Deep learning ensembles for detecting brain metastases in longitudinal multi-modal MRI studies

Area of Science:

• Neuro-oncology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Metastatic brain tumors are more prevalent than primary brain tumors and exhibit aggressive growth.
• Manual analysis of MRI scans for brain metastases is challenging, subjective, and lacks reproducibility.
• Current methods struggle with the heterogeneity and complexity of metastatic brain lesions seen in MRI.

Purpose of the Study:

• To develop and validate an automated pipeline for detecting and analyzing brain metastases in longitudinal MRI studies.
• To improve the accuracy and reproducibility of brain metastasis assessment.
• To provide physicians with a quantitative tool for tracking disease progression and treatment efficacy.

Main Methods:

• An ensemble of deep learning architectures (detection and segmentation) was utilized.
• The pipeline was trained and validated on 275 multi-modal MRI scans from 87 patients.
• A novel data stratification approach and quality metrics were introduced for robust model evaluation.

Main Results:

• The proposed pipeline demonstrated high-quality detection of brain metastases.
• The system accurately tracked disease progression in longitudinal MRI studies.
• The open-source, automated approach enhances reproducibility and physician support.

Conclusions:

• The developed deep learning pipeline offers a fully automatic and quantitative solution for brain metastasis analysis.
• This system can significantly support clinicians in the laborious process of disease monitoring and treatment assessment.
• The approach addresses the limitations of manual MRI analysis, improving efficiency and reliability.