Radiomics and machine learning for osteoporosis detection using abdominal computed tomography: a retrospective...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Osteoporosis diagnosis typically requires dedicated bone density scans.
• Abdominal CT scans are common, offering potential for incidental findings.

Purpose of the Study:

• To develop and validate a predictive model for osteoporosis detection.
• Utilize radiomic features from lumbar spine CT images within abdominal CT exams.
• Employ machine learning (ML) approaches for osteoporosis prediction.

Main Methods:

• Retrospective analysis of 509 patients from two centers.
• Extraction of radiomic features from lumbar spine CT images.
• Construction and evaluation of seven ML models (LR, Bernoulli, Gaussian NB, SGD, decision tree, SVM, KNN) using AUC and DCA.

Main Results:

• Logistic Regression (LR) model showed excellent performance (AUC 0.960) in internal validation for differentiating osteoporosis from normal BMD and osteopenia.
• LR and Gaussian NB models achieved high AUCs (0.905 and 0.839) in differentiating normal BMD from osteopenia and osteoporosis.
• LR model demonstrated superior net benefit in differentiating osteoporosis from normal BMD and osteopenia.

Conclusions:

• Radiomic-based ML models can predict osteoporosis from abdominal CT images.
• This methodology presents a viable option for opportunistic osteoporosis screening.
• Leveraging existing CT scans can enhance early detection and management of osteoporosis.