CBCT Condylar Segmentation Holds Up in Unilateral Condylar Hyperplasia

In patients with unilateral condylar hyperplasia, the ability to measure condylar volume consistently is more than a technical exercise. It is central to tracking abnormal mandibular growth, judging when surgery is warranted, and estimating whether a proportional condylectomy has actually altered the bony asymmetry driving malocclusion and facial distortion.

This PLOS ONE reliability study evaluated two semi-automatic segmentation strategies for cone-beam computed tomography, one anchored to the sigmoid notch and the other to the mandibular lingula. The core question was simple and clinically relevant: which landmark-based method can be repeated with less measurement noise when condylar anatomy is already distorted by disease and surgery?

The answer was encouraging for both approaches. Each method achieved excellent intra-rater reliability, but the sigmoid notch method produced slightly tighter agreement and lower random error.

Why Volumetric Precision Matters in Condylar Hyperplasia

Mandibular condylar hyperplasia creates a moving target. The condyle grows independently and asymmetrically, which can enlarge one side of the mandible, shift the occlusion, and produce progressive lower facial asymmetry. In practical terms, clinicians are trying to quantify a structure that is not only small and irregular, but also surrounded by the glenoid fossa, articular disc, soft tissue interfaces, and dense temporal bone that complicate three-dimensional segmentation.

Traditional panoramic radiography has not offered dependable reproducibility for this task. CBCT is better suited to three-dimensional assessment, yet condylar segmentation still depends on where the operator chooses to place lower and medial boundaries. Small differences in landmark choice can translate into meaningful changes in measured volume, especially in a structure as anatomically compressed as the mandibular condyle.

The present analysis directly addressed that limitation by comparing two segmentation workflows in a real clinical cohort of patients treated with proportional condylectomy. The design focused on reliability, not treatment efficacy, but the measurement question itself carries downstream clinical weight. If a segmentation method is unstable, the volume change attributed to disease or surgery becomes difficult to trust.

What the Study Measured Across 20 Condyles

Ten adult patients with UCH were randomly selected from a private practice cohort in Santiago, Chile. Each patient contributed affected and non-affected sides, and the dataset included scans before and after unilateral proportional condylectomy. In total, 20 condyles were evaluated. The imaging workflow used open-source software: 3D Slicer for segmentation and ITK-Snap for slice-by-slice refinement and volumetric measurement.

A single blinded operator performed all measurements twice, separated by a four-week interval. This setup matters because it isolates intra-rater consistency rather than inter-operator variation. The investigators also used the Dahlberg formula to quantify random error and a two-way mixed-effects intraclass correlation coefficient model to assess repeatability.

The segmentation pathway began with automatic landmark generation and manual refinement of 13 cephalometric points, followed by head orientation using the Frankfort horizontal, midsagittal, and transporionic planes. The mandible was then segmented, refined in all three dimensions, and separated into condylar units using one of two boundary definitions. In method SN, the sigmoid notch served as the landmark for the lower, anterior, and medial borders. In method ML, the mandibular lingula served the same role.

Slightly Less Variability With the Sigmoid Notch Method

The main result was a high level of reproducibility for both segmentation methods. Method SN reached an ICC of 0.96, with a 95% confidence interval of 0.92 to 0.98. Method ML reached an ICC of 0.91, with a 95% confidence interval of 0.79 to 0.96. Both values fall in the excellent range, which is reassuring for clinicians who depend on serial imaging to follow mandibular asymmetry.

The difference between the two methods emerged in measurement stability. Method ML had a larger random error, with 220.18 mm3 compared with 38.33 mm3 for method SN. On repeat measurement, ML values were 9.9% higher on average, while SN values were 4.7% higher. In a field where the condylar unit is already difficult to isolate, that extra variability can matter when small volume changes are being interpreted as biologically meaningful.

One sentence captures the practical takeaway: the sigmoid notch method was the steadier of the two.

Bedside Implications for Surgical Follow-Up and 3D Imaging Workflows

For oral and maxillofacial teams, the study strengthens the case for using volumetric CBCT analysis when evaluating complex mandibular asymmetry. It also suggests that a carefully defined anatomical landmark can reduce the drift that often undermines repeated measurements. If a method can be reproduced by the same examiner with excellent agreement, it becomes more credible for preoperative planning, postoperative surveillance, and research on surgical outcomes.

The mandibular lingula still has clinical value. It performed well enough to qualify as a reliable alternative and may remain attractive where workflow simplicity or access to landmarking tools favors its use. Even so, the higher random error argues for caution when condylar volume differences are subtle or when longitudinal changes are modest.

The study also reflects a broader imaging reality in craniofacial surgery: measurement quality depends not only on scanner resolution, but on anatomy-aware segmentation rules. A condyle is not a tidy geometric object. It sits at a crowded anatomical junction, and the boundary chosen to define it will influence every subsequent comparison.

Limits of the Evidence and What the Dataset Can Support

This was a small retrospective sample, drawn from 10 patients in a private practice setting. The analysis therefore supports reliability, not generalizability across every scanner, operator, or patient population. All measurements were made by a single trained examiner, so the data do not address inter-rater reproducibility. The study also does not prove that one method is superior for every clinical purpose; it shows only that both methods can be used consistently, with a modest edge for the sigmoid notch approach.

The publication identifier for this study is 10.1371/journal.pone.0363132. Future work will need larger cohorts, multiple raters, and direct comparison across additional segmentation strategies before condylar volume thresholds can be used with greater confidence in routine decision-making.


Reporting by Dr. Aishwarya N., MBBS for Medicine World Foundation, with additional data provided by PLOS ONE (Medicine).
PRIMARY CLINICAL EVIDENCE & REGISTRY DATA