Identification of the most significant magnetic resonance imaging (MRI) radiomic features in oncological patients with vertebral bone marrow metastatic disease: a feasibility study
- 222 Downloads
Recently, radiomic analysis has gained attention as a valuable instrument for the management of oncological patients. The aim of the study is to isolate which features of magnetic resonance imaging (MRI)-based radiomic analysis have to be considered the most significant predictors of metastasis in oncological patients with spinal bone marrow metastatic disease.
Materials and methods
Eight oncological patients (3 lung cancer; 1 prostatic cancer; 1 esophageal cancer; 1 nasopharyngeal cancer; 1 hepatocarcinoma; 1 breast cancer) with pre-radiotherapy MR imaging for a total of 58 dorsal vertebral bodies, 29 metastatic and 29 non-metastatic were included. Each vertebral body was contoured in T1 and T2 weighted images at a radiotherapy delineation console. The obtained data were transferred to an automated data extraction system for morphological, statistical and textural analysis. Eighty-nine features for each lesion in both T1 and T2 images were computed as the median of by-slice values. A Wilcoxon test was applied to the 89 features and the most statistically significant of them underwent to a stepwise feature selection, to find the best performing predictors of metastasis in a logistic regression model. An internal cross-validation via bootstrap was conducted for estimating the model performance in terms of the area under the curve (AUC) of the receiver operating characteristic.
Of the 89 textural features tested, 16 were found to differ with statistical significance in the metastatic vs non-metastatic group. The best performing model was constituted by two predictors for T1 and T2 images, namely one morphological feature (center of mass shift) (p value < 0.01) for both datasets and one histogram feature minimum grey level (p value < 0.01) for T1 images and one textural feature (grey-level co-occurrence matrix joint variance (p value < 0.01) for T2 images. The internal cross-validation showed an AUC of 0.8141 (95% CI 0.6854–0.9427) in T1 images and 0.9116 (95% CI 0.8294–0.9937) in T2 images.
The results suggest that MRI-based radiomic analysis on oncological patients with bone marrow metastatic disease is able to differentiate between metastatic and non-metastatic vertebral bodies. The most significant predictors of metastasis were found to be based on T2 sequence and were one morphological and one textural feature.
KeywordsVertebral metastases Radiomics Magnetic resonance Quantitative imaging Oncology Radiotherapy
The authors would like to thank Roberto Gatta (Department of Radiation Oncology—Gemelli-ART, Catholic University of Rome, School of Medicine, Foundation University Hospital “A. Gemelli”) for the experienced support during data analysis.
Conflict of interests
All the authors declare they do not have conflict of interests.
All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.
- 7.Ma W, Ji Y, Qi L, Guo X, Jian X, Liu P (2018) Breast cancer Ki67 expression prediction by DCE-MRI radiomics features. Clin Radiol S0009–9260(18):30220–30224Google Scholar
- 8.Abdollahi H, Mahdavi SR, Mofid B, Bakhshandeh M, Razzaghdoust A, Sadipour A, Tanha K (2018) Rectal wall MRI radiomics in prostate cancer patients: prediction of and correlation with early rectal toxicity. Int J Radiat Biol 3:1–36Google Scholar
- 14.Romanos O, Solomou E, Georgiadis P, Kardamakis D, Siablis D (2013) Magnetic resonance imaging and image analysis of post—radiation changes of bone marrow in patients with skeletal metastases. J BUON 18(3):788–794Google Scholar
- 18.R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/
- 20.Priestley M (1981) Spectral analysis and time series. Volume 1: univariate series. Academic Press, CambridgeGoogle Scholar
- 21.Dinapoli N, Alitto AR, Vallati M, Gatta R, Autorino R, Boldrini L, Damiani A, Valentini V (2015) Moddicom: a complete and easily accessible library for prognostic evaluations relying on image features. Conf Proc IEEE Eng Med Biol Soc 2015:771–774Google Scholar