Evaluation of an automatic registration-based algorithm for direct measurement of volume change in tumors

Int J Radiat Oncol Biol Phys. 2012 Jul 1;83(3):1038-46. doi: 10.1016/j.ijrobp.2011.07.040. Epub 2011 Dec 14.

Abstract

Purpose: Assuming that early tumor volume change is a biomarker for response to therapy, accurate quantification of early volume changes could aid in adapting an individual patient's therapy and lead to shorter clinical trials. We investigated an image registration-based approach for tumor volume change quantification that may more reliably detect smaller changes that occur in shorter intervals than can be detected by existing algorithms.

Methods and materials: Variance and bias of the registration-based approach were evaluated using retrospective, in vivo, very-short-interval diffusion magnetic resonance imaging scans where true zero tumor volume change is unequivocally known and synthetic data, respectively. The interval scans were nonlinearly registered using two similarity measures: mutual information (MI) and normalized cross-correlation (NCC).

Results: The 95% confidence interval of the percentage volume change error was (-8.93% to 10.49%) for MI-based and (-7.69%, 8.83%) for NCC-based registrations. Linear mixed-effects models demonstrated that error in measuring volume change increased with increase in tumor volume and decreased with the increase in the tumor's normalized mutual information, even when NCC was the similarity measure being optimized during registration. The 95% confidence interval of the relative volume change error for the synthetic examinations with known changes over ±80% of reference tumor volume was (-3.02% to 3.86%). Statistically significant bias was not demonstrated.

Conclusion: A low-noise, low-bias tumor volume change measurement algorithm using nonlinear registration is described. Errors in change measurement were a function of tumor volume and the normalized mutual information content of the tumor.

Publication types

  • Evaluation Study
  • Research Support, N.I.H., Extramural
  • Validation Study

MeSH terms

  • Algorithms*
  • Breast Neoplasms / pathology*
  • Breast Neoplasms / therapy
  • Confidence Intervals
  • Diffusion Magnetic Resonance Imaging / methods*
  • Female
  • Humans
  • Image Processing, Computer-Assisted / methods*
  • Linear Models
  • Models, Statistical
  • Retrospective Studies
  • Tomography, X-Ray Computed / methods
  • Tumor Burden*