Automatic recognition of relevant segments in ophthalmic surgery videos

Researchers from the fields of computer science and medicine are collaborating to make better use of microscopic videos that record surgical procedures relating to the eyes for the purpose of teaching, research and documentation.

Ophthalmic surgery is usually performed with the aid of a microscope that can be equipped with a video camera in order to record the surgical procedure. Videos captured in this way are highly significant for medical education, research and documentation. The research team working with Klaus Schöffmann and Mario Taschwer (Department of Information Technology) is collaborating with physicians from the Klinikum hospital in Klagenfurt to develop methods for the automatic analysis of these videos. The project bears the title “Relevance Detection in Ophthalmic Surgery Videos” and is supported by the Austrian Science Fund (FWF), which is funding positions for three doctoral students over a period of three years.

Efforts are focused specifically on the automatic detection of relevant temporal segments in surgery videos. Klaus Schöffmann explains: “Relevant content should be learnt automatically through the application of machine learning methods (neural networks in particular), with surgery videos that have been annotated by surgeons serving as training data.” Important video segments might include, for instance, irregular operation phases, which deviate from the procedure used in quasi-standardized eye surgery. Mario Taschwer goes on to add: ”As well as recognising deviations from regular surgical procedures, we also hope to identify and differentiate between irregularities that occur with a certain frequency. To achieve this, we aim to develop and evaluate automatic classification procedures for ophthalmic surgery videos.”

The videos analysed in this way will not only simplify the tracking of video segments featuring “conspicuous” phases of surgery, but thanks to the automatic relevance detection, it will also be possible to improve the compression and storage process. The research project is due to commence in October, with initial results expected approximately one year later.