Scalp surface estimation and head registration using sparse sampling and 3D statistical models.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Computer Vision

Background:

• Accurate head registration and scalp surface estimation are crucial for neuronavigation in brain stimulation and recording.
• Current neuronavigation systems often require manual fiducial targeting and patient-specific MRI scans, limiting widespread adoption.

Purpose of the Study:

• To develop a practical technique for accurate head registration and scalp surface estimation without manual landmark annotation or individual MRI scans.
• To improve the accessibility and efficiency of neuronavigation procedures.

Main Methods:

• Estimating scalp shape from surface samples acquired using standard pointer tools.
• Utilizing statistical head model priors for accurate registration.
• Leveraging object class priors for non-trivial shape acquisition from limited data.

Main Results:

• Achieved an average reconstruction root-mean-square error of 2.95 mm in a virtual study with 1152 subjects, outperforming a common neuronavigation technique by 2.70 mm.
• Demonstrated an average root-mean-square error of 2.89 mm on 50 subjects using conventional tools, improving to 2.63 mm with landmark-based registration.
• Characterized error under various conditions and provided guidelines for efficient sampling.

Conclusions:

• The proposed method offers accurate head registration and scalp surface estimation without requiring manual landmarks or patient-specific MRIs.
• The technique shows broad applicability and enhances the accuracy of neuronavigation systems.
• Results support the method's effectiveness in both simulation and experimental settings.