BoneCoT AI model distinguishes primary from metastatic bone lesions with 40% AUC improvement over radiologists in multicenter study

BoneCoT AI improved AUC for distinguishing primary vs metastatic bone lesions by 40% over radiologists (multi-center, 10 hospitals), with 20% AUC gain over prior methods across 26 tasks. Trained on 29.3M CT images.

• Development and multi-center validation across 10 hospitals, pretrained on 29.3 million CT slices from 30,267 patients.
• BoneCoT outperformed existing AI models by 20% AUC across 26 tasks, and improved primary vs. metastatic distinction by 40% AUC over experienced radiologists.
• Limitation: performance has not been assessed in a real-world prospective clinical setting.

RadPigeon summaries are original and for information only. They are not clinical advice.