The delineation of largely deformed brain midline using regression-based line detection network

Area of Science:

• Medical Imaging and Artificial Intelligence
• Neuroscience and Computational Anatomy

Background:

• The brain's midline is a key anatomical reference in computed tomography (CT) scans.
• Brain diseases such as hematoma and tumors can cause midline shift, a critical indicator for clinical evaluation.
• Existing methods for detecting midline shift are limited by anatomical variability and large brain deformations, lacking intuitive delineation.

Purpose of the Study:

• To develop an automated and robust method for delineating the brain's midline, particularly in cases of significant brain deformation.
• To overcome the limitations of existing landmark- or symmetry-based methods by providing intuitive midline measurements.

Main Methods:

• Proposed a novel regression-based line detection network (RLDN) for robust midline delineation.
• Treated midline delineation as a skeleton extraction task, enhanced with a multiscale bidirectional integration module for feature representation.
• Incorporated a regression task for accurate and continuous midline detection, validated on the public CQ 500 dataset and a private hospital cohort.

Main Results:

• Achieved a mean line distance error of 1.17 ± 0.72 mm and an F1-score of 0.78 on the CQ 500 test set.
• Demonstrated a mean line distance error of 4.15 ± 3.97 mm and an F1-score of 0.61 on the private dataset.
• Showed statistically significant improvements (P < 0.05) compared to existing methods on both datasets.

Conclusions:

• The developed RLDN offers a robust solution for midline delineation in severely deformed brains.
• The method achieves state-of-the-art performance, enabling potential for automated diagnosis of brain diseases.
• This automated approach facilitates clinical evaluation by providing accurate and reliable midline measurements.