Midsagittal plane extraction from brain images based on 3D SIFT.

Area of Science:

• Neuroimaging
• Computer Vision
• Medical Image Analysis

Background:

• Midsagittal plane (MSP) extraction is crucial for human brain symmetry analysis.
• Existing methods often rely on gray similarity or surface matching, limiting accuracy and speed.
• A robust and efficient MSP extraction technique is needed for advanced neuroimaging applications.

Purpose of the Study:

• To develop a fast and robust midsagittal plane extraction method for 3D brain images.
• To improve human brain symmetry analysis using distinctive 3D Scale-Invariant Feature Transform (SIFT) features.
• To achieve real-time performance and high accuracy in MSP extraction.

Main Methods:

• Utilized 3D Scale-Invariant Feature Transform (SIFT) for distinctive feature extraction.
• Employed parallel 3D SIFT matching and iterative least-median of squares plane regression to determine the fissure plane.
• Developed a novel metric for symmetry magnitude using relative scales, orientations, and flipped descriptors of 3D SIFT features.
• Implemented k-dimensional tree (KD-tree) on GPUs for efficient feature clustering, indexing, and parallel matching.

Main Results:

• Achieved real-time performance for midsagittal plane extraction.
• Demonstrated improved accuracy compared to state-of-the-art methods.
• Reported an average yaw angle error below 0.91° and an average roll angle error no more than 0.89°.
• Validated the method on synthetic and in vivo datasets, including normal and pathological cases.

Conclusions:

• The proposed 3D SIFT-based method offers a fast, robust, and accurate solution for midsagittal plane extraction.
• This technique significantly advances human brain symmetry analysis capabilities.
• The real-time performance and high precision make it suitable for clinical and research neuroimaging.