Multimodal Radiomics and Explainable AI for Osteoporosis Risk Assessment from Routine CT Scans

Roshan Bharathi R,    Nitin Kumar M V

(312423104501) ,       (312423104155)

St.Joseph’s Institute of Technology OMR, Chennai – 600119

Guide: Gowri A

ABSTRACT:Osteoporosis remains severely underdiagnosed despite affecting over 200 million people worldwide, largely because the gold-standard diagnostic tool — dual-energy X-ray absorptiometry (DXA) — is inaccessible in resource-constrained settings. Routine Computed Tomography (CT) scans, performed daily for unrelated indications, present an untapped opportunity for opportunistic screening. We present OsteoRadXAI, an end-to-end multimodal framework that repurposes existing clinical CT acquisitions for automated osteoporosis risk assessment. The pipeline integrates (1) nnU-Net–based vertebral segmentation, (2) IBSI-compliant high-dimensional radiomic feature extraction, and (3) attention-based multimodal fusion of imaging biomarkers with clinical covariates. To bridge the clinical-trust gap, Explainable AI (XAI) techniques — SHAP, LIME, and Grad-CAM — provide transparent feature-level and pixel-level explanations. Validated on 2,268 CT scans across multiple datasets, OsteoRadXAI achieves an AUC of 0.91 ± 0.03 and a mean absolute error (MAE) of 21.3 ± 3.1 mg/cm² for bone mineral density (BMD) estimation on external validation, substantially outperforming HU-thresholding baselines. A web-based clinical prototype with DICOM integration delivers reports within 3.8 minutes per scan. A clinician user study (n = 20) rated the XAI explanations at 4.0/5.0, indicating high acceptability.Keywords: Osteoporosis; Opportunistic Screening; Radiomics; Explainable AI; CT Imaging; Bone Mineral Density; Deep Learning; SHAP; Vertebral Segmentation; Multimodal Fusion