A Novel Swarm Intelligence-Driven Feature Selection for Interpretable Machine Learning in Multiparametric MRI-Based...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Glioblastoma multiforme (GBM) is an aggressive brain tumor with poor prognosis.
• Accurate prediction of overall survival (OS) is crucial for treatment planning in GBM patients.
• Multiparametric MRI-derived radiomic features (RFs) hold potential for non-invasive prognostic assessment.

Purpose of the Study:

• To develop and validate an interpretable machine learning (ML) model for OS estimation in GBM patients.
• To integrate a hybrid swarm intelligence (SI)-based feature selection with MRI-derived RFs.
• To achieve robust risk stratification using an interpretable ML approach.

Main Methods:

• A cohort of 276 GBM patients with pre-treatment MRI data was analyzed.
• A hybrid SI-based feature selection method was combined with a regularized Cox regression model (Cox-LASSO).
• Model performance was evaluated using concordance index (C-index) and validated through cross-validation and external testing.

Main Results:

• The developed ML model achieved a C-index of 0.71 in the test dataset and 0.67 upon external validation.
• The model demonstrated statistically significant risk stratification for OS (p < 0.001).
• The final model incorporated patient age and ten independent radiomic features.

Conclusions:

• The study successfully developed an interpretable ML model integrating SI-based feature selection and MRI radiomics for GBM OS prediction.
• The model achieved significant risk stratification, outperforming traditional methods.
• This represents one of the first studies to utilize an SI-based interpretable ML model for GBM survival stratification.