Automatic recognition of midline shift on brain CT images

Area of Science:

• Medical Imaging
• Biomechanical Modeling
• Computational Anatomy

Background:

• Midline shift is a critical indicator of brain compression severity.
• Accurate quantitative assessment of midline shift is essential for clinical decision-making.
• Existing methods may lack precision in modeling complex brain deformations.

Purpose of the Study:

• To propose and validate a novel computational method for identifying the deformed midline in brain images.
• To model brain deformation using biomechanical properties and cerebrospinal fluid spaces.
• To develop an algorithm for accurate quantitative assessment of midline shift.

Main Methods:

• Decomposition of the deformed midline into three segments: two straight (dura mater) and one curved (Bezier curve).
• Minimization of summed square differences across midline pixels to simulate bilateral symmetry.
• Application of a genetic algorithm to optimize Bezier curve control points.

Main Results:

• The novel algorithm successfully identified deformed midlines in 65 out of 81 (80%) pathological brain images.
• The method achieved a high accuracy of 95% in recognizing midline shift.
• The algorithm demonstrated robustness in evaluating images from a diverse patient cohort.

Conclusions:

• The proposed method offers a reliable and accurate approach for quantitative midline shift assessment.
• This technique can serve as a valuable tool to support clinical decision-making in neuroimaging.
• Further validation in larger, multi-center studies could enhance its clinical adoption.