Comparison of radiomic feature aggregation methods for patients with multiple tumors

Area of Science:

• Oncology
• Radiology
• Medical Imaging Analysis

Background:

• Radiomic feature analysis shows promise in predicting cancer outcomes from diagnostic images.
• Optimal methods for combining radiomic features in patients with multifocal tumors are not well-established.
• Rising incidence of multifocal metastatic cancer necessitates improved personalized prognosis.

Purpose of the Study:

• To compare six mathematical methods for combining radiomic features from multiple brain metastases.
• To evaluate the performance of these aggregation methods in predicting patient-level outcomes.

Main Methods:

• Analysis of radiomic features from 3,596 tumors across 831 patients with multiple brain metastases.
• Comparison of six tumor feature aggregation methods.
• Evaluation using three survival models: Cox proportional hazards, Cox with LASSO regression, and random survival forest.

Main Results:

• The weighted average of the largest three metastases achieved the highest concordance index across all survival models.
• Concordance indices ranged from 0.627 to 0.652, indicating effective prediction.
• This finding remained consistent across varying numbers of metastases and tumor volumes.

Conclusions:

• Radiomic features can be effectively combined to estimate patient-level outcomes in multifocal brain metastases.
• The volume-weighted average of the largest three tumors is a promising aggregation method.
• Further research is needed to validate this method across different imaging modalities and cancer types.