eduzhai > Applied Sciences > Engineering >

Segmentation of Surgical Instruments for Minimally-Invasive Robot-Assisted Procedures Using Generative Deep Neural Networks

  • Save

... pages left unread,continue reading

Document pages: 7 pages

Abstract: This work proves that semantic segmentation on minimally invasive surgicalinstruments can be improved by using training data that has been augmentedthrough domain adaptation. The benefit of this method is twofold. Firstly, itsuppresses the need of manually labeling thousands of images by transformingsynthetic data into realistic-looking data. To achieve this, a CycleGAN modelis used, which transforms a source dataset to approximate the domaindistribution of a target dataset. Secondly, this newly generated data withperfect labels is utilized to train a semantic segmentation neural network,U-Net. This method shows generalization capabilities on data with variabilityregarding its rotation- position- and lighting conditions. Nevertheless, one ofthe caveats of this approach is that the model is unable to generalize well toother surgical instruments with a different shape from the one used fortraining. This is driven by the lack of a high variance in the geometricdistribution of the training data. Future work will focus on making the modelmore scale-invariant and able to adapt to other types of surgical instrumentspreviously unseen by the training.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×