T4: Predicting slide-level tumor proportion score in NSCLC IHC-stained WSI¶

Objective:
Develop a model to predict the Tumor Proportion Score (TPS) from immunohistochemically (IHC) stained whole slide images (WSIs) of non-small cell lung cancer (NSCLC). TPS is defined as the percentage of PD-L1-positive tumor cells relative to the total number of tumor cells in the slide.

Patient Population:
534 lung cancer patients from Radboudumc. Biopsies were taken from both primary lung tumors and metastatic sites.

Imaging Data:

• 534 PD-L1 IHC-stained WSIs provided as .tif files
• Resolution: 0.5 microns per pixel
• Each slide is accompanied by a binary tissue mask (.tif):
  • Label 0: Background
  • Label 1: Tissue region

Test Data:
Unlabeled WSIs and tissue masks in the same format. Participants must predict the binned TPS (0,1,2) value for each slide.

Reference Standard:
A binned TPS value per case, extracted from pathology reports and based on visual estimates by experienced lung pathologists. In clinical settings, these estimates are typically categorized into three bins:

| 0: TPS < 1% | 1: 1% ≤ TPS < 50% | 2: TPS ≥ 50% |

Evaluation Metrics:
Cohen’s quadratic weighted kappa between predicted and reference binned TPS is evaluated.