Iterative active deformational methodology for tumor delineation: Evaluation across radiation treatment stage and volume

Abstract

Purpose: To introduce, implement, and assess an iterative modification to the active deformational image segmentation method as applied to cervical cancer tumors. Materials and Methods: A comparison by Jaccard similarity (JS) between this active deformational method and manual segmentation was performed on tumors of various sizes across preradiation, 3 weeks postradiation, and 6 weeks postradiation using a General Linear Mixed Model across 121 studies from 52 patients with Stage IIB-IV cervical cancers. Results: The deformable segmentation method produced promising levels of agreement including J S factors of 0.71 ± 0.11 in the preradiation studies. The analysis illustrated a rate of improvement in JS with increasing tumor volume that differed between the preradiation and 6 weeks postradiation stage (P = 0.0474). In the large preradiated tumors each additional cm 3 of volume was associated with an increase or improvement in J S of 0.0008 (95% confidence interval [CI]: 0.0003, 0.0014). In the smaller postradiation tumors, each additional cm3 of volume was associated with a more robust improvement in JS of 0.0046 (95% CI: 0.0009, 0.0082). Conclusion: Agreement was strongly affected by tumor volume, and its performance was most impacted across volume in the later stages of radiation therapy. The deformation-based segmentation method appears to demonstrate utility for delineating cervical cancer tumors, particularly in the earliest stages of radiation treatment, where agreement is greatest. © 2008 Wiley-Liss, Inc.