Label-Free Prediction of Immunotherapy Response in Lung Cancer

Abstract

The advent of immune checkpoint blockade revolutionizes the landscape of cancer treatment. However, there are currently no biomarkers that can accurately predict the response of immunotherapy. In this work, we demonstrate label-free prediction of immunotherapy response in lung cancer using artificial intelligence-equipped multidimensional optical time-stretch imaging flow cytometry. First, the hypothesis of identifying immune activation of leukocytes via label-free images is confirmed using the in vitro coculture model. Then, with the support of the deep information mining capabilities of convolutional neural networks, we achieve prediction accuracies of 87 and 80% in lung cancer patients for the response and nonresponse to immunotherapy, respectively, significantly outperforming prediction using peripheral blood biomarkers. Furthermore, the experimental results on lung adenocarcinoma and lung squamous cell carcinoma patients show that our method is capable of predicting immunotherapy response with high accuracy across various types of lung cancer. We believe that our method can be applied to other types of cancer and will effectively enhance the specificity and efficacy of immunotherapy, thereby benefiting a large number of patients.