Iterative machine learning based rotational alignment of brain 3D CT data.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Neuroimaging Analysis

Background:

• Accurate rotational alignment of brain Computed Tomography (CT) images is critical for reliable diagnostic analysis.
• Current manual alignment methods are time-consuming and prone to inter-observer variability.
• Standardized imaging positions are essential for both automated and manual interpretation of neurological data.

Purpose of the Study:

• To develop and validate a novel, unsupervised, two-step iterative approach for automatic 3D rotational alignment of brain CT images.
• To improve the efficiency and accuracy of standardizing brain CT data for diagnostic purposes.
• To reduce the time required for image alignment compared to manual expert methods.

Main Methods:

• An unsupervised Midsagittal Plane (MSP) localization method was employed to determine axial and coronal rotation angles.
• This involved detecting and pairing medially symmetrical feature points within the CT data.
• A regression convolutional neural network (CNN) was utilized to estimate the sagittal rotation angle.

Main Results:

• The proposed algorithm demonstrated a low error in estimated rotations, averaging approximately 1 degree.
• The automated alignment process was significantly faster than manual alignment by radiologists, taking around 2 minutes per case.
• Validation on a dataset of manually aligned CT scans confirmed the method's accuracy and efficiency.

Conclusions:

• The novel two-step iterative approach provides an accurate and efficient solution for automatic 3D rotational alignment of brain CT images.
• This automated method has the potential to streamline diagnostic workflows and improve the consistency of neuroimaging analysis.
• The algorithm's speed and precision offer a valuable alternative to manual image alignment by clinical experts.