Machine Learning-Based Prognosis Prediction in Glioblastoma Multiforme Patients by Integrating Clinical Data with...

Area of Science:

• Radiology and Imaging
• Machine Learning in Oncology
• Neuro-oncology

Background:

• Glioblastoma multiforme (GBM) is an aggressive brain tumor characterized by intratumoral heterogeneity.
• Tumor hypoxia is a critical factor in GBM heterogeneity, significantly impacting patient prognosis.
• Accurate prediction of survival is crucial for effective treatment planning in GBM.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting one-year survival in GBM patients.
• To investigate the utility of radiomics features from fluoromisonidazole (FMISO)-PET and MRI, alongside clinical data, for survival prediction.
• To identify key predictors of one-year survival in GBM.

Main Methods:

• Retrospective analysis of data from 35 GBM patients from the ACRIN 6684 trial.
• Utilized FMISO-PET, MRI (T1, T2, FLAIR) images, and clinical information.
• Employed machine learning algorithms (SVM, RF, LR) and evaluated model performance using ROC curves and AUC.

Main Results:

• The model using clinical data achieved the highest predictive performance with an AUC of 0.921.
• FMISO-PET radiomics models showed strong performance (AUC = 0.870).
• Female sex and younger age were significantly associated with better one-year survival (p < 0.05).

Conclusions:

• Machine learning models, particularly those incorporating clinical features, demonstrate significant potential for predicting one-year survival in GBM.
• These predictive models can aid in tailoring personalized therapeutic strategies for GBM patients.
• Further validation with larger cohorts is recommended to confirm the generalizability of the findings.