Can machine learning models improve early detection of brain metastases using diffusion weighted imaging-based...

Area of Science:

• Radiology
• Machine Learning
• Oncology

Background:

• Brain metastases are a significant cause of cancer morbidity, affecting up to 40% of patients.
• Early detection of brain metastases is crucial for improving patient outcomes and survival rates.
• Diffusion-weighted imaging (DWI) radiomics offers a potential avenue for earlier detection.

Purpose of the Study:

• To investigate the efficacy of machine learning (ML) models for the early detection of brain metastases.
• To utilize diffusion-weighted imaging (DWI) radiomics for identifying brain metastases.
• To develop and validate ML models for differentiating metastatic tissue from healthy brain tissue.

Main Methods:

• Retrospective analysis of longitudinal diffusion imaging from 116 patients treated for brain metastases.
• Extraction of radiomic features from apparent diffusion coefficient (ADC) maps and analysis of feature changes over time.
• Training and validation of four classification algorithms (SVM, Random Forest, AdaBoost, XGBoost) using radiomic, clinical, and anatomical data.

Main Results:

• Machine learning models achieved up to 87.7% accuracy on the training set and 85.8% on an unseen test set.
• XGBoost and Random Forest models demonstrated superior performance, with high accuracy and AUC scores on validation sets.
• Key features identified included radiomic changes occurring months before metastases were visible on contrast-enhanced T1-weighted imaging.

Conclusions:

• Diffusion-based radiomics and ML models show encouraging potential for early brain metastasis detection.
• These findings suggest that longitudinal diffusion imaging and ML can enhance patient care through earlier diagnosis and monitoring.
• Future research will focus on improving model performance, robustness, and clinical applicability.