特征选择
机器学习
医学
降维
人工智能
计算机科学
Lasso(编程语言)
医学物理学
万维网
标识
DOI:10.1016/j.kint.2021.06.042
摘要
We read with great interest the article by Zheng et al., 1 Zheng J. Yu H. Batur J. et al. A multicenter study to develop a non-invasive radiomic model to identify urinary infection stone in vivo using machine-learning. Kidney Int. 2021; 100: 870-880 Abstract Full Text Full Text PDF Scopus (16) Google Scholar published in Kidney International. This study leveraged a noninvasive radiomic model to preoperatively predict infection stones. Despite the encouraging results, several methodological issues should be noted. A robust radiomic biomarker across various image acquisitions and feature selection methods is crucial for the reliability of subsequent modeling. The authors should include the radiomic features that did not show significant differences due to machine and acquisition parameters. More sophisticated and rigorous dimensionality reduction techniques (such as Pearson correlation coefficient analysis) need to be implemented because the least absolute shrinkage and selection operator (LASSO) results showed that some of the selected features are still highly correlated and thus would not contribute to adding more information. Given that the authors applied 4 feature selection algorithms, we recommend using the features that were repeatedly significant among all classifiers. Regarding the decision curve analysis, the comparison of net benefits between the radiomic model and the radiomic signature as well as the clinical model would be better to demonstrate the clinical usefulness of the radiomic model. Last but not least, the interpretation of the selected radiomic features might help us comprehend the underlying mechanism of the prediction. A multicenter study to develop a non-invasive radiomic model to identify urinary infection stone in vivo using machine-learningKidney InternationalVol. 100Issue 4PreviewUrolithiasis is a common urological disease, and treatment strategy options vary between different stone types. However, accurate detection of stone composition can only be performed in vitro. The management of infection stones is particularly challenging with yet no effective approach to pre-operatively identify infection stones from non-infection stones. Therefore, we aimed to develop a radiomic model for preoperatively identifying infection stones with multicenter validation. In total, 1198 eligible patients with urolithiasis from three centers were divided into a training set, an internal validation set, and two external validation sets. Full-Text PDF The authors reply:Kidney InternationalVol. 100Issue 5PreviewWe thank Zhang et al.1 for their interest in our study.2 Usually, feature reproducibility assessment is implemented for data dimension reduction. However, because the margins of a urinary stone in computed tomography images are clear, satisfactory interobserver feature extraction reproducibility was achieved in our study, with interclass correlation coefficients ranging from 0.848 to 1.000. Therefore, all extracted radiomics features were used for the subsequent modeling. Moreover, the 24 selected features had only a low pairwise correlation (mean absolute Spearman, ρ = 0.196), indicating that these features provide complementary information. Full-Text PDF
科研通智能强力驱动
Strongly Powered by AbleSci AI