A pathomic approach for tumor-infiltrating lymphocytes classification on breast cancer digital pathology images

• DOI: 10.1016/j.heliyon.2023.e14371
• Publication Date: 2023-03-09T08:52:04Z
• Authors: Mario Verdicchio; Valentina Brancato; Carlo Cavaliere; Francesco Isgrò; Marco Salvatore; Marco Aiello
• Keywords: Digital Pathology; Tumor-infiltrating lymphocytes; Resampling

Background and objectivesThe detection of tumor-infiltrating lymphocytes (TILs) could aid in the development objective measures infiltration grade can support decision-making breast cancer (BC). However, manual quantification TILs BC histopathological whole slide images (WSI) is currently based on a visual assessment, thus resulting not standardized, reproducible, time-consuming for pathologists. In this work, novel pathomic approach, aimed to apply high-throughput image feature extraction techniques analyze microscopic patterns WSI, proposed. fact, features provide additional information concerning underlying biological processes compared WSI interpretation, providing more easily interpretable explainable results than most frequently investigated Deep Learning methods literature.MethodsA dataset containing 1037 regions interest with tissue compartments annotated 195 TNBC HER2+ hematoxylin eosin (H&E)-stained was used. After segmenting nuclei within tumor-associated stroma using watershed-based 71 were extracted from each nucleus reduced Spearman's correlation filter followed by nonparametric Wilcoxon rank-sum test least absolute shrinkage selection operator. The relevant used classify candidate as either or non-TILs 5 multivariable machine learning classification models trained 5-fold cross-validation (1) without resampling, (2) synthetic minority over-sampling technique (3) downsampling. prediction performance assessed ROC curves.Results21 selected, them related well-known properties having regular shape, clearer margins, high peak intensity, homogeneous enhancement different textural pattern other cells. best obtained Random-Forest AUC 0.86, regardless resampling technique.ConclusionsThe presented approach holds promise H&E-stained pathologists reliable, rapid clinical assessment BC.