For instance, the tumor-stroma ratio (TSR) is associated with prognosis among patients with MIBC. TSR is typically estimated using visual assessment with a predefined cut-off point of 50% stromal content to divide samples into stromal-high or stromal-low categories. However, this technique is relatively subjective and may be inaccurate. Therefore, Zheng et al. developed an automated TSR assessment technique using machine learning.
The researchers collected retrospective clinical data and whole slide images (WSIs) for 261 patients from The Cancer Genome Atlas (TCGA) and 133 patients from the Renmin Hospital of Wuhan University (RHWU). The RHWU cohort was used as the discovery set for the machine learning algorithm, while the TCGA cohort was used as an external validation set. The model was first trained on defining tumor and stromal regions based on the distribution of four cell types in WSIs to generate TSR estimations. There was a high concordance between the machine-learning classification of regions and manual annotation by pathologists. Similarly, there was high agreement between the machine-learning TSR and manually derived TSR.
Next, analysis of the RHWU cohort revealed that the optimal cut-off point for stromal content was 45.7%. According to this cut-off point, 70% of RHWU patients and 71% of TCGA patients were classified as stroma-high. Low TSR was associated with shorter overall survival in both cohorts. Multivariate Cox analysis on the TCGA cohort revealed that TSR remained a significant prognostic factor after accounting for other variables. TSR analysis also enabled risk identification in patient subgroups (clinical stages, age, gender), adding value to existing staging systems. The researchers subsequently analyzed gene expression in the TCGA cohorts. They found that the expression of six genes (DACH1, DEEND2A, NOTCH4, DTWD1, TAF6L, and MARCHF5) was significantly correlated with TSR, thereby providing insights into potential cellular mechanisms underlying the effects of TSR.
Objective and accurate assessment of TSR in MIBC patients is potentially important for personalized therapy. The widespread use of machine learning analysis of WSIs is feasible and potentially more readily available compared to other techniques for biomarker assessment, such as sequencing.
Written by: Bishoy M. Faltas, MD, Director of Bladder Cancer Research, Englander Institute for Precision Medicine, Weill Cornell Medicine
Reference:
- Zheng Q, Jiang Z, Ni X, et al. Machine Learning Quantified Tumor-Stroma Ratio Is an Independent Prognosticator in Muscle-Invasive Bladder Cancer. Int J Mol Sci. 2023;24(3):2746. Published 2023 Feb 1. doi:10.3390/ijms24032746