Session Item

Clinically significant facial asymmetry (FA) occurs in 75% of children treated with radiotherapy for tumours in the head-and-neck region causing functional and aesthetic problems. Few dose-response relationships for facial bones have been developed. During follow-up, FA is assessed subjectively, but childhood cancer survivors also routinely receive MRI scans years after treatment. We hypothesize that deformable image registration (DIR) of those routine follow-up MR images can provide an objective measure of FA. The aim of this work was to validate quantitative measures derived from DIR as FA measures, compare them to anatomical landmarks, and report baseline (i.e., not treatment-related) FA in children.

T1w MR images were available for 100 unirradiated children from the Adolescent Brain Cognitive Development (ABCD) study (9-10 years). Two observers identified five anatomical landmarks on the left and right: angle (AM) and head of the mandible (HM), frontozygomatic process (FZ), frontonasal process (FN), and the superior orbital fissure (SO), Fig. 1A. We measured inter-observer variation (IOV) for landmarks on the right.

We quantified FA by mirroring the image and A) measuring the distance from the right landmarks to the mapped left landmarks, and B) reading the vector magnitude and Jacobian after DIR (NiftyReg, bending energy=0.01) of the mirrored and original images (Fig.1B). For these, we extracted the mean vector magnitude and Jacobian in a small sphere (radius 1.75mm) centred at each landmark.

Since the vector magnitude maps measure LR displacement of anatomy, which is analogous to landmark FA, we expected to find a linear relationship between vector magnitude and landmark distance. The correlation between the absolute value of 1-Jacobian (indicating shrinkage or expansion) and landmark distance was also explored. Baseline FA was analysed by sex and child size.

The median IOV ranged between 0.9-10.1mm. Three landmarks with IOV less than 5mm were selected for consecutive analysis.

A linear trend between landmark distance and vector magnitude was identified for HM (Fig.2A). This was not found for AM and FZ, as these landmarks are difficult to identify. No correlations were found between the Jacobian and landmark distance - the Jacobian provides a measure of local deformation information that cannot be measured with landmarks (Fig.2B). Overall, landmarks overestimate FA, as they are directly impacted by observer errors (Fig.2C). The vector magnitude was the most objective measure of FA and detected 0.4–7.8mm deviation that did not correlate with sex or brain volume (size surrogate).

Vector magnitude maps derived from DIR of MRI provide an objective measure of facial asymmetry. Jacobian maps provide additional information independent of landmark measures. Using DIR, we report up to 7.8mm of mandibular asymmetry in unirradiated children. These methods will be used to derive dose-response relationships for facial bones of childhood cancer survivors.