Multimodal hip CT analysis combining radiomics, clinical data, and deep learning detects osteoporosis with high accuracy

In a study of 567 patients, integrating hip CT radiomics with clinical data via a GradientBoosting Nomogram model yielded the highest validation accuracy (0.849) and AUC (0.911) for opportunistic osteoporosis screening, outperforming 2D and 3D deep learning models alone.

• Among 2D deep learning models, DenseNet201 achieved a validation accuracy of 0.817 and AUC of 0.884; the best 3D model (3D ResNet34) reached an accuracy of 0.806 with AUC of 0.889.
• The Nomogram model integrated a radiomic signature with clinical variables such as age and gender using a GradientBoosting machine learning algorithm.
• The study was a single-institution retrospective development and internal validation; external or prospective validation was not performed.

Automated summary

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