Efficacy of MRI-based deep learning algorithm for detecting acute ischemic stroke: evaluation among diverse readers.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Neurology

Background:

• Deep learning algorithms (DLAs) show promise in medical image analysis.
• The user-dependency of DLA performance across diverse medical backgrounds requires evaluation.
• Accurate detection of acute ischemic stroke (AIS) is critical for timely treatment.

Purpose of the Study:

• To evaluate the efficacy of an MRI-based DLA in detecting AIS.
• To assess DLA performance across readers with varying levels of radiology expertise.
• To determine if DLA assistance impacts diagnostic confidence and inter-reader agreement.

Main Methods:

• Retrospective analysis of 407 MRI scans from patients with suspected AIS.
• Nine readers (radiology residents, clinicians, non-neuroradiologists) independently reviewed scans with and without DLA assistance.
• Comparison of diagnostic performance metrics (AUC, sensitivity, specificity, confidence) and inter-reader agreement.

Main Results:

• The DLA significantly improved clinicians' AUC, sensitivity, and diagnostic confidence.
• Inter-reader agreement across all reader groups was significantly enhanced by the DLA.
• The DLA did not significantly affect the performance of radiology residents or non-neuroradiologists.

Conclusions:

• An MRI-based DLA significantly boosts diagnostic performance and confidence for clinicians in detecting AIS.
• The DLA improves inter-reader agreement among diverse medical professionals.
• DLAs can be valuable tools for enhancing stroke detection, particularly for less experienced readers.