Validation of active shape model techniques for intracochlear anatomy segmentation in computed tomography images

Area of Science:

• Medical Imaging
• Computational Anatomy
• Otolaryngology

Background:

• Cochlear implants (CIs) improve hearing in severe-to-profound hearing loss.
• Hearing outcomes correlate with electrode position relative to intracochlear anatomy.
• Accurate visualization of intracochlear anatomy is essential but challenging *in vivo*.

Purpose of the Study:

• To comprehensively evaluate the performance of active shape models (ASM) for segmenting intracochlear anatomy.
• To optimize ASM parameters using an extended dataset of micro-CT specimens.
• To assess the clinical reliability of the optimized ASM.

Main Methods:

• Utilized a dataset of 16 manually segmented cochlea specimens on micro-CT.
• Optimized ASM parameters to enhance segmentation accuracy.
• Evaluated the optimized ASM on a clinical dataset of 134 CT images.

Main Results:

• Achieved mean CT segmentation performance with 0.36 mm point-to-point error, 0.10 mm surface error, and 0.83 Dice score.
• Demonstrated diminishing returns of larger library sizes on segmentation performance.
• Identified the candidate search process as the primary limitation, not model representation.

Conclusions:

• Provided a comprehensive validation of ASM for intracochlear anatomy segmentation.
• Highlighted the clinical reliability of the ASM method.
• Emphasized the importance of understanding ASM limitations for clinical application and future development.