Surgical robot to learn suture treatment by watching video

The suture work after treatment of the affected area during surgery is the final finish and a simple repetition of work. Intel and UC Berkeley created a robot to solve this monotony. The robot can automatically close patient wounds. It is said that AI can learn skilled suture work images and imitate it with high precision.

A system for learning AI from a video called Motion2Vec differs in that it uses a function that learns through visual observation, while the existing method required learning in advance with static methods or imitated motion trajectories. There is.

The approximate movement of an object seen by humans can be imitated at a glance. On the other hand, the robot only understands this as a pixel stream flowing at several tens of fps, and it needs to interpret the image and substitute it for its own movement. To realize this, the research team used the Siamese Network. The Siamese network is an old machine learning technique that calculates whether two images are similar or similar as a function of distance as one of the methods for training AI neural networks.

This network is used to discriminate whether or not they are the same person by comparing the face reflected on the surveillance cam with the face in the ID photo, for example, to evaluate and learn the similarity among the two. In this study, the aim was to raise the robot movement to a level close to humans by comparing the image of the doctor’s arm movement with the image of the surgical robot arm movement. Although only 78 images were learned, the segmentation accuracy was 85.5% and the operation average error was 0.94cm.

The research team does not believe that this will eliminate surgeries by hand in the future. Surgical robots trained with this technology still need a few more years before they can actually work in the operating room. Moreover, it is limited to monotonous sutures after the operation is almost finished. Of course, if AI is trained in an appropriate way, there is a possibility that it will eventually take on various auxiliary tasks such as suturing as well as removing dead tissue. However, just being able to take charge of these tasks is expected to help surgeons reduce operating time and focus on complex tasks. Related information can be found here .