This study aimed to develop a computed tomography (CT)-based radiomics model capable of precisely predicting hyperprogression and pseudoprogression (PP) in patients with non-small cell lung cancer (NSCLC) treated with immunotherapy. We retrospectively analyzed 105 patients with NSCLC, from three institutions, treated with immune checkpoint inhibitors (ICIs) and categorized them into training and independent testing set. Subsequently, we processed CT scans with a series of image-preprocessing techniques, and 6008 radiomic features capturing intra- and peritumoral texture patterns were extracted. We used the least absolute shrinkage and selection operator logistic regression model to select radiomic features and construct machine learning models. To further differentiate between progressive disease (PD) and hyperprogressive disease (HPD), we developed a new radiomics model. The logistic regression (LR) model showed optimal performance in distinguishing PP from HPD, with areas under the receiver operating characteristic curve (AUC) of 0.95 (95% confidence interval [CI]: 0.91-0.99) and 0.88 (95% CI: 0.66-1) in the training and testing sets, respectively. Additionally, the support vector machine model showed optimal performance in distinguishing PD from HPD, with AUC of 0.97 (95% CI: 0.93-1) and 0.87 (95% CI: 0.72-1) in the training and testing sets, respectively. Kaplan‒Meier survival curves showed clear stratification between PP predicted by the radiomics model and true progression (HPD and PD) (hazard ratio = 0.337, 95% CI: 0.200-0.568, p < 0.01) in overall survival. Our study demonstrates that radiomic features extracted from baseline CT scans are effective in predicting PP and HPD in patients with NSCLC treated with ICIs.
Keywords: Computed tomography; hyperprogression; immunotherapy; non-small cell lung cancer; pseudoprogression; radiomics.
© 2024 The Author(s). Published with license by Taylor & Francis Group, LLC.