Radiomics-based machine learning model for efficiently classifying transcriptome subtypes in glioblastoma patients...

Area of Science:

• Neuro-oncology
• Radiomics
• Machine Learning in Medicine

Background:

• Glioma transcriptome subtypes are crucial biomarkers for diagnosis and prognosis.
• Current identification methods face limitations like long detection times and biopsy challenges.
• Intratumoral heterogeneity in glioblastoma complicates subtype classification.

Purpose of the Study:

• To evaluate an eXtreme Gradient Boosting (XGBoost)-based radiomics model for classifying glioblastoma (GBM) transcriptome subtypes.
• To overcome limitations of existing transcriptome subtype identification methods.
• To improve diagnostic and prognostic accuracy for GBM patients.

Main Methods:

• Retrospective analysis of TCGA-GBM and IvyGAP glioblastoma cohorts.
• Segmentation of MRI scans into enhancement of the tumor core (ET), non-enhancing tumor core (NET), and peritumoral edema (ED).
• Application of 704 handcrafted radiomics features and two-level feature selection (Spearman correlation, F-score tests).

Main Results:

• Identified 13 optimal radiomics features for classification.
• Achieved high predictive accuracies for transcriptome subtypes: Classical (70.9%), Mesenchymal (73.3%), Neural (88.4%), and Proneural (88.4%).
• Demonstrated improved model performance compared to existing methods on the same dataset.

Conclusions:

• The XGBoost radiomics model with feature selection shows high potential for classifying glioblastoma transcriptome subtypes.
• This approach offers a promising, non-invasive method for subtype identification.
• Further research into radiomics-based GBM models is warranted.