Automated Midline Shift Detection in Head CT Using Localization and Symmetry Techniques Based on User-Selected Slice.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Radiology
• Neurological Pathology Detection

Background:

• Midline shift (MLS) is a critical intracranial displacement, often caused by space-occupying lesions or traumatic brain injuries.
• Timely detection of MLS is vital for patient outcomes, as delays in diagnosis and intervention can be detrimental.
• Existing methods face challenges in assessing MLS severity, especially when the septum pellucidum is effaced.

Purpose of the Study:

• To develop a deep learning algorithm capable of detecting midline shift (MLS) across its full severity spectrum, from mild to severe.
• To address the limitation of traditional methods that rely on the septum pellucidum, which may be effaced in severe MLS cases.
• To create an AI tool for rapid and accurate identification of MLS in patient CT scans.

Main Methods:

• Utilized a cohort of 981 patient CT scans with diverse cerebral pathologies.
• Selected individual CT slices, prioritizing those with visible lateral ventricles.
• Trained a You Only Look Once (YOLO) object detection model on 400 annotated scans to identify lateral ventricles and the skull-axis midline.

Main Results:

• The AI model achieved an area under the curve (AUC) of 0.79 for MLS detection.
• Demonstrated a sensitivity of 0.73 and specificity of 0.72 in differentiating MLS severity.
• The model effectively captured moderate and severe MLS while distinguishing them from mild and normal cases.

Conclusions:

• An AI model was successfully developed to reliably identify the lateral ventricles and cerebral midline in CT scans.
• The model accurately identifies and stratifies clinically significant MLS, distinguishing emergent from non-emergent cases.
• This AI tool has the potential to serve as a foundation for integrated clinical systems to expedite the review of urgent cases.