• Yunji Seol, Jin Ho Song, Kyu Hye Choi, Young Kyu Lee, Byung-Ock Choi, and Young-Nam Kang.
• Department of Biomedicine and Health Sciences, The Catholic University of Korea, 222, Banpo-daero, Seocho-gu, Seoul, Korea.
• Eur Spine J. 2024 Aug 1; 33 (8): 322132293221-3229.

PurposeThe purpose of the study was to develop a predictive model for vertebral compression fracture (VCF) prior to spinal stereotactic body radiation therapy (SBRT) using radiomics features extracted from pre-treatment planning CT images.MethodsA retrospective analysis was conducted on 85 patients (114 spinal lesions) who underwent spinal SBRT. Radiomics features were extracted from pre-treatment planning CT images and used to develop a predictive model using a classification algorithm selected from nine different machine learning algorithms. Four different models were trained, including clinical features only, clinical and radiomics features, radiomics and dosimetric features, and all features. Model performance was evaluated using accuracy, precision, recall, F1-score, and area under the curve (AUC).ResultsThe model that used all features (radiomics, clinical, and dosimetric) showed the highest performance with an AUC of 0.871. The radiomics and dosimetric features model had the superior performance in terms of accuracy, precision, recall, and F1-score.ConclusionThe developed predictive model based on radiomics features extracted from pre-treatment planning CT images can accurately predict the likelihood of VCF prior to spinal SBRT. This model has significant implications for treatment planning and preventive measures for patients undergoing spinal SBRT. Future research can focus on improving model performance by incorporating new data and external validation using independent data sets.© 2023. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.

31 keywords...

hide…