Department of Molecular Imaging and Nuclear Medicine, Tianjin Medical University Cancer Institute and Hospital, Tianjin, 300020, China.
National Clinical Research Center for Cancer, Tianjin Key Laboratory of Cancer Prevention and Therapy, Tianjin's Clinical Research Center for Cancer, Tianjin, China.
Ann Nucl Med. 2022 Feb;36(2):172-182. doi: 10.1007/s12149-021-01688-3. Epub 2021 Oct 30.
Human epidermal growth factor receptor 2 (HER2) expression status determination significantly contributes to HER2-targeted therapy in breast cancer (BC). The purpose of this study was to evaluate the role of radiomics and machine learning based on PET/CT images in HER2 status prediction, and to identify the most effective combination of machine learning model and radiomic features.
A total of 217 BC patients who underwent PET/CT examination were involved in the study and randomly divided into a training set (n = 151) and a testing set (n = 66). For all four models, the model parameters were determined using a threefold cross-validation in the training set. Each model's performance was evaluated on the independent testing set using the receiver operating characteristic (ROC) curve, and AUC was calculated to get a quantified performance measurement of each model.
Among the four developed machine learning models, the XGBoost model outperformed other machine learning models in HER2 status prediction. Furthermore, compared to the XGBoost model based on PET alone or CT alone radiomic features, the predictive power for HER2 status by using XGBoost model based on PET/CTmean or PET/CTconcat radiomic fusion features was dramatically improved with an AUC of 0.76 (95% confidence interval [CI] 0.69-0.83) and 0.72 (0.65-0.80), respectively.
The established machine learning classifier based on PET/CT radiomic features is potentially predictive of HER2 status in BC.
人表皮生长因子受体 2（HER2）表达状态的确定对乳腺癌（BC）的 HER2 靶向治疗有重要贡献。本研究旨在评估基于 PET/CT 图像的放射组学和机器学习在 HER2 状态预测中的作用，并确定机器学习模型和放射组学特征的最佳组合。
本研究共纳入 217 例接受 PET/CT 检查的 BC 患者，并随机分为训练集（n=151）和测试集（n=66）。对于所有四个模型，使用训练集的三折交叉验证来确定模型参数。使用受试者工作特征（ROC）曲线评估每个模型在独立测试集上的性能，并计算 AUC 以获得每个模型的量化性能测量。
在所开发的四个机器学习模型中，XGBoost 模型在 HER2 状态预测方面优于其他机器学习模型。此外，与基于单独 PET 或 CT 的放射组学特征的 XGBoost 模型相比，基于 PET/CTmean 或 PET/CTconcat 放射组学融合特征的 XGBoost 模型对 HER2 状态的预测能力显著提高，AUC 分别为 0.76（95%置信区间 [CI] 0.69-0.83）和 0.72（0.65-0.80）。
基于 PET/CT 放射组学特征的建立机器学习分类器可能对 BC 中的 HER2 状态具有预测性。
