Premium
Predictive value of 0.35 T magnetic resonance imaging radiomic features in stereotactic ablative body radiotherapy of pancreatic cancer: A pilot study
Author(s) -
Simpson Garrett,
Spieler Benjamin,
Dogan Nesrin,
Portelance Lorraine,
Mellon Eric A.,
Kwon Deukwoo,
Ford John C.,
Yang Fei
Publication year - 2020
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1002/mp.14200
Subject(s) - medicine , magnetic resonance imaging , receiver operating characteristic , nuclear medicine , pancreatic cancer , confidence interval , radiology , radiation therapy , radiomics , radiosurgery , cancer
Purpose The aim of this study was to evaluate the potential and feasibility of radiomic features extracted from low field strength (0.35 T) magnetic resonance images (MRIs) in predicting treatment response for patients with pancreatic cancer undergoing stereotactic body radiotherapy (SBRT). Methods Twenty patients with unresected, non‐metastatic pancreatic ductal adenocarcinoma (PDAC) were enrolled, all of whom received neoadjuvant chemotherapy followed by five‐fraction MR‐guided SBRT with a radiation dose range of 33−50 Gy. For each patient, five daily setup scans were acquired from a hybrid 0.35 T MRI/radiotherapy unit. Tumor heterogeneity quantified with radiomic features extracted from the gross tumor volume (GTV) was averaged over the course of treatment. Random forest (RF) and adaptive least absolute shrinkage and selection operator (LASSO) classification models were constructed to identify radiomics features predictive of treatment response. Predictive capability of the top‐performing features was then evaluated using the receiver operating characteristic area under curve (AUC) obtained using leave‐one‐out cross‐validation. Results Half of the 20 patients showed response to treatment, defined by tumor regression on histopathology or tumor response on follow‐up dynamic contrast‐enhanced computed tomography (CT). The most predictive features selected by the RF method were GLCM energy and GLSZM gray‐level variance. The RF‐based model achieved an AUC = 0.81 with a 95% confidence interval of [0.594 to 1] The LASSO algorithm selected GLCM energy as the only predictive feature, achieving an AUC = 0.81 with 95% confidence interval of [0.596 to 1]. Conclusion The findings of this study suggest that radiomic features extracted during MR‐guided SBRT may contain predictive information about response of PDAC patients to treatment. Using the images acquired during treatment of PDAC patients supports continued expansion of radiomic analysis based on low field strength MR images and may hold the potential for providing timely indications of response to treatment.