Radiomics based on BPX images to detect abnormal bone mineral density.

Abstract

Abnormal bone mineral density (BMD) is a major contributor to bone fragility and fractures. While dual-energy X-ray absorptiometry (DXA) and quantitative computed tomography (QCT) are the primary diagnostic modalities, both methods are associated with additional radiation exposure and costs. This study investigates the feasibility of using radiomics to establish an automated tool for identifying patients at high risk for BMD abnormality based on biplanar X-ray radiography (BPX) images. A total of 906 BPX scans from 453 subjects, including 275 females, were included in this prospective study, with QCT results as the ground truth (GT). Radiomic features were extracted from the anteroposterior and lateral views of the L1-L5 vertebrae using Pyradiomics. The most relevant features were selected using the least absolute shrinkage and selection operator (LASSO) filter. Then, radiomics-only and clinical-radiomics models were established to diagnose BMD abnormality. The performances of the models were calculated and evaluated. The radiomics-only model achieved an accuracy (ACC) of 0.88 for both sexes and 0.91 for women. The clinical-radiomics models achieved an ACC of 0.87 for both sexes and 0.89 for women. The radiomics-only model based on BPX images effectively distinguishes BMD abnormality and demonstrates potential as a decision-support tool in real-world physical examination populations.