CT-based radiomics model for differential diagnosis of fat-poor renal angiomyolipoma and homogeneous-density clear cell renal cell carcinoma
10.13929/j.issn.1003-3289.2020.05.022
- VernacularTitle: 基于CT影像组学模型鉴别肾乏脂肪血管平滑肌脂肪瘤与均质肾透明细胞癌
- Author:
Lei YAN
1
Author Information
1. School of Medicine, Qingdao University
- Publication Type:Journal Article
- Keywords:
Angiomyolipoma;
Kidney neoplasms;
Radiomics;
Tomography, X-ray computed
- From:
Chinese Journal of Medical Imaging Technology
2020;36(5):732-737
- CountryChina
- Language:Chinese
-
Abstract:
Objective: To observe the value of enhanced CT-based radiomics nomogram incorporated with radiomics signatures and clinical factors in differential diagnosis of fat-poor angiomyolipoma (fp-AML) and homogeneous density clear cell renal cell carcinoma (hd-ccRCC) before surgical operation. Methods: Data of 71 patients with fp-AML (fp-AML group, n=32) and hd-ccRCC (hd-ccRCC group,n=39) proved by pathology were retrospectively collected. Three-dimensional ROI were contoured manually at cortical, nephrographic and excretory phase images (CP, NP and EP), and the radiomics features were extracted. Then inter- and intra- class correlation coefficients (ICC) were used to exclude the inter-observer and intra-observer difference of ROI feature extraction. The LASSO regression method was used to select radiomics features. A regression formula was constructed by multivariate Logistic regression analysis. Radiomics scores of CP, NP, EP and all 3 phases were calculated. A combined radiomics nomogram was developed by incorporating clinical factors and radiomics score using Logistic multivariate regression model. The Hosmer-Lemeshow test was performed to assess the goodness-of-fit of the nomogram. The calibration of the nomogram was assessed with calibration curves. The differential effectiveness of the radiomics nomogram was evaluated on the basis of ROC curves. The decision curve was performed to evaluate the net benefits of the nomogram for differentiating fp-AML from hd-ccRCC. Results: Totally 1 029 features including intensity, shape, texture and wavelet features were extracted from all phases. Radiomics features with ICC greater than 0.75 were enrolled into the LASSO regression model. Totally 6, 6, 5 and 7 optimal features extracted from cortical, nephrographic and excretory phases and all 3 phases were selected, and AUC was 0.83 (95%CI[0.73,0.92]), 0.80 (95%CI[0.70,0.91] ), 0.78 (95%CI[0.68,0.89] ) and 0.86 (95%CI[0.77,0.95] ), respectively. AUC of the nomogram based on radiomics score of all 3 phases and clinical factors was 0.90 (95%CI[0.81,0.99]), and decision curve indicated that this radiomics nomogram had a satisfactory overall net benefit for differentiating fp-AML from hd-ccRCC before surgical operation. Conclusion: CT-based radiomics nomogram, which incorporates the radiomics signatures and clinical factors, shows favorable predictive value for differentiating fp-AML from hd-ccRCC, which might be helpful to accurate diagnosis before surgical operation.
