Machine Learning Model Based on Optimized Radiomics Feature from ¹⁸F-FDG-PET/CT and Clinical Characteristics Predicts...

Area of Science:

• Nuclear medicine imaging
• Radiomics analysis
• Oncology

Background:

• Multiple myeloma (MM) is a hematologic malignancy.
• Accurate prognostic prediction is crucial for managing newly diagnosed MM (NDMM) patients.
• Integrating diverse data sources can enhance prognostic models.

Purpose of the Study:

• To assess the prognostic value of radiomics features from ¹⁸F-FDG-PET/CT.
• To evaluate the added benefit of combining radiomics with clinical data and conventional metrics for NDMM prognosis.

Main Methods:

• Retrospective analysis of 98 NDMM patients' baseline clinical and ¹⁸F-FDG-PET/CT data.
• Development of three Cox regression models: radiomics alone (Rad Model), clinical data alone (Cli Model), and combined (Cli-Rad Model).
• Model performance evaluated using C-index and Net Reclassification Index (NRI).

Main Results:

• The combined Cli-Rad Model demonstrated superior prognostic performance compared to Rad Model and Cli Model individually.
• C-indices in the training cohort were 0.790 (Cli-Rad) vs. 0.675 (Rad) vs. 0.736 (Cli).
• The combination showed improved classification accuracy, indicated by NRI > 0 and significant AUC values in both training and validation cohorts.

Conclusions:

• Radiomics features from baseline ¹⁸F-FDG-PET/CT are valuable prognostic indicators for NDMM.
• Combining radiomics with clinical characteristics enhances prognostic prediction accuracy.
• This integrated approach holds clinical utility for predicting MM prognosis.