Researchers have launched the largest open-access dataset for tool instance segmentation in laparoscopic surgery.
Precise tool instance segmentation is an essential technology for advanced computer-assisted interventions. Researchers from King’s College London and Tongji University have now unveiled CholecInstanceSeg.
This resource is set to accelerate the development of intelligent surgical systems that can assist or even automate parts of minimally invasive procedures.
Minimally invasive surgery (MIS), such as laparoscopic cholecystectomy, offers patients faster recovery and fewer complications.
However, it also presents challenges for surgeons, who must navigate complex anatomy using a limited field of view from a laparoscope.
To enhance safety and precision, AI-powered systems are being developed to recognise and track surgical tools in real time.
However, progress has been hindered by a lack of high-quality, annotated datasets until now.
CholecInstanceSeg addresses this gap by providing 41,900 annotated video frames from 85 real-world surgeries, capturing over 64,400 individual tool instances.
Unlike previous datasets, which often relied on animal models or lacked detailed annotations, this dataset is derived from actual human procedures. It includes semantic masks and unique instance IDs for each tool.
The dataset builds on the widely used CholecT50 and Cholec80 video collections, enhancing them with precise, frame-by-frame annotations.
The research team conducted rigorous quality control and inter-annotator agreement checks to ensure the reliability of the labels.
They also benchmarked the dataset using several state-of-the-art segmentation algorithms, demonstrating its value for training and evaluating AI models.
By making CholecInstanceSeg freely available, the researchers hope to foster innovation in computer-assisted surgery, including real-time tool tracking, surgical workflow analysis and even autonomous robotic interventions.
The dataset is expected to serve as a cornerstone for future research in surgical AI, helping to bridge the gap between experimental models and real-world clinical applications.
Oluwatosin Alabi, lead author, said: ‘Our hope with CholecInstanceSeg is that it will really supercharge the development of AI systems that can assist surgeons during the Cholecystectomy procedure.
‘With better tools to understand and track what’s happening in surgery, we’re looking at a future where patient outcomes are improved, and procedures become safer and more precise for everyone involved.’
