The present study revealed that the AI tool is capable of effectively detecting bone metastases on CT scans, showing comparable performance to both expert and general radiologists. The algorithm demonstrated a lesion-wise sensitivity of 0.82 at one false positive per scan. Importantly, sensitivities remained high for both sclerotic and osteolytic lesion types. 

These results align with previously reported AI sensitivities of 0.79-0.92, but achieved lower false positive rates. For example, while Chmelik et al. (2018) reported 0.80 and 0.92 sensitivities for osteolytic and sclerotic spinal metastases respectively, their false positive rates of 1.6-3.4 per vertebra were significantly higher than in our study [12]. 

Our study’s comprehensive approach, examining both types of metastases across all bones, addresses limitations of previous research that often focused on single morphologies or specific anatomical regions. However, the generalizability of the present findings is constrained by the small sample size and retrospective design.

In conclusion, this novel AI solution demonstrates excellent diagnostic capabilities across diverse lesion types, matching the performance of experienced radiologists in bone metastasis detection. By providing a robust safety net for both cancer follow-up and incidental metastasis identification, AI could significantly optimize radiological workflows. Future research should explore the performance of AI-assisted clinicians and examine metastasis detection across a broader spectrum of lesion sizes.