Predicting vertebral fractures and bone density from chest X-rays using artificial intelligence

Abstract

In Japan, the number of patients with osteoporosis is increasing as the population ages, but the therapeutic intervention rate is not high. For efficient and effective interventions, we hypothesized that the evaluation of osteoporosis can be performed (1) without special equipment and (2) without changing the conventional flow of medical care, thereby helping to solve this social problem. Focusing on chest X-rays, which are common and frequently taken, two AI algorithms to evaluate osteoporosis from chest X-rays. The first AI algorithm was developed to predict vertebral fractures from chest X-rays to detect vertebral fractures. A total pf 5,791 chest X-rays were used, and the AI algorithm predicted the presence of vertebral fractures in chest X-rays with Area Under the Curve = 0.98. The second AI algorithm was developed to predict bone mineral density from chest X-rays to evaluate bone mass. A total of 17,899 chest X-rays and bone mineral density measurements were used, and the AI algorithm predicted bone mineral density with a correlation coefficient (R) of 0.75. The AI algorithms developed in this study were able to detect vertebral fractures and evaluate bone mineral density with high accuracy from chest X-rays. Accurate detection and evaluation of these factors are important aspects of planning appropriate interventions for osteoporosis.