Quantitative Assessment of Tumor-Infiltrating Lymphocytes Using Machine Learning Predicts Survival in Muscle-Invasive Bladder Cancer

Qingyuan Zheng, Rui Yang, Xinmiao Ni, Yang Song, Panpan Jiao, Jiejun Wu, Lin Xiong, Jingsong Wang, Jun Jian, Zhengyu Jiang, Lei Wang, Zhiyuan Chen, Xiuheng LiuJournal of Clinical Medicine 2022 · DOI

TL;DR

This paper introduces a machine learning (ML) approach to quantitatively assess tumor-infiltrating lymphocytes (TILs) in muscle-invasive bladder cancer (MIBC) from whole-slide images. This ML-driven method provides a robust and reproducible prognostic marker, potentially identifying patients who could benefit from immunotherapy.

What Problem Does This Paper Address?

Muscle-invasive bladder cancer (MIBC) is a serious malignancy with declining survival rates as the tumor invades deeper bladder layers. While tumor-infiltrating lymphocytes (TILs) are known to be important prognostic indicators, their assessment in clinical practice is often subjective and prone to inter-observer variability. This lack of standardization hinders their widespread use and limits their potential to guide treatment decisions, such as the use of immunotherapy. The paper aims to overcome this by developing an automated, quantitative method for TIL assessment using machine learning.

Methods

The researchers developed a pipeline for analyzing whole-slide images (WSIs) of bladder cancer tissue. They utilized the open-source QuPath software and a built-in machine learning algorithm to classify cells within the tumor microenvironment into four categories: tumor cells, lymphocytes, stromal cells, and "ignore" (background/artifacts). This classification was trained on H&E-stained slides.

Based on this cell classification, four distinct TILs variables were defined:

  • eTILs%: Proportion of TILs relative to tumor cells.
  • esTILs%: Proportion of TILs relative to stromal cells.
  • etTILs%: Proportion of TILs relative to all detected cells (tumor, stromal, and TILs).
  • eaTILs (mm²): Density of TILs within the tumor region.

These ML-derived TILs (eTILs) were then analyzed for their prognostic value using Kaplan-Meier survival curves and Cox regression models, both univariate and multivariate.

Datasets

The study utilized two independent cohorts for training and validation:

  • RHWU cohort: 133 MIBC patients from Renmin Hospital of Wuhan University. This served as the training and internal validation set.
  • TCGA cohort: 247 MIBC patients from The Cancer Genome Atlas. This served as the external validation set.

Key Insights and Ideas

  • Bridging the Subjectivity Gap: The core innovation is using ML to move beyond subjective visual assessment of TILs, offering a standardized and quantitative approach.
  • Prognostic Power Across Stages: The ML-derived TILs (specifically eTILs) demonstrated prognostic significance not only overall but also within specific subgroups, including early-stage (Stage II) and advanced-stage (Stage III-IV) MIBC patients.
  • Independence from Traditional Markers: The eTILs score proved to be an independent prognostic factor, meaning it provides valuable information beyond established markers like TNM staging, age, and lymphovascular invasion.
  • Potential for Personalized Therapy: The ability to identify high-risk patients who might benefit from immunotherapy opens doors for more personalized treatment strategies.

Key Results

  • All four ML-defined TILs variables showed significant prognostic associations with overall survival in both the RHWU and TCGA cohorts (p < 0.001).
  • Higher TILs scores were consistently associated with better prognosis.
  • In multivariate Cox regression, eTILs were independently associated with overall survival, even after adjusting for clinicopathological factors (p < 0.001).
  • The eTILs variable was found to be a strong prognostic factor that was not redundant with pre-existing clinicopathological features (p < 0.05 in subgroup analyses).
  • Among the four TILs variables tested, eTILs and etTILs were suggested to be potentially more robust methods for future studies.

Limitations and Future Work

The authors acknowledge several limitations:

  • Segmentation/Classification Errors: The ML algorithm's accuracy can be affected by variations in WSI staining and the inherent heterogeneity of tumor cells, potentially leading to misclassification.
  • Lack of Differentiation: The current model does not differentiate between stromal TILs (sTILs) and intratumoral TILs (iTILs), which could be important for prognosis.
  • Retrospective Nature: Both cohorts were retrospective, and WSIs from different institutions may have staining variations.
  • Cut-point Specificity: The optimal cut-points identified are specific to these cohorts and may need adjustment for broader clinical application, potentially by treating TILs scores as continuous variables.

Future work suggested includes:

  • Complementing TILs variables with molecular typing of cells.
  • Conducting larger prospective cohort studies to refine the best approach for identifying patients who would benefit from immunotherapy.
  • Developing appropriate cut-points for broader clinical use, possibly considering continuous variable analysis.

Summary updated

Abstract

(1) Purpose: Although assessment of tumor-infiltrating lymphocytes (TILs) has been acknowledged to have important predictive prognostic value in muscle-invasive bladder cancer (MIBC), it is limited by inter- and intra-observer variability, hampering widespread clinical application. We aimed to evaluate the prognostic value of quantitative TILs score based on a machine learning (ML) algorithm to identify MIBC patients who might benefit from immunotherapy or the de-escalation of therapy. (2) Methods: We constructed an artificial neural network classifier for tumor cells, lymphocytes, stromal cells, and “ignore” cells from hematoxylin-and-eosin-stained slide images using the QuPath open source software. We defined four unique TILs variables based on ML to analyze TILs measurements. Pathological slide images from 133 MIBC patients were retrospectively collected as the discovery set to determine the optimal association of ML-read TILs variables with patient survival outcomes. For validation, we evaluated an independent external validation set consisting of 247 MIBC patients. (3) Results: We found that all four TILs variables had significant prognostic associations with survival outcomes in MIBC patients (p < 0.001 for all comparisons), with higher TILs score being associated with better prognosis. Univariate and multivariate Cox regression analyses demonstrated that electronic TILs (eTILs) variables were independently associated with overall survival after adjustment for clinicopathological factors including age, sex, and pathological stage (p < 0.001 for all analyses). Results analyzed in different subgroups showed that the eTILs variable was a strong prognostic factor that was not redundant with pre-existing clinicopathological features (p < 0.05 for all analyses). (4) Conclusion: ML-driven cell classifier-defined TILs variables were robust and independent prognostic factors in two independent cohorts of MIBC patients. eTILs have the potential to identify a subset of high-risk stage II or stage III-IV MIBC patients who might benefit from adjuvant immunotherapy.
BibTeX
@article{zheng2022quantitative,
  author = {Qingyuan Zheng and Rui Yang and Xinmiao Ni and Yang Song and Panpan Jiao and Jiejun Wu and Lin Xiong and Jingsong Wang and Jun Jian and Zhengyu Jiang and Lei Wang and Zhiyuan Chen and Xiuheng Liu},
  title = {{Quantitative Assessment of Tumor-Infiltrating Lymphocytes Using Machine Learning Predicts Survival in Muscle-Invasive Bladder Cancer}},
  journal = {Journal of Clinical Medicine},
  year = {2022},
  doi = {10.3390/jcm11237081},
}
Figure