Related Experiment Video
Updated: Apr 24, 2026
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
A multi-task masked autoencoder with GAN-based augmentation for PD-L1 prediction from chest CT images
Ying-Zhen Ye1, Pei-Yu Chou1, De-Xiang Ou1
1Department of Biomedical Engineering, College of Medicine and College of Engineering, National Taiwan University, No. 1, Sec. 4, Roosevelt Rd., Taipei, Taiwan.
Abstract:
Targeted and immune-based therapies, such as PD-1/PD-L1 inhibitors, have become standard treatments for advanced non-small cell lung cancer (NSCLC). However, accurately identifying patients who benefit from these therapies remains challenging due to tumor heterogeneity and variability in PD-L1 staining. To address this issue, we propose a non-invasive, model-driven computer-aided diagnosis framework that predicts PD-L1 expression directly from CT images under limited labeled data conditions. This study included 188 NSCLC patients from two university hospitals, of whom 49 had PD-L1 expression ≥ 50% and 139 had < 50%. We introduce a Multi-task Masked Autoencoder (MTMAE) with three key components: (1) a self-supervised masked image modeling strategy to leverage unlabeled data and improve data efficiency, (2) an integrated segmentation task to enhance tumor-focused feature learning, and (3) a Gabor-based generative adversarial network for data augmentation to improve generalization. The proposed model achieved an AUC of 0.735 and an accuracy of 0.724, outperforming traditional supervised pretraining (AUC 0.695) and single-task MAE (AUC 0.712). These results demonstrate that combining self-supervised learning, multi-task learning, and GAN-based augmentation enables a reproducible and standardized model-based prediction of clinically reported PD-L1 status from CT images, providing a non-invasive complementary tool for treatment decision-making.
Insights
A new computer-aided diagnosis framework predicts PD-L1 expression in non-small cell lung cancer (NSCLC) from CT scans. This non-invasive approach aids treatment decisions by analyzing tumor imaging data.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Immune checkpoint inhibitors targeting PD-1/PD-L1 are standard for advanced non-small cell lung cancer (NSCLC).
- Patient selection for these therapies is difficult due to tumor heterogeneity and variable PD-L1 staining.
- Accurate prediction of PD-L1 expression is crucial for effective treatment stratification.
Purpose of the Study:
- To develop a non-invasive, model-driven computer-aided diagnosis (CADx) framework.
- To predict PD-L1 expression directly from CT images in NSCLC patients.
- To address challenges posed by limited labeled data and improve treatment decision-making.
Main Methods:
- A Multi-task Masked Autoencoder (MTMAE) framework was developed using CT images from 188 NSCLC patients.
- The MTMAE incorporates self-supervised masked image modeling for data efficiency.
- Integrated segmentation and Gabor-based generative adversarial network (GAN) augmentation enhance feature learning and generalization.
Main Results:
- The proposed MTMAE model achieved an Area Under the Curve (AUC) of 0.735 and an accuracy of 0.724.
- This performance surpassed traditional supervised pretraining (AUC 0.695) and single-task Masked Autoencoder (MAE) (AUC 0.712).
- The model demonstrated effective prediction of PD-L1 status from CT images.
Conclusions:
- Combining self-supervised learning, multi-task learning, and GAN augmentation provides a robust method for PD-L1 prediction.
- This model-based approach offers a reproducible and standardized non-invasive tool for predicting PD-L1 status.
- The framework serves as a valuable complementary tool for guiding NSCLC treatment decisions.
