With cataract interventions increasing, improvements in safety and efficiency of the surgery are active research topics. Examples include automated post-operative analysis, surgical training, and real-time decision support for which automated comprehension of surgical activities is an important part.
The method proposed by Michel Hayoz, Mathias Gallardo, Pablo Neila Marquez and Raphal Sznitman from the ARTORG AIMI lab together with Martin Zinkernagel, Ophthalmology Inselspital, takes the surgical video stream from the microscope and recognizes the surgical activity based on the current and past video frames. The team proposes to train a deep learning model to jointly recognize surgical tools and the activity in the scene. By exploiting the relationship between the two, the model achieved promising performance, winning the CATARACTS Workflow Challenge 2020 worth USD 1250.
MICCAI 2020 cataracts challenge: https://cataracts2020.grand-challenge.org/
Proposed solution by the ARTORG AIMI lab
Artificial Intelligence in Medical Imaging