胸部/腹部CT画像における全包含的椎体骨折検出用ソフトウェア「Ofeye 3.0」の開発と多施設検証

Missing report for fragility vertebral fracture (VF) on chest/abdominal CT is common when the indication is not spine disorders. This represents a missed opportunity to alert patients to take preventive measures to improve bone health and prevent further severe fractures. In this study, we aim to develop a software program for automated detection of VF with existing chest/abdominal CT scans and validate its detection performance.

An automated sagittal Central Slab reconstruction (CSR) method for CT axial images was developed. For reference VF reading, VFs inclusive of those of with <20 % vertebral height loss and those of endplate fracture with minimal vertebral height loss were identified. VFs were also differentiated from osteoarthritic wedging and endplatitis short vertebrae. Prior knowledge of VF detection models for lateral radiograph were transferred to a new 'Ofeye 3.0' model optimized for VF detection on CT image. Training CT images were obtained from nine centers, totaling 3313 cases without VF and 835 cases with VF. For external validation, CT images were from five centers totaling 732 cases without VF and 224 cases with VF.

The automated CSR method showed advantages in demonstrating structural changes of the endplate and adjacent structures. For detecting VF in chest/abdominal CT scans, counting case-by-case and compared with the reference reading, the average performance of Ofeye 3.0 was accuracy 0.967, sensitivity 0.906, and specificity 0.986. Most of false negative or false positive cases were minimal or mild VF, or with image artifacts, or with VF close to the peripheral of CSR images.

Despite the challenging requirements for the software to detect all-inclusive VF, our results compare favorably with other published automated VF detection models.

We developed a software program for automated all-inclusive VF detection on chest and/or abdominal CT image data and conducted a multi-center external validation study. This software is proved to have high VF detection precision. By alerting patients of the VFs likely related to osteoporosis and in turn the patients taking measures to prevent further fracture, the integration of this software into radiological practice will improve patient outcomes and reduce healthcare costs.