Ultrafast MRI using deep learning echoplanar imaging for a comprehensive assessment of acute ischemic stroke

Area of Science:

• Radiology and Medical Imaging
• Neurology
• Artificial Intelligence in Medicine

Background:

• Acute ischemic stroke (AIS) necessitates rapid diagnostic imaging.
• Conventional MRI protocols for AIS can be time-consuming.
• Artificial intelligence (AI) offers potential for accelerating medical imaging.

Purpose of the Study:

• To evaluate an AI-enhanced ultrafast (UF) MRI protocol for AIS management.
• To compare the diagnostic performance of the UF protocol against a standard reference protocol.
• To assess the UF protocol's efficacy in detecting AIS and differential diagnoses.

Main Methods:

• A 3-T MRI protocol was used, comparing a 7-min 54-sec reference sequence (T2-FLAIR, DWI, SWI) with a 1-min 54-sec AI-enhanced UF sequence.
• Two blinded neuroradiologists assessed DWI, FLAIR, and T2*/SWI sequences for lesion detection, characterization, and presence of hemorrhage or thrombus.
• Diagnostic agreement between the UF and reference protocols was quantified using kappa coefficients.

Main Results:

• Excellent agreement (κ=0.98) was observed between UF and reference protocols for AIS detection and distribution on DWI.
• High agreement (κ=0.93) was found for detecting vascular hyperintensities on FLAIR.
• Substantial agreement (κ=0.64) for thrombus detection and fair agreement (κ=0.38) for hemorrhagic transformation on T2*/SWI were noted; differential diagnoses showed perfect agreement (κ=1).

Conclusions:

• The AI-enhanced UF MRI protocol enables effective detection and characterization of AIS.
• The UF protocol's performance in identifying stroke features is equivalent to reference sequences.
• Differential diagnoses are detected similarly by both UF and reference MRI protocols.