Radiation and Cancer Physics

SS 27 - Physics 7 - Special Session: Outcome Analysis and Modeling

190 - Application of Machine Learning and Radiomics to Distinguish True Progression From Treatment Effect After Stereotactic Radiation Therapy for Brain Metastases

Tuesday, October 23
2:45 PM - 2:55 PM
Location: Room 214 C/D

Application of Machine Learning and Radiomics to Distinguish True Progression From Treatment Effect After Stereotactic Radiation Therapy for Brain Metastases
L. C. Peng1, P. Huang2, V. Parekh3, K. Sheikh1, B. R. Baker1, T. Kirschbaum1, F. Silvestri1, J. Son1, A. Robinson1, H. Ames4, J. Grimm1, L. Chen1, C. Shen1, M. Soike5, E. McTyre6, K. J. Redmond1, M. Lim1, M. A. Jacobs3, J. Lee1, and L. R. Kleinberg1; 1Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University School of Medicine, Baltimore, MD, 2Department of Oncology - Biostatistics and Bioinformatics Division, Johns Hopkins University School of Medicine, Baltimore, MD, 3Department of Radiology, Johns Hopkins University School of Medicine, Baltimore, MD, 4Department of Pathology, Johns Hopkins University School of Medicine, Baltimore, MD, 5Department of Radiation Oncology, Wake Forest School of Medicine, Winston-Salem, NC, 6Wake Forest Baptist Medical Center, Winston-Salem, NC

Purpose/Objective(s): Growth due to treatment effect after stereotactic radiosurgery (SRS) for brain metastases is a common phenomenon often indistinguishable from true progression on imaging. Radiomics is an emerging field which promises to improve upon conventional imaging. In this study, we hypothesized that a radiomics-based prediction model would outperform a prediction model based solely on clinical data in distinguishing true progression after SRS.

Materials/Methods: Patients at a single institution treated with SRS for brain metastases from 2003-2017 were reviewed to identify cases with subsequent resection for suspected progression. Additional cases of likely treatment effect were also included where lesions grew but subsequently regressed spontaneously. Lesions were contoured on T1 post-contrast and T2 FLAIR MRI at time of maximal growth, and radiomic features were extracted via the PyRadiomics library. A subset of features was then selected for training random forest (RF) classifiers. Performance of this radiomics-based classifier (RadiomicRF) was assessed with leave-one-out cross-validation (LOOCV) and compared with the LOOCV performance of a clinical RF classifier (ClinicalRF) trained only with the following features: age, sex, prior whole brain radiation, prescribed dose (in BED10), prescription isodose line, primary pathology, lesion location, lesion volume, time to progression, and receipt of immunotherapy.

Results: We identified 82 treated lesions across 66 patients, with 77 lesions having pathologic confirmation. At least some component of true progression was identified in 52/82 (63%) cases. There were 1904 radiomic features extracted per contoured lesion. Through iterative analysis of correlation clusters, 10 features with minimal redundancy and low correlation were selected for training the RadiomicRF in a process blinded to pathologic outcomes. In performance testing, ClinicalRF had an accuracy of 58%, with sensitivity 29%, specificity 75%, and area under the receiver operating characteristic curve (AUC) 0.59 on LOOCV. By contrast, RadiomicRF on LOOCV had an accuracy of 77%, sensitivity 85%, specificity 63%, and AUC 0.84. The radiomic feature most predictive for TP was gray level size-zone matrix (GLSZM) non-uniformity, a marker for tumor heterogeneity.

Conclusion: Radiomics holds promise for differentiating between treatment effect and true progression in brain metastases treated with SRS. A predictive model built on radiomic features from an institutional cohort substantially outperformed a clinical predictive model on LOOCV. These results warrant further validation in independent datasets. Such work could prove invaluable for guiding management of individual patients and assessing outcomes of novel interventions.

Author Disclosure: L.C. Peng: Employee; Johns Hopkins University School of Medicine. P. Huang: None. V. Parekh: None. B.R. Baker: None. T. Kirschbaum: None. F. Silvestri: None. J. Son: None. A. Robinson: None. H. Ames: None. J. Grimm: Employee; Academic Urology. Research Grant; Novocure, Accuray. Honoraria; Accuray. Travel Expenses; Accuray. Patent/License Fees/Copyright; DVH Evaluator. E. McTyre: Employee; Wake Forest University Baptist Medical Center. K.J. Redmond: Research Grant; Elekta AB, Accuray. Honoraria; AstraZeneca, Accuray. Travel Expenses; Elekta AB, Accuray. M.A. Jacobs: Research Grant; NIH. In-kind Donation; NVIDIA. J. Lee: Research Grant; Canon, Inc. L.R. Kleinberg: Research Grant; Novocure, Accuray. Honoraria; Accuray. Advisory Board; Novocure.

Send Email for Luke Peng


Assets

190 - Application of Machine Learning and Radiomics to Distinguish True Progression From Treatment Effect After Stereotactic Radiation Therapy for Brain Metastases



Attendees who have favorited this

Please enter your access key

The asset you are trying to access is locked. Please enter your access key to unlock.

Send Email for Application of Machine Learning and Radiomics to Distinguish True Progression From Treatment Effect After Stereotactic Radiation Therapy for Brain Metastases