Soldiers that are in the field do not always have access to hospitals or professionally-trained medical personnel. In order to save their lives in case of injury, we would like to create an autonomous system that can help prolong their lives until they can reach adequate help. We have several mechanisms to be able to do this, but the one we will be focusing on is inserting a needle in the femoral artery autonomously with a robot to be able to perform minimally invasive cardiac surgery. Atur Dubrawski's work focuses on creating a map of the internal vessel structure of the leg (via segmentation and 3D reconstruction) and tracking the needle to ensure that it gets inserted at our point of interest and does not perforate the artery.

Some of the major challenges we face are lack of labelled data. In order to train better models, we are proposing to create simulated data. The goals we have are to:

  1. Refactor existing code to be more computationally efficient.
  2. Use mathematical methods to improve the data quality (ex. interpolate existing holes in the data)
  3. Use deep learning to create baselines for our experiments. You will be working with ultrasound images that come from three different mediums: a blue phantom, a leg phantom, and pig surgeries.

Programming experience is required. Some understanding of robot kinematics and image transformations is important, while knowledge of ROS and deep learning would be beneficial but not required.