Clinical Scorecard: AI in the Operating Room

At a Glance

Key Highlights

• In 20 recordings from Poland and Ukraine, LensNarrate identified the surgical phase in 83.2% of analyzed frames.
• Temporal ordering increased agreement between recognized surgical segments and the reference record from 6.5% to 58.0%.
• Phase-recognition accuracy was 71.6% in ten recordings from Ningbo Eye Hospital, where instruments, materials, and technique details differed.
• In blinded assessment by three ophthalmologists, LensNarrate performed slightly better than Qwen2.5-VL and VideoCLIP, but clinical-acceptability ratings remained relatively low.
• The system sometimes generated unsupported details, including implanted lens specifications, medication composition, and incision angle.

Guideline-Based Recommendations

Diagnosis

Management

• The article describes LensNarrate as a support tool for preparing a draft operative report, not for independently generating final documentation.
• The article describes a human–AI workflow in which the surgeon verifies the report, adds missing information, and approves the final documentation.

Monitoring & Follow-up

• The article identifies larger multicenter studies and prospective evaluation of documentation time and omitted details as next steps.

Risks

• The article reports that the system may generate details that cannot be reliably determined from video or repeat patterns from training data that do not match the procedure.
• Phase-recognition accuracy was lower in the China dataset than in the Poland and Ukraine recordings.

Patient & Prescribing Data

Not reported; the study evaluated cataract surgery videos rather than patient treatment outcomes.

No treatment or prescribing recommendations are reported. LensNarrate generates draft documentation from surgical video and requires physician review.

Clinical Best Practices

• The article describes generated operative reports as drafts requiring surgeon verification.
• The article reports that additional models highlight instruments and ocular structures so physicians can verify the visual basis of generated descriptions.
• The article identifies multicenter studies and prospective evaluation of documentation time and omitted details as next steps.

Related Resources & Content