Image Segmentation and Registration

We specifically selected CECT scans demonstrating the earliest radiographically evident ORN characteristic lesion(s) as reported by radiologists and further confirmed by physical examination by physicians in Head & Neck Surgery as well as in Dental Oncology [ORN CECT].

The original delivered DICOM-RT clinical treatment plans were restored from Pinnacle treatment planning system (Pinnacle, Phillips Medical Systems, Andover, MA) into commercially available image registration software (VelocityAI™ 3.0.1). Diagnostic CECT scans at baseline, post-RT2, post-RT6, and ORN were also imported. Radiographically evident bony lesions were delineated manually by a radiation oncologist (HE) to constitute the ORN volumes of interest (VOIs). Physical exam and other available imaging modalities such as dental-dedicated panoramic X-rays were utilized to guide the segmentation of VOIs.

Planning CT was co-registered with ORN CECT using deformable image registration algorithm of VelocityAI™ 3.0.1. The 3D reconstructed dose grid of RT plan was then overlaid to the ORN CECT. A neighboring radiographically intact mandibular subvolume within the same isodose distribution volume was manually segmented and designated as “Control VOI” at the ORN CECT. Subsequently, baseline, post-RT2, post-RT6 CECT scans were co-registered with ORN CECT using rigid registration algorithms of VelocityAI™ 3.0.1. Both “ORN” & “Control” VOIs were propagated from ORN CECT to other CECT scans at all three prior time points (Figure 2).

Imaging workflow. Registration of CECT scan at time of diagnosis of ORN to radiation dose grid as well as previous CECT scans at: baseline, 2-month, and 6-month post-RT for each patient with subsequent propagation of ORN & “Control” VOIs.