Back to all papers

Interpretable machine learning for bone mineral density regression via computed tomography radiomics: A retrospective cross-sectional study.

November 18, 2026pubmed logopapers

Authors

Yao L,Zhang Z,Cheng L,Yu Y

Affiliations (2)

  • Division of Spine, Department of Orthopaedics, Tongji Hospital, School of Medicine, Tongji University, China.
  • Department of General Practice, Yangpu Hospital, School of Medicine, Tongji University, China.

Abstract

ObjectiveThe limited interpretability and generalizability of machine learning models for bone mineral density regression hinder their clinical application. This study aimed to develop and evaluate a machine learning model for lumbar bone mineral density regression using computed tomography radiomics and to assess its interpretability through Shapley Additive exPlanations analysis.MethodsIn this retrospective, single-center, cross-sectional study, 150 patients who underwent routine abdominal computed tomography and dual-energy X-ray absorptiometry scans between April 2018 and November 2023 were included. Lumbar vertebral bodies were segmented using 3D-slicer, followed by radiomics feature extraction. Machine learning models, including EXtreme gradient boosting, light gradient boosting machine, gradient boosting decision tree, support vector machine, and random forest, were used for bone mineral density regression and T-score classification. Shapley Additive exPlanations analysis was used to assess global and local interpretability.ResultsThe EXtreme gradient boosting algorithm showed the best performance for bone mineral density regression, with an R<sup>2</sup> of 0.80, mean absolute error of 0.07, mean squared error of 0.01, and Pearson's r of 0.90. The T-score classification achieved a weighted average precision, recall, and F1 score of 0.75, 0.71, and 0.72, respectively. Shapley Additive exPlanations analysis provided global insights into feature contributions and local interpretability for individual samples.ConclusionsThis study introduced an interpretable method using routine computed tomography scans and the EXtreme gradient boosting algorithm to predict lumbar bone mineral density. The top radiomic features identified through Shapley Additive exPlanations analysis, such as Coarseness and RunEntropy, may contribute to accurate bone mineral density prediction.

Topics

Bone DensityTomography, X-Ray ComputedMachine LearningLumbar VertebraeJournal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAISlice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.