Developing a metric for bone union in mandibular reconstruction using quantitative CT.

BACKGROUND

Objective quantification of bone union after mandibular reconstruction is important for evaluating reconstructive outcomes, yet current assessments are largely semi-quantitative.

OBJECTIVE

To explore the feasibility of using opportunistic quantitative computed tomography (CT)-derived Hounsfield unit (HU) measurements, with and without machine learning, to characterize bone union after fibula free flap mandibular reconstruction.

METHODS

In this proof-of-concept diagnostic mandibulectomy patients with variable clinical characteristics were selected from a prospectively maintained database at a quaternary referral center. CT scans from 2020-2024 were analyzed and quantitative HU measurements were obtained from buccal, lingual, and medullary bone at osteotomy sites. Bone union was graded using the Akashi scale. Logistic regression and random forest models were developed for binary and multiclass prediction, with performance assessed using area under the receiver operating characteristic curve (AUC), calibration metrics, and clustered cross-validation.

RESULTS

A total of 821 Hounsfield measurements from 280 axial CT slices were analyzed. Interrater agreement for Akashi scoring was 88.6% (κ = 0.79). Buccal HU was the strongest predictor, achieving an AUC of 0.74-0.75 in unadjusted analyses and 0.88-0.89 in adjusted logistic regression models. Random forest models achieved an AUC of 0.86 for union and 0.92 for complete union, with moderate to good calibration. Multiclass models showed good discrimination for non-union and complete union (AUC up to 0.86) but limited performance for partial union (AUC 0.68-0.73). Discriminative performance declined under clustered validation.

CONCLUSIONS

This exploratory study demonstrates the feasibility of using CT attenuation values to quantify bone union after mandibular reconstruction, supporting further validation in larger, multicenter cohorts.