Predicting Overall Survival Time in Glioblastoma Patients Using Gradient Boosting Machines Algorithm and Recursive...

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Glioblastoma multiforme (GBM) treatment response is inconsistent, leading to variable patient survival times.
• Predicting survival is crucial for effective GBM patient management and treatment planning.

Purpose of the Study:

• To integrate multimodal Magnetic Resonance Imaging (MRI) data with machine learning (ML) models to predict GBM patient survival.
• To identify key imaging features that correlate with glioblastoma survival outcomes.

Main Methods:

• Utilized multimodal MRI data from 29 GBM patients with known survival times.
• Applied Random Forest-Recursive Feature Elimination (RF-RFE) for feature selection, followed by a Gradient Boosting (GBoost) machine for survival prediction.
• Performed univariate and multivariate Cox regression analyses to assess the impact of imaging features on overall survival (OS).

Main Results:

• The RF-RFE GBoost model achieved 75% accuracy in predicting GBM patient survival time.
• Regional Cerebral Blood Volume (rCBV) in low perfusion areas showed significant differences between patient groups and strongly correlated with survival time.
• Feature selection methods combined with multimodal MRI enhanced classifier performance.

Conclusions:

• Integration of multimodal MRI and advanced feature selection techniques significantly improves the accuracy of GBM survival prediction.
• Specific imaging biomarkers, such as rCBV in low perfusion areas, are critical predictors of glioblastoma patient outcomes.
• This approach offers a promising tool for more personalized treatment strategies in neuro-oncology.