Multimodal model combining CT and clinical data predicts 90-day respiratory failure in DM-ILD without anti-MDA5 antibody results

A multimodal early-fusion random forest model combining admission CT and clinical data (excluding anti-MDA5 antibodies) predicted 90-day respiratory failure in dermatomyositis-associated interstitial lung disease (DM-ILD), achieving AUC 0.967 in internal testing.

• SHAP analysis identified arthritis, pulmonary function indices, laboratory markers, and latent CT features as the most influential predictors.
• Model was built without anti-MDA5 antibody results, targeting settings where testing is unavailable or delayed.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.