Bone Metastasis Detection at CT with Deep Learning ModelsTrained Using Multicenter, Multimodal Reference Standards: Development and Evaluation

This study aimed to develop and evaluate deep learning models for automated bone metastasis detection on CT imaging. Models were trained using data collected from a multicenter setting, with bone scan and PET/CT serving as multimodal reference standards to generate ground-truth annotations. AVIEW Research was utilized for CT image data processing and analysis pipeline construction. The developed models achieved high sensitivity and specificity on both internal and external test sets, with performance quantified using AUC metrics. Training with multicenter, multimodal reference standards was shown to improve model generalizability across diverse patient populations and scanner types. The results demonstrate the potential of deep learning-based bone metastasis detection as a clinical decision-support tool that can assist radiologists in identifying osseous metastatic lesions on CT, potentially improving diagnostic efficiency and consistency in oncologic imaging workflows.