AI Model Detects STAS in Lung Cancer Histopathological Images

A new model called the Diffusion Attention Expert Model (DAEM) has been introduced by researchers to identify spread through air spaces (STAS) in histopathological images of lung cancer, tackling the challenges of labor-intensive and error-prone diagnostic processes. This model incorporates a diffusion attention expert module that aggregates full attention to extract multi-scale features from frozen sections (FSs) and paraffin sections (PSs), alongside a dual-branch architecture to enhance feature representation. In tests on an internal dataset, DAEM recorded AUCs of 0.8946 for FSs and 0.9112 for PSs. Its validation across external multi-center datasets from eight institutions demonstrates robust generalizability and interpretability. The study is available on arXiv (2605.16444).