AUTHOR=Dikland Felix Anne , Fekih Cyrine , Wellenstein Marius René Jacques , Souza da Silva Ricella , Machado-Neves Raquel , Fraga João , Oliveira Domingos , Montezuma Diana , Pinto Isabel Macedo , Woodburn Jonathan 

TITLE=A scoping review of TSR analysis in colorectal cancer: implications for automated solutions

JOURNAL=Oncology Reviews

VOLUME=Volume 19 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/oncology-reviews/articles/10.3389/or.2025.1605383

DOI=10.3389/or.2025.1605383

ISSN=1970-5557

ABSTRACT=The tumour-stroma ratio (TSR), which refers to the composition of stromal tissue and tumour epithelium of a malignant lesion, is gaining recognition as a promising biomarker in pathology. In 2018, recommendations for quantifying TSR in colorectal carcinoma were published, yet diverse quantification methods are still in use today. To assess the prognostic value of TSR, evaluate the impact of scoring variations, and explore efforts to automate TSR quantification, a scoping review was conducted. A total of 950 articles were identified through PubMed and Scopus, of which 76 met the inclusion criteria for this review. Of these, 56 employed manual scoring methods, while 20 utilised semi-automated or fully automated TSR quantification techniques. The TSR has been consistently identified as a strong prognostic indicator for disease-free survival. Its association with poor prognosis may be linked to its correlation with metastatic status, perineural invasion, and vascular invasion in stroma-high lesions. Variability in TSR scoring protocols was most evident in the selection of the region of interest and the type of histological specimen, both of which had a direct impact on final TSR scores. Moreover, significant inter-observer variability was observed in manual semi-quantitative TSR assessments, with Kappa scores ranging from 0.42 to 0.88. Automated TSR scoring pipelines have been proposed to standardise scoring protocols and reduce inter-observer variability. Deep learning models have demonstrated promising results, with pixel-wise and patch-wise accuracies exceeding 95%. Even though deep learning approaches have shown high performance, discrepancies remain, as evidenced by Kappa scores ranging from 0.239 to 0.472. In conclusion, the variation in TSR scoring protocols, along with a wide range of inter-observer variability, limits the broader clinical application of TSR. While automated TSR quantification methods show promise, they are still in the early stages, particularly in relation to region of interest selection and stratifying patients into risk categories. As these methods evolve, adjustments to TSR scoring cut-off values may be necessary to improve consistency. This scoping review highlights the prognostic significance of TSR in colorectal carcinoma while emphasizing the challenges posed by variability in scoring methods and the need for further advancements in automated quantification.