Predicting Postoperative Functional Outcomes in High-Grade Glioma Using RS-fMRI and Machine Learning

Overview

This study demonstrates that machine learning models leveraging resting state fMRI data can accurately predict postoperative functional outcomes in patients with high-grade gliomas. Using random forest classifiers trained on functional connectivity metrics, tumor overlap with resting state networks, tumor volume, and age, the models distinguished patients with favorable (KPS ≥ 70) versus unfavorable (KPS < 70) functional status after surgery.

Background

High-grade gliomas (HGG) represent the majority of brain tumor cases and are characterized by aggressive growth and poor prognosis, with median survival around 14 months. Surgical resection remains the mainstay of treatment but poses risks of functional impairment, which correlates with overall patient outcomes. Conventional task-based fMRI for preoperative functional mapping has limitations, whereas resting state fMRI (RS-fMRI) offers a noninvasive alternative that can be performed under sedation and maps multiple networks simultaneously. Machine learning techniques have shown promise in extracting predictive features from complex imaging data, potentially aiding in forecasting postoperative functional status to guide surgical planning.

Data Highlights

Key Findings

• Random forest models trained on RS-fMRI derived features accurately classified postoperative functional outcomes in HGG patients.
• Functional connectivity measures and tumor overlap with resting state networks were strong predictors of postoperative Karnofsky Performance Status.
• Age and tumor volume contributed additional predictive value to the models.
• Autoencoder-based dimensionality reduction effectively handled high-dimensional RS-fMRI data for model training.
• Permutation feature importance analysis identified key imaging and clinical variables influencing functional outcome predictions.

Clinical Implications

Preoperative prediction of functional outcomes using RS-fMRI combined with machine learning can enhance surgical planning by balancing maximal tumor resection with preservation of neurological function. This approach may improve patient counseling regarding expected postoperative quality of life and aid in individualized treatment strategies. Additionally, RS-fMRI’s ability to map multiple networks under sedation broadens applicability in patients unable to perform task-based fMRI.

Conclusion

Machine learning models utilizing resting state fMRI data provide a promising tool for forecasting postoperative functional status in high-grade glioma patients, potentially guiding surgical decision-making and improving patient outcomes.

References

1. Ostrom et al. 2020 -- CBTRUS Statistical Report: Primary Brain and Other Central Nervous System Tumors Diagnosed in the United States
2. Louis et al. 2021 -- WHO Classification of Tumors of the Central Nervous System
3. Additional references [1-23] as cited in the article