Peripheral quantitative computed tomography (pQCT) is a non-invasive and relatively safe imaging technique for measuring volumetric bone mineral density (vBMD) and predicting fracture risk. However, its limited availability restricts its practical use. This study aimed to develop an Artificial Intelligence (AI) system to predict vBMD from standard X-rays. The AI system was developed using data from 2539 individuals and tested on 666 participants from the Vietnam Osteoporosis Study. Frontal digital X-rays of the knee were obtained using the FCR Capsula XLII (Fujifilm Corp., Tokyo, Japan). vBMD was measured at the tibia using the XCT 2000 (Stratec Medizintechnik, Pforzheim, Germany). The deep learning models predicted vBMD at the distal (4%) and proximal (66%) tibia. The predicted vBMD was termed 'xBMD'. The correlation between xBMD and vBMD was assessed using correlation coefficients, and the AI's ability to classify vertebral fractures based on the Genant semiquantitative method was evaluated using the area under the curve (AUC). At the distal tibia, xBMD showed a strong correlation with vBMD, with a correlation coefficient of 0.88 (95% CI, 0.87 to 0.89) for total bone and 0.86 (95% CI, 0.84 to 0.88) for trabecular bone. At the proximal tibia, the correlation coefficients were 0.81 (95% CI, 0.78 to 0.83) for total bone and 0.78 (95% CI, 0.75 to 0.81) for cortical bone. The AI system effectively identified grade ≥ 2 fractures in women (AUC: 0.85 [95% CI, 0.74 to 0.95]), while the performance was modest in men (AUC: 0.59 [95% CI, 0.50 to 0.68]). The findings suggest that predicting vBMD from plain radiographs is feasible, and the developed AI system can extend the use of pQCT in resource-limited settings.