To evaluate the feasibility of fixed-horizon surgical step prediction in recorded microscope videos of middle cerebral artery aneurysm clipping operations, highlighting the potential impact of AI in enhancing surgical precision.
Key Findings:
The multimodal model achieved the highest mean accuracy of 0.683 and weighted F1 score of 0.673, indicating strong predictive capabilities.
The annotation-only model had a mean accuracy of 0.606 and weighted F1 score of 0.577, suggesting the importance of prior knowledge.
The video-only model had a mean accuracy of 0.477 and weighted F1 score of 0.447, highlighting the limitations of using video features alone.
The multimodal model showed the best sequence-level alignment with a normalized edit distance of 0.430, reinforcing its superiority in predictive performance.
Interpretation:
Fixed-horizon surgical step prediction during MCA aneurysm clipping is feasible under controlled input conditions, with multimodal modeling providing the strongest predictive performance, which could enhance surgical training and decision-making.
Limitations:
The study represents upper-bound performance and requires validation in fully automated recognition-to-prediction pipelines, as well as consideration of potential biases in the data.
The analysis was limited to a small sample size of 25 surgeries, which may affect the generalizability of the findings.
Conclusion:
The findings indicate the potential for improved predictive modeling in surgical procedures, suggesting avenues for future research to validate and expand upon these results.
