Radiomics feature analysis for survival prediction in multiple myeloma: An automated PET/CT approach

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• ¹⁸F FDG PET/CT imaging is an emerging tool for evaluating Multiple Myeloma (MM).
• Accurate risk stratification is crucial for optimizing MM patient management.

Purpose of the Study:

• To develop and evaluate an automated radiomics pipeline for MM risk stratification.
• To identify the most informative imaging regions and features for predicting progression-free survival in MM patients.

Main Methods:

• Utilized whole-body ¹⁸F FDG PET/CT images from 227 MM patients.
• Automated segmentation of skeletal structures (spine) and radiomic feature extraction.
• Trained and cross-validated various survival models (DeepSurv, Random Survival Forests) using radiomic features.

Main Results:

• The best performing model (DeepSurv) used PET image texture features from the spine and surrounding regions.
• Achieved a mean c-index of 0.657, indicating good prognostic performance.
• Image heterogeneity, particularly texture features, strongly correlated with increased risk of disease progression.

Conclusions:

• Automated radiomics analysis, especially using PET texture features from the spine, effectively stratifies MM risk.
• DeepSurv and Random Survival Forests models outperformed traditional Cox-based models.
• The study provides an open-source pipeline for reproducible radiomics analysis in MM.