Unsupervised Clustering of Vertebral Hounsfield Units in Opportunistic Chest CT for Stratifying Bone Mass Subtypes in a 2 Years' Period.

There is an urgent need for a convenient and incidental method to assess the bone health status of the population, especially in primary-level hospitals lacking specialized bone density testing equipment. This study aims to investigate the association between multiple vertebral Hounsfield Unit (HU) value clusters and bone mass subtypes using an unsupervised learning approach, providing a practical tool for incidental osteoporosis screening in clinical settings. This retrospective study included subjects who underwent chest CT and quantitative CT (QCT) from January 2023 to December 2024. Vertebral HU values (T7-T12) were measured on chest CT images. Intergroup comparisons (normal, osteopenia, and osteoporosis) in clinical findings and CT values were performed using Pearson χ2 test and one-way analysis of variance. An unsupervised k-means clustering was applied to vertebral CT values across the cohort. The study comprised 455 participants (260 males, 195 females) with a median age of 60 years (interquartile range, 51-67 years), who were classified into three groups: normal bone mass, 253 cases; osteopenia, 152 cases; osteoporosis, 50 cases. Among 455 participants, age inversely correlated with bone mass. Vertebrae HU values (T7-T12) exhibited significant stepwise declines from normal to osteopenia to osteoporosis (OP) groups. The clustering analysis revealed five distinct subtypes: cluster 1 strongly correlated with OP (45 of 72 cases), cluster 4 with osteopenia (107 of 146 cases), and clusters 2, 3, and 5 with normal bone mass (31 of 31 cases; 90 of 107 cases; 97 of 99 cases). Unsupervised clustering of T7-T12 vertebral HU values effectively stratifies bone mass subtypes, offering an efficient, CT-based screening method for skeletal health assessment, especially valuable in resource-limited primary-level hospitals lacking dedicated bone densitometry.