Noninvasive Assessment of HER2 Expression Status in Gastric Cancer Using 18F-FDG Positron Emission Tomography/Computed Tomography-Based Radiomics: A Pilot Study

Xiaojing Jiang,Tianyue Li,Jianfang Wang,Zhaoqi Zhang,Xiaolin Chen,Jingmian Zhang,Xinming Zhao

DOI: https://doi.org/10.1089/cbr.2023.0162

2024-01-10

Cancer Biotherapy & Radiopharmaceuticals

Abstract:Purpose: Immunohistochemistry (IHC) is the main method to detect human epidermal growth factor receptor 2 (HER2) expression levels. However, IHC is invasive and cannot reflect HER2 expression status in real time. The aim of this study was to construct and verify three types of radiomics models based on 18 F-fuorodeoxyglucose ( 18 F-FDG) positron emission tomography/computed tomography (PET/CT) imaging and to evaluate the predictive ability of radiomics models for the expression status of HER2 in patients with gastric cancer (GC). Patients and Methods: A total of 118 patients with GC were enrolled in this study. 18 F-FDG PET/CT examination was underwent before surgery. The LIFEx software package was applied to extract PET and CT radiomics features. The minimum absolute contraction and selection operator (least absolute shrinkage and selection operator [LASSO]) algorithm was used to select the best radiomics features. Three machine learning methods, logistic regression (LR), support vector machine (SVM), and random forest (RF) models, were constructed and verified. The Synthetic Minority Oversampling Technique (SMOTE) was applied to address data imbalance. Results: In the training and test sets, the area under the curve (AUC) values of the LR, SVM, and RF models were 0.809, 0.761, 0.861 and 0.628, 0.993, 0.717, respectively, and the Brier scores were 0.118, 0.214, and 0.143, respectively. Among the three models, the LR and RF models exhibited extremely good prediction performance. The AUC values of the three models significantly improved after SMOTE balanced the data. Conclusions: 18 F-FDG PET/CT-based radiomics models, especially LR and RF models, demonstrate good performance in predicting HER2 expression status in patients with GC and can be used to preselect patients who may benefit from HER2-targeted therapy.

oncology,pharmacology & pharmacy,radiology, nuclear medicine & medical imaging,medicine, research & experimental