Time-series CT radiomics deep learning model predicts post-SBRT rib fractures in NSCLC

A deep learning model using longitudinal CT radiomics predicted rib fracture risk after stereotactic body radiotherapy in NSCLC patients, achieving an AUC of 0.792 on validation. The model-predicted high-risk group had a hazard ratio of 10.82 compared to the low-risk group.

• The KA-TMoE model integrated three timepoint CT scans with radiomic grouping knowledge across 67 patients and over 1600 individual ribs.
• Model output remained independently associated with fracture risk in multivariable logistic analysis (odds ratio 12.05; p=0.002).
• Decision curve analysis demonstrated net benefit, and time-to-event analysis confirmed high-risk classification (HR 10.82; p<0.001).

Automated summary

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