用于评估结直肠癌免疫浸润的无标注全切片图像分析方法

Tumor-infiltrating lymphocytes (TILs) play a crucial role in host antitumor processes. High level of TILs is associated with better outcomes for patients. We aim to automatically quantify TILs without any nuclei annotation and further construct an immune score to predict the survival of patients with colorectal cancer.

We developed an artificial intelligence-based pipeline to quantify the level of TILs in hematoxylin-eosin (H&E) images with immunohistochemistry (IHC)-guided nuclei annotation-free auxiliary labels. Patients enrolled in the study were divided into two cohorts (development cohort: n = 557; validation cohort: n = 439). The proposed tumor stroma immune score (TSI score) reflects the immune infiltration in the tumor stroma of patients.

A strong correlation of TILs was observed between IHC and H&E images made by consecutive sections staining. Moreover, multivariate analysis confirmed that TSI score was an independent prognostic factor, and the higher TSI score was associated with the better prognosis (development cohort: hazard ratio for high v low 0.54 [95% CI, 0.42 to 0.80], P = .001; validation cohort: 0.68 [95% CI, 0.27 to 0.94], P = .031). The model with TSI score showed a higher C-index (development cohort: 0.700 v 0.679; validation cohort: 0.689 v 0.677).

IHC images have the value of providing TIL density reference information for spatial corresponding H&E images. TSI score has the ability to predict overall survival.