SU-E-T-349: Effective Dose-Volume-Histogram Prediction Method Using Euclidean Distance Volume Histogram for Volumetric Modulated Arc Therapy to Treat Prostate Cancer

Authors

  • Shin J.,

    1. Department of Radiation Oncology, Mayo Clinic Arizona, Phoenix, AZ 85054
    2. Department of Bioinformatics, Arizona State University, Scottsdale, AZ 85054
    3. Department of Computer Science, Arizona State University, Tempe, AZ
    Search for more papers by this author
  • Liang J. PhD,

    1. Department of Radiation Oncology, Mayo Clinic Arizona, Phoenix, AZ 85054
    2. Department of Bioinformatics, Arizona State University, Scottsdale, AZ 85054
    3. Department of Computer Science, Arizona State University, Tempe, AZ
    Search for more papers by this author
  • Schild S. MD,

    1. Department of Radiation Oncology, Mayo Clinic Arizona, Phoenix, AZ 85054
    2. Department of Bioinformatics, Arizona State University, Scottsdale, AZ 85054
    3. Department of Computer Science, Arizona State University, Tempe, AZ
    Search for more papers by this author
  • Wong W. MD,

    1. Department of Radiation Oncology, Mayo Clinic Arizona, Phoenix, AZ 85054
    2. Department of Bioinformatics, Arizona State University, Scottsdale, AZ 85054
    3. Department of Computer Science, Arizona State University, Tempe, AZ
    Search for more papers by this author
  • Liu W. PhD

    1. Department of Radiation Oncology, Mayo Clinic Arizona, Phoenix, AZ 85054
    2. Department of Bioinformatics, Arizona State University, Scottsdale, AZ 85054
    3. Department of Computer Science, Arizona State University, Tempe, AZ
    Search for more papers by this author

Abstract

Purpose:

Manually checking the plan dose-volume histograms (DVH) quality is time consuming, thereby limiting the patient throughput and leading to inconsistent plan quality across institutions. We propose to use Euclidean distance volume histogram (EDVH) to automate this process in volumetric-modulated arch therapy (VMAT) to treat prostate cancer patients by predicting DVH of a new patient based on prior expert patients.

Methods:

52 prostate cancer patients treated by VMAT at our institution were randomly selected. 3D distance transformation was performed on the simulation CTs to get Euclidean distance for each voxel inside every organ-at-risk with respect to the tumor. EDVH was then constructed, analogous to computing a DVH. In order to evaluate our proposed method, we utilize the leave-one-patient-out cross validation. For each test patient, we select 5 patients with the closest EDVHs from the remaining 51 patients and determine whether the DVH of this patient is within the band bounded by these 5 patients’ DVHs with a ratio defined as AREA5/AREA6, where AREA5 is the area-between-curves (ABC) bounded by the DVHs of the 5 selected patients and AREA6 is the ABC bounded by the DVHs of the 5 selected patients plus the test patient. The higher the ratio is, the better prediction performance is.

Results:

From the process of leave-one-patient-out cross validation, we obtained a prediction ratio (0.94±0.11) for each patient. 48 out of 52 patients (92.3%) for bladder and 46 out of 52 patients (88.5%) for rectum had prediction ratios above 0.80. 41 out of 52 patients (78.9%) for bladder and 38 out of 52 patients (73.1%) for rectum had prediction ratios above 0.90.

Conclusion:

EDVH can effectively capture the overall geometric relationship between the tumor and organ-at-risks, which can be used to confidently predict the DVH based on prior expert patients for VMAT to treat prostate cancer.

NIH/NCI K25CA168984, Eagles Cancer Research Career Development, The Lawrence W. and Marilyn W. Matteson Fund for Cancer Research, Mayo ASU Seed Grant, and The Kemper Marley Foundation.

Ancillary