Radiomic-based approaches in the multi-metastatic setting: a quantitative review

Area of Science:

• Medical imaging analysis
• Quantitative imaging
• Radiomics

Background:

• Traditional radiomics focuses on single lesions, potentially missing inter-lesion heterogeneity in metastatic disease.
• No standardized methods exist for combining radiomic features from multiple lesions, leading to diverse and inconsistent approaches.
• This review addresses the need for best practices in multi-lesion radiomic analysis across varied datasets.

Purpose of the Study:

• To quantitatively review and assess methodologies for integrating radiomic data from multiple lesions.
• To evaluate the replicability of existing multi-lesion radiomic analysis methods.
• To provide insights and guide future research toward establishing best practices in this domain.

Main Methods:

• Conducted a comprehensive literature search to identify methods for multi-lesion radiomic data integration.
• Replicated identified methods using author's code or paper-based reconstruction.
• Applied and compared these methods across three distinct metastatic datasets.

Main Results:

• Compared ten mathematical methods for combining radiomic features across 16,894 lesions in 3,930 patients.
• Performance varied by dataset and lesion burden; no single method was universally superior.
• Averaging methods excelled in colorectal liver metastases, concatenation in soft tissue sarcoma, and total tumor volume in head and neck cancers.

Conclusions:

• Effective selection or combination of radiomic features can predict outcomes in multi-metastatic patients.
• The optimal approach for multi-lesion radiomic analysis is dependent on the specific metastatic setting.
• This study provides critical insights into the challenges and effective strategies for radiomic analysis in multi-metastatic disease.