Predictive power of combined inflammatory markers and magnetic resonance imaging features for glioma grading using machine learning: a retrospective study

Yaohua Liu, Yuting Wang, Junle Zhu, Shuang Qin, Yi Hong, Saifei Sun, Qin Zhang, Qi LvBMC Medical Imaging 2025 · DOI

TL;DR

This study demonstrates that combining routine MRI features with simple blood-based inflammatory markers can significantly improve the accuracy of predicting glioma grade using machine learning. The support vector machine (SVM) model, in particular, achieved high performance, suggesting a promising non-invasive approach for better glioma prognosis and personalized treatment.

What Problem Does This Paper Address?

Gliomas, the most common primary brain tumors, pose a significant clinical challenge due to their high recurrence rates and poor survival. Accurately grading these tumors (determining their aggressiveness) is crucial for treatment planning and prognosis. Traditionally, this relies on invasive surgical biopsies and histopathological analysis, which can be subject to variability and sampling errors. While MRI provides valuable insights into tumor characteristics, and inflammatory markers are increasingly recognized for their prognostic value, there's a need for a more integrated, non-invasive approach to predict glioma grade effectively. This paper aims to bridge this gap by exploring the combined predictive power of MRI features and inflammatory markers using machine learning.

Methods

The researchers employed a retrospective study design, analyzing data from 179 glioma patients. They extracted key MRI features, including contrast enhancement ratio (ER) and relative apparent diffusion coefficient (rADC), known indicators of tumor vascularity and cellularity, respectively. Simultaneously, they analyzed several inflammatory and nutritional markers from blood tests, such as the neutrophil-to-lymphocyte ratio (NLR), lymphocyte-to-CRP ratio (LCR), and hemoglobin-albumin-lymphocyte-platelet index (HALP).

These features were then fed into three classical machine learning classifiers: k-nearest neighbors (KNN), support vector machines (SVM) with a radial basis function kernel (r-SVM), and random forest (RF). The dataset was split into training (75%) and testing (25%) sets, with five-fold cross-validation used during training to ensure robustness. Feature selection was performed using ANOVA F-tests and Pearson correlation to identify the most discriminatory variables. Model performance was evaluated using metrics like Area Under the Curve (AUC), accuracy, and F1 score.

Datasets

The study utilized a dataset of 179 glioma patients from Shanghai Tongji Hospital. This dataset was split into training and testing cohorts for model development and evaluation. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Synergy of Modalities: The core idea is that combining imaging data (MRI) with easily obtainable biological data (blood markers) can provide a more comprehensive picture of glioma biology than either modality alone.
  • Inflammation as a Key Player: The study highlights the significant role of systemic inflammation, as reflected by specific blood markers, in influencing glioma grade and progression.
  • Machine Learning for Complex Relationships: Machine learning is crucial for uncovering subtle, non-linear relationships between these diverse features and the complex biological processes underlying glioma grading.
  • SVM's Superiority: The findings suggest that SVM, known for its ability to handle high-dimensional data, is particularly well-suited for this classification task, outperforming KNN and RF in this study.

Key Results

  • The Support Vector Machine (SVM) model achieved the highest predictive performance on the test set, with an AUC of 0.92 and an F1 score of 0.91.
  • Key features that showed significant discriminatory power between low-grade and high-grade gliomas included age, rADC, monocyte count, LCR, and HALP.
  • The study demonstrated that the combined approach significantly improved predictive accuracy compared to using individual modalities.
  • Kaplan-Meier survival analysis showed significantly longer overall survival (OS) and progression-free survival (PFS) for patients with low-grade gliomas compared to high-grade gliomas.

Limitations and Future Work

The authors acknowledge several limitations:

  • Single-Institution Study: The findings may not be generalizable to other populations due to recruitment from a single hospital.
  • Retrospective Nature: This design can introduce selection bias.
  • Limited Imaging Data: The relatively small number of imaging datasets prevented external validation, which is crucial for confirming model robustness.

Future work suggested includes:

  • External Validation: Conducting studies across multiple centers to validate the models.
  • Incorporating Molecular Data: Integrating genetic markers like IDH1 mutation status and MGMT promoter methylation to further refine predictive accuracy.
  • Exploring Other Imaging Modalities: Investigating the use of perfusion MRI and MR spectroscopy for a more comprehensive assessment.
  • Developing a Clinical Decision Support Tool: Creating a user-friendly tool to aid clinicians in predicting glioma grade and guiding treatment.

Summary updated

BibTeX
@article{liu2025predictive,
  author = {Yaohua Liu and Yuting Wang and Junle Zhu and Shuang Qin and Yi Hong and Saifei Sun and Qin Zhang and Qi Lv},
  title = {{Predictive power of combined inflammatory markers and magnetic resonance imaging features for glioma grading using machine learning: a retrospective study}},
  journal = {BMC Medical Imaging},
  year = {2025},
  doi = {10.1186/s12880-025-01946-0},
}
Figure