Erschienen in:
09.11.2018 | Computer Applications
Preoperative radiomic signature based on multiparametric magnetic resonance imaging for noninvasive evaluation of biological characteristics in rectal cancer
verfasst von:
Xiaochun Meng, Wei Xia, Peiyi Xie, Rui Zhang, Wenru Li, Mengmeng Wang, Fei Xiong, Yangchuan Liu, Xinjuan Fan, Yao Xie, Xiangbo Wan, Kangshun Zhu, Hong Shan, Lei Wang, Xin Gao
Erschienen in:
European Radiology
|
Ausgabe 6/2019
Einloggen, um Zugang zu erhalten
Abstract
Objectives
To develop and validate radiomic models in evaluating biological characteristics of rectal cancer based on multiparametric magnetic resonance imaging (MP-MRI).
Methods
This study consisted of 345 patients with rectal cancer who underwent MP-MRI. We focused on evaluating five postoperative confirmed characteristics: lymph node (LN) metastasis, tumor differentiation, fraction of Ki-67-positive tumor cells, human epidermal growth factor receptor 2 (HER-2), and KRAS-2 gene mutation status. Data from 197 patients were used to develop the biological characteristics evaluation models. Radiomic features were extracted from MP-MRI and then refined for reproducibility and redundancy. The refined features were investigated for usefulness in building radiomic signatures by using two feature-ranking methods (MRMR and WLCX) and three classifiers (RF, SVM, and LASSO). Multivariable logistic regression was used to build an integrated evaluation model combining radiomic signatures and clinical characteristics. The performance was evaluated using an independent validation dataset comprising 148 patients.
Results
The MRMR and LASSO regression produced the best-performing radiomic signatures for evaluating HER-2, LN metastasis, tumor differentiation, and KRAS-2 gene status, with AUC values of 0.696 (95% CI, 0.610–0.782), 0.677 (95% CI, 0.591–0.763), 0.720 (95% CI, 0.621–0.819), and 0.651 (95% CI, 0.539–0.763), respectively. The best-performing signatures for evaluating Ki-67 produced an AUC value of 0.699 (95% CI, 0.611–0.786), and it was developed by WLCX and RF algorithm. The integrated evaluation model incorporating radiomic signature and MRI-reported LN status had improved AUC of 0.697 (95% CI, 0.612–0.781).
Conclusion
Radiomic signatures based on MP-MRI have potential to noninvasively evaluate the biological characteristics of rectal cancer.
Key Points
• Radiomic features were extracted from MP-MRI images of the rectal tumor.
• The proposed radiomic signatures demonstrated discrimination ability in identifying the histopathological, immunohistochemical, and genetic characteristics of rectal cancer.
• All MRI sequences were important and could provide complementary information in radiomic analysis.