Machine learning and glioma imaging biomarkers.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Magnetic resonance imaging (MRI) is crucial in neuro-oncology, providing anatomical and physiological details.
• Machine learning (ML) algorithms can identify complex image features for accurate classification.
• ML aids in determining tumor characteristics, grade, and prognosis from initial MRI scans.

Purpose of the Study:

• To review the application of machine learning (ML) in neuro-oncology imaging biomarkers.
• To explore ML's role in diagnosis, prognosis, and treatment response monitoring for brain tumors.
• To assess ML's utility in differentiating treatment effects from tumor progression.

Main Methods:

• Systematic literature search of PubMed and MEDLINE databases up to September 2018.
• Focus on studies applying ML to high-grade glioma biomarkers for prediction and monitoring.
• Analysis of research utilizing MRI features for ML-based classification and prediction.

Main Results:

• ML is frequently used with MRI features for accurate classification and identification of image biomarkers.
• Significant research applies ML to predict molecular profiles, tumor grade, and prognosis from initial MRIs.
• ML is actively studied for distinguishing treatment response from post-treatment effects in glioma imaging.

Conclusions:

• Current evidence for ML in neuro-oncology biomarkers is largely retrospective and from single centers.
• ML models have not yet demonstrated a clear advantage over traditional statistical methods in neuro-oncology.
• Development requires large, well-annotated datasets, necessitating multidisciplinary, multi-center collaborations.