Introduction:Glioma Grade 4 (GG4) tumors, which include both IDH-mutated and IDH wild-type astrocytomas, are the most prevalent and aggressive form of primary brain tumor. Radiomics is gaining ground in neuro-oncology. The integration of this data into machine learning models has the potential to improve the accuracy of prognostic models for GG4 patients. Karnofsky Performance Status (KPS), an established preoperative prognostic factor for survival, is commonly used in these patients. In this study, we developed a nomogram to identify patients with improved functional performance as indicated by an increase in KPS after surgery by analyzing radiomic features from preoperative 3D MRI scans.Methods:Quantitative imaging features were extracted from the -3D T1 GRE sequence of 157 patients from a single center and were used to develop the machine learning (ML) model. To improve applicability and create a nomogram, multivariable logistic regression analysis was performed to build a model incorporating clinical characteristics and radiomics features.Results:We labeled 55 cases in which KPS was improved after surgery (35%, KPS-flag = 1). The resulting model was evaluated according to test series results. The best model was obtained by XGBoost using the features extracted by pyradiomics, with a Matthew coefficient score (MCC) of 0.339 (95% CI: 0.330–0.3483) in cross-validation. The out-of-sample evaluation on the test set yielded an MCC of 0.302. A nomogram evaluating the improvement of KPS post-surgery was built based on statistically significant variables from multivariate logistic regression including clinical and radiomics data (c-index = 0.760, test set).Conclusions:MRI radiomic analysis represents a powerful tool to predict postoperative functional outcomes, as evaluated by KPS.
引言：胶质瘤4级（GG4）肿瘤，包括IDH突变型和IDH野生型星形细胞瘤，是最常见且最具侵袭性的原发性脑肿瘤。影像组学在神经肿瘤学领域正日益受到重视。将此类数据整合到机器学习模型中，有望提高GG4患者预后模型的准确性。卡氏功能状态评分（KPS）作为一项公认的术前生存预后因素，常被用于此类患者的评估。本研究通过分析术前3D MRI扫描的影像组学特征，开发了一种列线图，旨在识别术后KPS评分提高、功能状态改善的患者。 方法：从单中心157例患者的3D T1 GRE序列中提取定量影像特征，并用于构建机器学习模型。为提高适用性并创建列线图，采用多变量逻辑回归分析构建了一个整合临床特征与影像组学特征的模型。 结果：我们标记了55例术后KPS改善的病例（占35%，KPS标志=1）。根据测试集结果对所得模型进行评估。最佳模型通过XGBoost算法结合pyradiomics提取的特征获得，交叉验证的Matthew相关系数（MCC）为0.339（95% CI：0.330–0.3483）。测试集的样本外评估MCC为0.302。基于多变量逻辑回归中具有统计学意义的变量（包括临床和影像组学数据），构建了评估术后KPS改善情况的列线图（c指数=0.760，测试集）。 结论：MRI影像组学分析是预测术后功能结局（以KPS评估）的有效工具。
肿瘤(癌症)患者之家