Automatic Quantification of Computed Tomography Features in Acute Traumatic Brain Injury

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Traumatic brain injury (TBI) frequently presents with intracranial abnormalities visible on CT scans.
• Visual assessment of CT scans for TBI lacks quantitative interpretation and is subject to observer variability.
• Objective quantification of TBI lesions and secondary injuries is crucial for patient management.

Purpose of the Study:

• To present icobrain, a fully automated deep learning method for quantifying acute intracranial lesions, cistern volume, and midline shift on noncontrast CT scans.
• To evaluate the accuracy and reliability of icobrain compared to expert assessments.

Main Methods:

• Development of a deep learning algorithm (icobrain) for automated analysis of noncontrast CT images.
• Utilized a multi-center dataset from the CENTER-TBI study for validation.
• Compared icobrain's quantitative measurements (lesion volume, cistern volume, midline shift) against expert annotations.

Main Results:

• icobrain demonstrated high accuracy in quantifying acute intracranial lesions (median volume difference: 0.07 mL, correlation: 0.91) and cistern segmentation (median volume difference: -0.01 mL, correlation: 0.94).
• Midline shift computation showed a median error of -0.22 mm with a strong correlation (0.93) to expert assessments.
• The automated method provides reliable quantitative data, overcoming limitations of visual assessment.

Conclusions:

• icobrain offers a fully automated and reliable solution for quantifying key metrics in traumatic brain injury from CT scans.
• The method's high accuracy and correlation with expert assessments suggest significant value for routine clinical practice in TBI management.
• Automated image analysis tools like icobrain can enhance the objectivity and efficiency of TBI assessment.