Variability in Manual Segmentation of Temporal Bone Structures in Cone Beam CT Images.

Area of Science:

• Medical Imaging
• Anatomy
• Radiology

Background:

• Manual segmentation is the gold standard for anatomical labeling in clinical images.
• Variability in manual segmentation of temporal bone structures in CBCT images is not well-understood.
• Systematic evaluation of reviewer variability is crucial for reliable image analysis.

Purpose of the Study:

• To evaluate the intra- and inter-reviewer variability of manual segmentation for inner ear structures in CBCT scans.
• To assess the consistency of manual segmentation across multiple reviewers and sessions.
• To identify specific temporal bone structures with high segmentation variability.

Main Methods:

• 10 patients' preoperative CBCT scans of the inner ear were analyzed.
• Five reviewers manually segmented the cochlea, facial nerve, chorda tympani, mid-modiolar (MM) axis, and round window (RW).
• Segmentation was performed in two sessions at least one month apart, with variability assessed using Dice coefficient, volume similarity, and Hausdorff Distance.

Main Results:

• Manual segmentation of the cochlea demonstrated high consistency (mean DICE 0.91 intra-, 0.89 inter-reviewer).
• Facial nerve segmentation was also relatively consistent (mean DICE 0.83 intra-, 0.80 inter-reviewer).
• The chorda tympani showed the highest variability due to its small size; RW and MM axis also exhibited significant variability.

Conclusions:

• Significant variability exists in the manual segmentation of certain temporal bone structures.
• The findings highlight the need to consider reviewer variability when interpreting studies using manual segmentation.
• Standardization or advanced segmentation techniques may be necessary to improve consistency.