2023 World Conference on Lung Cancer (Posters)-P2.05. Immunotherapy Prognostication with Thoracic CT Contrastive Self-Supervised Learning in Patients with Non-Small Cell Lung Cancer

P2.05. Immunotherapy Prognostication with Thoracic CT Contrastive Self-Supervised Learning in Patients with Non-Small Cell Lung Cancer - PDF(Slides)

Back to course

Pdf Summary

In this study, researchers aimed to develop a reliable imaging biomarker for predicting the response to immunotherapy in patients with non-small cell lung cancer (NSCLC). They used a contrastive self-supervised learning (SSL) model that was pre-trained on a large dataset of CT lesions. The extracted 3D image representations of lung tumors, or SSL features, were then used in the analysis.<br /><br />The study included a cohort of 561 NSCLC patients who were treated with immune checkpoint inhibitors (ICIs). The cohort was divided into a training/tuning group and an independent test group. Baseline chest CT scans and clinical parameters, including age, sex, performance status, and PD-L1 expression, were collected and analyzed.<br /><br />The researchers found that the SSL features predicted overall survival (OS) with an area under the curve (AUC) of 0.583 in the independent test cohort. In comparison, tumor volume and clinical factors had AUCs of 0.552 and 0.666, respectively. PD-L1 expression resulted in an AUC of 0.625. However, when the SSL features were combined with clinical variables, the predictive performance improved to an AUC of 0.68.<br /><br />Furthermore, the combination of SSL features and clinical variables successfully stratified the cohort into high and low-risk groups, as demonstrated by the Kaplan-Meier estimator.<br /><br />Overall, these findings suggest that SSL representations of baseline chest CT, when combined with clinical risk factors, can significantly improve the predictive performance for survival in NSCLC patients receiving immunotherapy compared to clinical measures alone.<br /><br />In conclusion, this study highlights the potential of contrastive self-supervised learning in developing a robust imaging biomarker for predicting immunotherapy response in NSCLC patients.

Asset Subtitle

Tafadzwa Chaunzwa

Meta Tag

Speaker Tafadzwa Chaunzwa

Topic Metastatic NSCLC: Immunotherapy - Biomarker

Keywords

reliable imaging biomarker

immunotherapy response

NSCLC

contrastive self-supervised learning

SSL features

cohort

baseline chest CT scans

clinical parameters

predictive performance

survival