Automatic segmentation and radiomic texture analysis for osteoporosis screening using chest low-dose computed...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Bone density screening is crucial for early detection of osteoporosis.
• Current screening methods like dual-energy X-ray absorptiometry (DXA) have limitations.
• Low-dose computed tomography (LDCT) is widely used for lung cancer screening.

Purpose of the Study:

• To develop and validate a diagnostic tool for bone density screening using LDCT.
• To combine machine learning (ML) segmentation and radiomic texture analysis (RTA) for automated bone density assessment.
• To enable opportunistic bone density screening during routine LDCT scans.

Main Methods:

• Developed an autosegmentation model for thoracic vertebral body (VB) delineation on LDCT.
• Extracted radiomic features from VBs for a two-level hierarchical classifier.
• Classified patients into normal, osteopenia, and osteoporosis groups.
• Evaluated classifier performance using fivefold cross-validation.

Main Results:

• Automated VB segmentation achieved a 0.87 ± 0.01 Sorenson-Dice coefficient.
• The two-level classifier demonstrated high performance: AUC of 0.96 ± 0.01 for abnormal bone density and 0.98 ± 0.01 for osteoporosis.
• Overall testing prediction accuracy reached 0.90 ± 0.05.

Conclusions:

• ML segmentation and RTA provide a feasible approach for automated bone density prediction from LDCT.
• This method enables opportunistic screening, potentially integrating into lung cancer screening workflows.
• The tool can serve as an adjunct for patients ineligible for DXA.