Automated Critical Test Findings Identification and Online Notification System Using Artificial Intelligence in...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Non-contrast-enhanced head CT is crucial for diagnosing critical conditions.
• Deep learning algorithms are increasingly used in medical image analysis.
• Accurate and timely detection of HMH and SAI is vital for patient outcomes.

Purpose of the Study:

• To evaluate an AI tool's performance in detecting hemorrhage, mass effect, or hydrocephalus (HMH) on head CT.
• To assess the AI algorithm's capability in identifying suspected acute infarct (SAI).

Main Methods:

• A retrospective study utilized a deep learning (convolutional neural network) model.
• The AI was trained and validated on 2583 head CT images.
• Performance was tested on separate datasets for HMH and SAI detection using brain and stroke windows.

Main Results:

• For HMH, AI achieved 90% sensitivity and 85% specificity (AUC 0.91) with the brain window.
• For SAI, the best performance (stroke window) was 62% sensitivity and 96% specificity (AUC 0.81).
• AI demonstrated strong performance for HMH, with reasonable, though lower, sensitivity for SAI detection.

Conclusions:

• Deep learning AI shows potential for detecting critical findings on non-contrast head CT.
• A dedicated algorithm is necessary for SAI detection, which showed lower sensitivity than HMH detection.
• Further prospective studies are warranted to validate AI's role in screening head CT examinations.