Assessment of breast cancer mesenchymal tumor infiltrating lymphocytes based on regional segmentation and nuclear segmentation classification

Abstract

Histological assessment of mesenchymal tumor infiltrating lymphocytes (sTIL) can serve as a proxy for host immune response and has been shown to be prognostic and potentially chemically predictive in Her2 positive and triple negative breast cancers. Current manual assessment practices are discrepancy between Intra- and Inter-observer. To address this challenge, this paper proposes a region segmentation- based and nuclear segmentation classification method for sTIL assessment of H&E-stained breast cancer pathology images.The experimental results of mesenchymal region segmentation showed that the LinkNet-based segmentation method could effectively segment the mesenchymal region of breast cancer pathology images with a segmentation accuracy DICE index of 0.9274. The experimental results of nuclear cell classification showed that the Random Forest classifier outperformed other nuclear classification methods with an Accuracy index of 0.837. And the best F1-score index was obtained for the classification of epithelial cells, lymphocytes, and mesenchymal cells, with 0.744, 0.752, and 0.639, respectively. in addition, the correlation between the pathologist score and the calculated sTIL score was analyzed in this paper, in which the pathologist consensus score and the calculated sTIL score was highly correlated, correspondingly a Spearman correlation coefficient r was 0.869, which was greater than the Spearman correlation coefficient r of 0.851 for the inter-pathologist score. The finds verified the validity of the sTIL score calculation based on the regional segmentation and nuclear segmentation classification method proposed in this paper.