Predicting the Cell-level PD-L1 Expression Status from H&E-stained Histopathological Slide

TPS (Tumor Proportion Score) serves as a crucial indicator to determine the eligibility of cancer patients for PD-L1 immunohistochemistry (IHC) treatment. In clinical practice, pathologists assess TPS primarily using IHC histopathological slides, which is time-consuming and incurs high treatment costs. In this paper, we introduce a novel approach for predicting PD-L1 positive cells using cost-effective and widely accessible Hematoxylin and Eosin (H&E) stained slides. Here, we first extract features of positive cells from PD-L1 slides. Then mapping the position and expression status of these cells onto H&E slide to generate cell-level pseudo-labels. Finally, we develop a deep correlation model that links PD-L1's cellular expression patterns to the morphological features identifiable in H&E stains. This model enables the direct extraction of PD-L1 markers from H&E-stained slides, not only saving pathologists valuable diagnostic time but also avoiding the high costs associated with PD-L1 staining processes. Extensive results highlight the potential of our model as a promising alternative for assessing PD-L1 in cancer.Clinical Relevance-The proposed cell level detection model can achieve high accuracy. This indicates that predicting PD-L1 expression status from H&E is feasible and helps simplify the detection of immunohistochemistry in clinical practice.