Prediction of prostate cancer aggressiveness with a combination of radiomics and machine learning-based analysis of dynamic contrast-enhanced MRI. Issue 11 (November 2019)

• Record Type: Journal Article
• Authors: Liu, B.; Cheng, J.; Guo, D.J.; He, X.J.; Luo, Y.D.; Zeng, Y.; Li, C.M.
• Is Part Of: Clinical radiology. Volume 74:Issue 11(2019)
• Page Start: 896.e1
• Page End: 896.e8
• Publication Date: 2019-11
• Subjects: Medical radiology -- Periodicals; Radiotherapy -- Periodicals; Radiotherapy -- Periodicals; Radiology -- Periodicals; Societies, Medical -- Periodicals; Medical radiology; Radiotherapy; Electronic journals; Periodicals; 616.0757
• DOI: 10.1016/j.crad.2019.07.011
• Languages: English
• ISSNs: 0009-9260

Abstract

Abstract : AIM: To investigate whether the combination of radiomics and automatic machine learning-based classification of original images from multiphase dynamic contrast-enhanced (DCE)-magnetic resonance imaging (MRI) can predict prostate cancer (PCa) aggressiveness before biopsy. MATERIALS AND METHODS: Forty consecutive biopsy-confirmed PCa patients were included. Biopsy was performed within 4 weeks after the DCE-MRI examinations. According to the time–signal-intensity curve, lesion segmentation was performed on the first and on the strongest phase of the enhancement on the original DCE-MRI images, and 1, 029 quantitative radiomics features were calculated automatically from each lesion, wherein there were three datasets available (Dataset-F, Dataset-S and Dataset-FS). The variance threshold method, select k-best method and least absolute shrinkage and selection operator (LASSO) algorithm were used to reduce the feature dimensions. Five machine learning approaches leveraging cross-validation were employed, and the clinical value of each model was evaluated by area under the receiver operating characteristic curve (AUC). Correlation analysis was performed between the features of the machine learning model that achieved the best classification performance and the Gleason score (GS) of the PCa lesion. RESULTS: Eight, four, and 16 features were selected as optimal subsets in Dataset-F, -S and -FS, respectively. Among all three datasets, logistic regression (LR)-based analysis with Dataset-FS had the highest predication efficacy (AUC=0.93). Ten features in Dataset-FS showed significantly positively correlation with GS. The model performance of Dataset-F was generally better than that in Dataset-S. CONCLUSIONS: A combination of radiomics and machine learning-analysis based analysis of the union of the first and strongest phases of original DCE-MRI images can predict PCa aggressiveness non-invasively, accurately, and automatically. Highlights: An innovative prostate cancer aggressiveness prediction approach is proposed. The first enhanced phase of DCE-MRI has great value in the prediction task. Radiomics features can characterize pathophysiology of prostate cancer.