Artificial intelligence for quantifying Crohn's-like lymphoid reaction and tumor-infiltrating lymphocytes in colorectal cancer

作者：Xu, Yao; Yang, Shangqing; Zhu, Yaxi; Yao, Su; Li, Yajun; Ye, Huifen; Ye, Yunrui; Li, Zhenhui; Wu, Lin; Zhao, Ke^*; Huang, Liyu^*; Liu, Zaiyi^*

来源：Computational and Structural Biotechnology Journal, 2022, 20: 5586-5594.

DOI：10.1016/j.csbj.2022.09.039

• Crohn?s-like lymphoid reaction
• Tumor-infiltrating lymphocytes
• Artificial intelligence
• Colorectal cancer
• Whole-slide images

Crohn's-like lymphoid reaction (CLR) and tumor-infiltrating lymphocytes (TILs) are crucial for the host antitumor immune response. We proposed an artificial intelligence (AI)-based model to quantify the den-sity of TILs and CLR in immunohistochemical (IHC)-stained whole-slide images (WSIs) and further con-structed the CLR-I (immune) score, a tissue level-and cell level-based immune factor, to predict the overall survival (OS) of patients with colorectal cancer (CRC). The TILs score and CLR score were obtained according to the related density. And the CLR-I score was calculated by summing two scores. The devel-opment (Hospital 1, N = 370) and validation (Hospital 2 & 3, N = 144) cohorts were used to evaluate the prognostic value of the CLR-I score. The C-index and integrated area under the curve were used to assess the discrimination ability. A higher CLR-I score was associated with a better prognosis, which was iden-tified by multivariable analysis in the development (hazard ratio for score 3 vs score 0 = 0.22, 95% con-fidence interval 0.12-0.40, P < 0.001) and validation cohort (0.21, 0.05-0.78, P = 0.020). The AI-based CLR-I score outperforms the single predictor in predicting OS which is objective and more prone to be deployed in clinical practice.

• 单位
  南方医科大学; 西安电子科技大学; 广东省人民医院; 中山大学; 6; 广东省心血管病研究所