AI Breakthrough Brings Automated Video Analysis to Cataract Surgery in Low-Resource Settings

Automated Phase Detection AI Video Analysis for Improved Cataract Surgery

Manual small incision cataract surgery (SICS) remains a common practice across many low- and middle-income countries, yet a lack of publicly available surgical video datasets has hindered research and quality improvement efforts. Addressing this gap, an international research team from the Sankara Eye Foundation India, University Hospital Bonn (UKB), and the University of Bonn has developed the first AI-based automated phase detection system specifically for SICS. Their findings are published in the journal Scientific Reports.

The Importance of Surgical Phase Detection

Analyzing surgical phases through AI enables objective comparison between surgeons, offers feedback on critical steps, and helps identify deviations from standard procedures. This capability marks the initial move toward automated evaluation of surgical quality.

Cataracts are the leading cause of blindness worldwide, particularly affecting populations in countries such as India, where access to resources and training can be limited. SICS is a cost-effective and preferred surgical technique in these regions, but outcomes may suffer due to these constraints. While AI-supported video analysis has been developed for phacoemulsification—the dominant technique in wealthier countries—similar tools were lacking for SICS until now.

Introducing the SICS-105 Dataset and AI Model

The research team has made publicly available the SICS-105 dataset, consisting of videos from 105 manual small incision cataract surgeries performed at Sankara Eye Hospitals. Using this dataset, they trained an advanced deep learning model called MS-TCN++, created by researchers at the University of Bonn. This model can identify various surgical phases, including preparation and lens-related steps, with over 85% accuracy.

"The analysis of surgical phases is important because it enables a quantitative comparison between different surgeons, feedback on identified critical steps, and the detection of deviations from surgical protocols," explains Dr. Kaushik Murali, president of medical administration at Sankara Eye Foundation India.

Collaboration Across Disciplines

The project showcases a transdisciplinary approach. Simon Mueller, the study’s first author, balances medical studies while pursuing a PhD that bridges computer science and ophthalmology. This collaboration reflects the synergy between technical expertise and clinical practice essential for advancing AI applications in healthcare.

The SICS-155 Challenge: Advancing AI in Surgical Video Analysis

Building on initial success, the research consortium is launching the SICS-155 Challenge at the MICCAI 2025 conference in South Korea. This global AI competition invites teams to develop and test algorithms for surgical phase recognition using an expanded dataset of 155 annotated SICS operations covering 18 distinct phases.

The surgical video annotations were created by ophthalmologists at Sankara Eye Foundation with software developed by Microsoft Research India and Sankara Eye Foundation. Participants are expected to submit their algorithms alongside a short paper detailing their methods.

Prof. Dr. Thomas Schultz from the University of Bonn emphasizes that this competition aims to accelerate progress in surgical video analysis specifically for low- and middle-income countries. The ultimate goal is to improve surgeon training and surgical outcomes in cataract care.

Future Directions

In addition to phase detection, ongoing work by computer scientists at Microsoft Research India and the University of Bonn targets automated detection of surgical instruments and complications. These advances will further enhance AI-driven analysis of SICS videos.

This initiative represents a significant step for global cataract surgery research, providing accessible data and tools that can drive improvements in surgical quality and patient outcomes where they are most needed.

• Public dataset: SICS-105 and SICS-155 datasets
• Learn more about AI in healthcare: Latest AI Courses