 This paper proposes a novel approach for decomposing musculoskeletal structures from radiographs into multiple individual muscle and bone structures. It utilizes a cycle GAN framework to solve the problem by translating between two datasets, one containing real X-ray images and another containing digitally reconstructed radiographs. Additionally, it introduces two features to improve the accuracy of the decomposition. Hierarchical learning and reconstruction loss with the gradient correlation similarity metric. The results of this study suggest that the proposed method can accurately decompose complex musculoskeletal structures from a single radiograph. Furthermore, it can measure the volume ratio of muscles, suggesting its potential application to muscle asymmetry assessment from an X-ray image. This article was authored by Naoki Nokonishi, Yoshito Otaik, Yuta Hayasa, and others. We are article.tv, links in the description below.