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

Kurzfassung

This work proves that semantic segmentation on Minimally Invasive Surgical Instruments can be improved by using training data augmented through domain adaptation. The benefit of these methods is two fold. Firstly, it suppresses the need of manually labeling thousands of images, by transforming synthetic data into realistic-looking data. To achieve this, a CycleGAN model is used, which transforms a source dataset to approximate the domain distribution of a target dataset. Secondly, this new generated data with perfect labels is utilized to train a semantic segmentation neural network, a U-Net. This method shows great generalization capabilities on data with great variability, i.e. the model is rotation- position- and lighting conditions invariant. Nevertheless, one of the caveats of this approach is that the model is unable to generalize well to other surgical instruments with a different shape of the one used for training. This seems to be driven by the lack of a high variance in the geometric distribution of the training data. To this end future work should focus on making the model more scale-invariant and able to adapt to other types of surgical instruments previously unseen by the training.