Automatic detection of acute ischemic stroke using non-contrast computed tomography and two-stage deep learning model

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neurology

Background:

• Detecting acute ischemic stroke (AIS) on computed tomography (CT) images presents significant challenges.
• Advanced computational methods are needed to improve the accuracy and efficiency of AIS detection.

Purpose of the Study:

• To develop and evaluate an automated system for detecting AIS using a two-stage deep learning model.
• To assess the impact of the AI system on radiologist performance in identifying AIS.

Main Methods:

• A two-stage deep learning model (You Only Look Once v3 and Visual Geometry Group 16) was trained on 189 head CT cases.
• The model was tested on 49 independent cases, with results compared to a radiologist's assessment.
• Radiologist sensitivity and false positive rates were evaluated with and without AI assistance.

Main Results:

• The AI model achieved a sensitivity of 37.3% with 1.265 false positives per case.
• Radiologist sensitivity improved from 33.3% to 41.3% when using the AI system.
• The AI system significantly improved radiologist detection sensitivity (p=0.0313).

Conclusions:

• The developed two-stage deep learning system effectively enhances radiologist sensitivity for AIS detection.
• This AI-powered approach shows promise for improving the diagnostic workflow for acute ischemic stroke.