Department of Radiology and Imaging Sciences, Sri Ramachandra Institute of Higher Education and Research, Porur, Chennai, 600 116, India.
Department of Engineering Design, Indian Institute of Technology-Madras, Chennai, 600 036, India.
Sci Rep. 2023 Nov 4;13(1):19062. doi: 10.1038/s41598-023-46391-7.
In an observational study conducted from 2016 to 2021, we assessed the utility of radiomics in differentiating between benign and malignant lung nodules detected on computed tomography (CT) scans. Patients in whom a final diagnosis regarding the lung nodules was available according to histopathology and/or 2017 Fleischner Society guidelines were included. The radiomics workflow included lesion segmentation, region of interest (ROI) definition, pre-processing, and feature extraction. Employing random forest feature selection, we identified ten important radiomic features for distinguishing between benign and malignant nodules. Among the classifiers tested, the Decision Tree model demonstrated superior performance, achieving 79% accuracy, 75% sensitivity, 85% specificity, 82% precision, and 90% F1 score. The implementation of the XGBoost algorithm further enhanced these results, yielding 89% accuracy, 89% sensitivity, 89% precision, and an F1 score of 89%, alongside a specificity of 85%. Our findings highlight tumor texture as the primary predictor of malignancy, emphasizing the importance of texture-based features in computational oncology. Thus, our study establishes radiomics as a powerful, non-invasive adjunct to CT scans in the differentiation of lung nodules, with significant implications for clinical decision-making, especially for indeterminate nodules, and the enhancement of diagnostic and predictive accuracy in this clinical context.
在一项从 2016 年到 2021 年进行的观察性研究中，我们评估了放射组学在区分 CT 扫描检测到的良性和恶性肺结节方面的效用。纳入的患者根据组织病理学和/或 2017 年 Fleischner 学会指南得出了关于肺结节的最终诊断。放射组学工作流程包括病变分割、感兴趣区域（ROI）定义、预处理和特征提取。通过随机森林特征选择，我们确定了 10 个用于区分良性和恶性结节的重要放射组学特征。在测试的分类器中，决策树模型表现出优越的性能，达到 79%的准确性、75%的敏感性、85%的特异性、82%的精度和 90%的 F1 分数。XGBoost 算法的实施进一步提高了这些结果，达到 89%的准确性、89%的敏感性、89%的精度和 89%的 F1 分数，特异性为 85%。我们的研究结果强调肿瘤纹理是恶性的主要预测因子，突出了基于纹理的特征在计算肿瘤学中的重要性。因此，我们的研究表明，放射组学是 CT 扫描在区分肺结节方面的有力、非侵入性辅助手段，对临床决策具有重要意义，特别是对不确定的结节，并提高了这一临床背景下的诊断和预测准确性。
