Brain metastasis detection using machine learning: a systematic review and meta-analysis.

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Accurate detection of brain metastasis (BM) is crucial for cancer patient management.
• Machine learning (ML) approaches applied to MRI show promise for BM detection.

Purpose of the Study:

• To systematically review the performance and quality of ML-based BM detection on MRI.
• To compare classical machine learning (cML) and deep learning (DL) for BM detection.

Main Methods:

• Systematic literature search up to April 27, 2020.
• Quality assessment using modified QUADAS-2 and CLAIM criteria.
• Pooled detectability calculated using an inverse-variance weighting model.

Main Results:

• Twelve studies were included, showing a shift from cML to DL post-2018.
• Pooled BM detectability was 88.7% for cML and 90.1% for DL.
• DL demonstrated a significantly lower false-positive rate per person (10 vs 135) compared to cML.

Conclusions:

• Comparable BM detectability was observed between cML and DL methods.
• Deep learning models offer a reduced false-positive rate for BM detection.
• Enhancements in study design and overall quality are needed for ML-based BM detection.