AI dataset targets nerve injury risk during thyroid surgery 

Objective:

To demonstrate the feasibility of AI-based identification of the recurrent laryngeal nerve (RLN) during endoscopic thyroidectomy using a curated dataset.

Key Findings:

• The RLN was confirmed in all cases, with a dataset capturing real-world surgical variability.
• DeepLabV3+ achieved 64% recall and 65% precision; Mask2Former achieved 67% recall and 84% precision.
• The transformer-based model showed significant improvement in precision but modest gains in recall.

Interpretation:

The study highlights the potential of AI in enhancing the identification of anatomical structures during thyroid surgeries, despite challenges in detecting small or obscured nerves.

Limitations:

• The dataset was derived from a single institution with only 28 cases and limited surgical approaches.
• Generalizability may be affected due to the use of a single imaging system.
• Excluded frames during quality filtering were not quantified, and model performance indicates room for improvement.

Conclusion:

The dataset lays a foundation for future AI-assisted surgical models and has implications for surgical education and clinical practice.