A multimodal fusion model integrating Vision Transformer, radiomics, and clinical features for predicting bone metastasis in prostate cancer - Report - MDSpire
Advertisement
A multimodal fusion model integrating Vision Transformer, radiomics, and clinical features for predicting bone metastasis in prostate cancer
Clinical Report: A Comprehensive Fusion Approach for Predicting Bone Metastasis
Overview
This study evaluates a multimodal fusion framework that integrates Vision Transformer, radiomics, and clinical data to predict bone metastasis in prostate cancer patients. The fusion model demonstrated superior performance compared to single-modal models.
Background
Prostate cancer is a leading cause of cancer-related mortality in men, with bone metastasis being the most common form of distant spread. Accurate prediction of bone metastasis is crucial for guiding treatment strategies, especially since many patients remain asymptomatic in early stages. Current diagnostic methods have limitations.
Data Highlights
Model
Training AUC
Validation AUC
Model_ViT
0.909
0.872
Model_Rad
0.885
0.842
Model_Clin
0.861
0.781
Model_Fusion
0.944
0.894
Key Findings
Model_ViT outperformed both Model_Rad and Model_Clin in predicting bone metastasis.
Model_Fusion achieved the highest AUC of 0.944 in the training set and 0.894 in the validation set.
DeLong test indicated significant performance improvement of Model_Fusion over single-modal models in the training set (P < 0.05).
Decision curve analysis showed Model_Fusion provided a higher net benefit compared to single-modal models.
SHAP analysis revealed that the predicted probability from ViT was the most influential factor in Model_Fusion.
Clinical Implications
The integration of multimodal data through the fusion model may enhance the accuracy of bone metastasis predictions in prostate cancer patients.
Conclusion
The study demonstrates that a fusion model combining Vision Transformer, radiomics, and clinical features offers a promising non-invasive method for predicting bone metastasis in prostate cancer.
