Radiation and Cancer Physics
PD 13 - Physics 8 - Poster Discussion - Outcome Analysis and Response Imaging
1112 - Machine Learning Methods Uncover Radio-Morphologic Dose Patterns in Salivary Glands That Predict Xerostomia in Head and Neck Cancer Patients
Tuesday, October 23
5:09 PM - 5:15 PM
Location: Room 217 C/D
Todd McNutt, PhD
Johns Hopkins University: Associate Professor: Employee
Elekta Oncology Systems: Research Grants; Philips Radiation Oncology Systems: Research Grants; Toshiba: Research Grants
Accuray-Tomotherapy: Patent/License Fees/Copyright; Sun Nuclear: Patent/License Fees/Copyright
AAPM-MAC: President Elect
Machine Learning Methods Uncover Radio-Morphologic Dose Patterns in Salivary Glands That Predict Xerostomia in Head and Neck Cancer Patients
T. R. McNutt1, W. Jiang2, P. Lakshminarayanan1, Z. Cheng3, M. R. Bowers1, H. Quon1, I. Shpitser4, S. Siddiqui5, P. Han1, and R. Taylor2; 1Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 2Johns Hopkins University, Baltimore, MD, 3Johns Hopkins Medicine, Baltimore, MD, 4Johns Hopkins University Department of Computer Science, Baltimore, MD, 5Johns hopkins University, Baltimore, MD
Purpose/Objective(s): Head and neck cancer (HNC) patients may experience xerostomia post radiotherapy, which leads to compromised quality of life. The purpose of this study is to explore how the spatial pattern of radiation dose (radio-morphology) in the major salivary glands influences xerostomia in head and neck cancer patients.
Materials/Methods: Existing studies generally used dose-volume histogram (DVH) features to study how radiation dose in organs affect xerostomia. However, DVH features lose the spatial information of dose within organs. A data-driven approach using spatially explicit dose features, i.e., actual radiation dose in voxels in parotid glands (PG) and submandibular glands (SMG), was used to predict if patients would develop xerostomia three months post radiotherapy. Utilizing planned radiation dose data and other non-dose covariates including baseline xerostomia grade of 427 HNC patients in our database, machine learning methods were used to investigate the influence of dose patterns across subvolumes in PG and SMG on xerostomia.
Results: Of the three supervised learning methods studied, ridge logistic regression yielded the best predictive performance. Ridge logistic regression was also preferred to evaluate the influence pattern of highly correlated dose on xerostomia. It showed a discriminative pattern of influence of dose in PG and SMG on xerostomia. Moreover, the superior, anterior portion of the contralateral PG and the medial portion of the ipsilateral PG were determined to be the most influential regions regarding dose effect on xerostomia. The area under the receiver operating characteristic curve (AUC) from 10-fold cross-validation is 0.70±0.04.
Conclusion: Radio-morphology combined with machine learning methods are able to suggest patterns of dose in PG and SMG that are most influential on xerostomia. The influence pattern identified by this data-driven approach and machine learning methods may help improve radiation treatment planning and reduce xerostomia after treatment.
Author Disclosure: T.R. McNutt: Research Grant; Elekta Oncology Systems, Philips Radiation Oncology Systems, Toshiba. Patent/License Fees/Copyright; Accuray-Tomotherapy, Sun Nuclear. President Elect; AAPM-MAC. W. Jiang: None. P. Lakshminarayanan: None. M.R. Bowers: Research Grant; Elekta Oncology Systems. H. Quon: None. I. Shpitser: None. P. Han: None. R. Taylor: None.