Click here to


Are you sure ?

Yes, do it No, cancel

CT/MRI-Based Radiomics Analysis to Predict Radiation Induced Xerostomia in Head and Neck Cancer Radiotherapy

K Sheikh*, L Peng , S Lee , P Han , Z Cheng , P Lakshminarayanan , T McNutt , H Quon , J Lee , Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University, Baltimore, MD


(Tuesday, 7/31/2018) 10:30 AM - 11:00 AM

Room: Exhibit Hall | Forum 2

Purpose: To analyze baseline CT/MRI-based radiomics features of salivary glands to predict radiation-induced xerostomia after head and neck cancer (HNC) radiotherapy.

Methods: CT and T1 post-contrast MR images along with NCI-CTCAE xerostomia grade (at 3-month follow-up) were prospectively collected in 140 HNC patients treated between 2009 and 2015 at our institution. CT images were corrected for metal artifact reduction to minimize the effect of dental artifacts in radiomic feature computation. CT and MR images were rigidly or deformably registered on which parotid/submandibular glands were contoured by a radiation oncologist. In total, 5744 CT and MR image features were extracted for ipsilateral(IL)/contralateral(CL) parotid and submandibular glands relative to the location of the primary tumor. In univariate analysis, features were ranked by likelihood for predicting severe xerostomia (grade≥2). Features that were correlated with xerostomia (p<0.1) were further reduced using a maximum relevance minimum redundancy method. The final set of image features were selected using a LASSO logistic regression algorithm. A generalized linear model with leave-one-out-cross-validation was applied to predict radiation-induced xerostomia. Three models were developed with image features extracted from (1) CT-only, (2) MR-only, and (3) CT+MR. The area under the receiver operating characteristic curve (AUC), sensitivity, and specificity were compared across models.

Results: Among extracted features, 14 CT and 12 MR image features were selected. The 5 most important features included MR CL parotid short-run-high-gray-level-emphasis and IL submandibular energy, CT CL submandibular gray-level-run-length-matrix and IL submandibular gray-level-co-occurrence-matrix entropy. The cross-validated AUC/sensitivity/specificity for CT-only, MR-only, and CT+MR were 0.79/0.84/0.63, 0.74/0.74/0.61, and 0.82/0.84/0.63 in xerostomia prediction.

Conclusion: In HNC patients, combining baseline CT and MR image features improved classification performance severe radiation-induced xerostomia. Further investigation is required to further validate if this strategy can be useful for personalized radiotherapy.

Funding Support, Disclosures, and Conflict of Interest: Partial support provided by the Radiation Oncology Institute. T McNutt acknowledges funding support from Philips and Toshiba Medical Systems Corp. J Lee acknowledges funding support from Toshiba Medical Systems Corp and Johns Hopkins Radiation Oncology Discovery Award.


MRI, Radiation Effects, CT


IM/TH- Image Analysis (Single modality or Multi-modality): Imaging biomarkers and radiomics

Contact Email