Research on predicting the progression of multiple myeloma treated with bortezomib based on multimodal ensemble learning
This multimodal ensemble model effectively predicts MM progression by integrating diverse diagnostic data.
The study developed a multimodal ensemble model integrating bone marrow smears, electrophoresis images, and baseline clinical data to predict 2‑year progression in 207 newly diagnosed multiple myeloma (NDMM) patients receiving bortezomib. Neural network back‑bones (ResNet, MobileNet, VGG16, DenseNet) extracted image features, while LASSO‑selected clinical features were modeled with Random Forest and Logistic Regression. The best models from each modality were fused using soft‑voting ensemble. The ensemble achieved an AUC of 0.8180 and accuracy of 0.7000, outperforming all single‑modality models. Electrophoresis‑image VGG16 (AUC 0.8082) and DenseNet (AUC 0.8088) were the strongest single modalities, followed by ResNet bone‑marrow (AUC 0.7295) and Logistic Regression clinical (AUC 0.6779). The approach offers a practical, cost‑effective decision‑support tool for clinicians to personalize treatment strategies in real‑world settings.
Accurate early prediction of progression in multiple myeloma patients treated with bortezomib can guide personalized treatment intensity, improve long‑term survival, and reduce unnecessary exposure to ineffective therapies.
Evidence level: Feltételezés / hírjelzés. Nincs önálló tudományos bizonyíték.
Related signals
- Multiple Myeloma Clinical Trials & Emerging Therapies
- S2209: Comparing Combinations of Drugs to Treat Newly Diagnosed Multiple Myeloma (NDMM)
- Improved survival in multiple myeloma following prior detection of precursor conditions: a nationwide real-world study
- Nanomedicine-Empowered CAR‑T Therapy for Multiple Myeloma: Toward Programmable, Durable, and Precision Immunotherapy
- Reaching Consensus on Defining, Treating High‑Risk Myeloma