-linear regression modelling. Neuroradiologists’ confidence remained high at R = 7X (p > 0.05) for AIS but decreased at R = 3X (p < 0.05) for chronic stroke. No IQMs correlated with confidence for AIS but all correlated to various degrees with confidence for chronic stroke, suggesting IQM performance does not necessarily indicate an image’s usefulness for a specific diagnostic task.