2024 World Conference on Lung Cancer (WCLC) - Posters-P1.06A.04 Deep Learning-based Whole Slide Image Analysis Predicts PD-L1 Status from H&E-Stained Histopathology Images in Lung Cancer

P1.06A.04 Deep Learning-based Whole Slide Image Analysis Predicts PD-L1 Status from H&E-Stained Histopathology Images in Lung Cancer

Back to course

Pdf Summary

The document outlines a study on predicting PD-L1 status in lung cancer using deep learning-based analysis of H&E-stained histopathology images. This research, led by Yi-Fan Qi and colleagues from various Chinese universities and hospitals, focuses on improving the prediction of PD-L1 status, which is crucial for determining the potential benefit of immune checkpoint inhibitors (ICIs) for lung cancer patients.<br /><br />The study utilized 348 whole slide images (WSIs) from lung cancer patients, which were divided into patches and analyzed using a convolutional neural network (CNN) based on ResNet50. Through an attention mechanism, the model predicts PD-L1 status by producing a score between 0 and 1, classifying scores above 0.5 as positive for PD-L1 expression.<br /><br />Key results show that the predictive model had an area under the curve (AUC) of 0.88 in the training set and 0.75 in the testing set, with a sensitivity of 71.05% and a positive predictive value of 79.41%. The study cohort primarily included stage III lung cancer patients, with adenocarcinoma being the most common pathology type.<br /><br />The research emphasizes the potential of deep learning methodologies in clinical oncology, particularly in predicting which patients might benefit from ICIs. Future research aims to expand the dataset and incorporate additional clinical and molecular data to enhance predictive accuracy and validate the models in prospective studies. By addressing the variability in manual assessment, this approach could standardize PD-L1 status prediction, offering a more reliable basis for therapeutic decision-making in lung cancer treatment.

Asset Subtitle

Yi-Fan Qi

Meta Tag

Speaker Yi-Fan Qi

Topic Pathology & Biomarkers

Keywords

PD-L1 status

lung cancer

deep learning

H&E-stained images

ResNet50

convolutional neural network

immune checkpoint inhibitors

predictive model

clinical oncology

adenocarcinoma