Digitalization of surgical features improves surgical accuracy via surgeon guidance and robotization

Xiaoyun Chen, Jiali Liu, Hongli Liang, Zelin Chen, Zitian Liu, Yingfeng Zheng, Zhangkai Lian, Lixia Luo, Weirong Chen, Mingxing Wu, Danying Zheng, Xialin Liu, Bing Cheng, Shengsong Huang, Xinyu Zhang, Ling Jin, Yan Luo, Zhenzhen Liu, Erping Long, Patrick Yu-Wai-Man, Weishi Zheng, Kai Huang, Haotian Lin, Yizhi LiuNpj Digit. Medicine 2025 · DOI

Introduction

Cataract surgery is one of the most common surgical procedures worldwide. A critical step in this surgery is the capsulorhexis, which involves creating a circular opening in the lens capsule to allow for the insertion of an artificial lens. Traditionally, this maneuver relies heavily on a surgeon's subjective experience and manual dexterity. Inconsistencies in size, centration, or shape can lead to suboptimal visual outcomes or post-operative complications.

Conceptual overview of transitioning from human experience to digital experience in surgery.

Researchers have introduced Meta Surgery (MetaS), an AI-driven framework designed to standardize this "art" by converting surgical expertise into digital features. By analyzing thousands of surgical videos, the system provides objective guidance for surgeons and enables autonomous robotic performance.

Methodology: The MetaS Framework

The MetaS system operates through three integrated modules designed to extract and apply digital surgical features:

  1. Intelligent Evaluation: Utilizing an InceptionResNetV2 architecture, the system was trained to classify surgical quality into "ideal," "acceptable," or "poor" categories based on historical expert data.
  2. Feature Extraction: By applying Mask R-CNN to "ideal" cases, researchers identified a mathematical "golden ratio." They discovered that the ideal capsulorhexis radius is consistently 0.58 times the radius of the corneal limbus (the edge of the iris).
  3. Guidance and Robotization: These digital features are translated into practical applications. For human surgeons, this includes real-time AR-style overlays (using a lightweight GhostNet backbone) or a physical "Lens Caliper." For robotization, coordinates are fed into a micromanipulator for autonomous execution.
Detailed workflow of the MetaS system, from data input to feature extraction and robotic execution.

Data and Insights

The study leveraged a massive dataset of 17,538 surgical videos (selected from an initial pool of over 26,000) from the Zhongshan Ophthalmic Center. A key finding from this data was that even experienced surgeons only achieved "ideal" results in approximately 16.7% of traditional manual procedures, highlighting a significant ceiling for improvement through objective standardization.

Key Results

The implementation of MetaS showed dramatic improvements in surgical precision across all testing phases:

  • Surgeon Performance: Using the MetaS-derived Lens Caliper increased the "ideal" rate from 16.7% to 64.1%. Real-time digital projection further boosted a single surgeon's ideal rate from 44.2% to 85.0%.
  • Robotic Accuracy: The robotic system achieved Level 2 (task-level) autonomy, successfully performing the task on ex vivo porcine eyes with high consistency in diameter (5.3 mm) and centration.
  • AI Reliability: The evaluation module demonstrated high diagnostic accuracy, with Area Under the Curve (AUC) scores ranging from 0.917 to 0.963 for quality classification.

Conclusion and Future Directions

MetaS represents a significant step toward the digitalization and standardization of microsurgery. While the system shows immense promise in routine cases, future work is required to address patient variability—such as small pupils or complicated cataracts—and to integrate guidance hardware directly into surgical microscope eyepieces. Transitioning from porcine models to human clinical trials remains the essential next step for validating long-term safety and visual outcomes.

Summary updated

BibTeX
@article{chen2025digitalization,
  author = {Xiaoyun Chen and Jiali Liu and Hongli Liang and Zelin Chen and Zitian Liu and Yingfeng Zheng and Zhangkai Lian and Lixia Luo and Weirong Chen and Mingxing Wu and Danying Zheng and Xialin Liu and Bing Cheng and Shengsong Huang and Xinyu Zhang and Ling Jin and Yan Luo and Zhenzhen Liu and Erping Long and Patrick Yu-Wai-Man and Weishi Zheng and Kai Huang and Haotian Lin and Yizhi Liu},
  title = {{Digitalization of surgical features improves surgical accuracy via surgeon guidance and robotization}},
  journal = {Npj Digit. Medicine},
  year = {2025},
  doi = {10.1038/s41746-025-01887-6},
}
Figure