5 Key Takeaways

• 1

  LensNarrate analyzes cataract surgery video, classifies actions into 12 surgical phases, and generates a temporally ordered narrative report.

• 2

  Phase-recognition accuracy was 83.2% in 20 recordings from Poland and Ukraine and 71.6% in 10 recordings from China.

• 3

  Temporal ordering increased agreement between recognized surgical segments and the reference record from 6.5% to 58.0%.

• 4

  Ophthalmologists rated LensNarrate slightly better than Qwen2.5-VL and VideoCLIP, but overall clinical acceptability remained relatively low.

• 5

  LensNarrate can generate unsupported details, so the authors describe physician review and approval as necessary before reports are finalized.