Aorta Mapping
Fall 2023, 16.485 - Visual Navigation for Autonomous Vehicles project
Skills Demonstrated
- Team Work
- Machine Learning - CNN
- TSDF
- C++
- Python
Project Description
In this work we use data taken while inserting a surgical probe into a synthetic aorta model to reconstruct a 3D model of the dissected aorta. Note that this is part of a larger ongoing project conducted by Tom Dillon in the Therapeutic Technology Design & Development lab at MIT. The data consists of 2D ultrasound cross section images with accompanying 6DOF position data. A key challenge is to extract the 2D contour of the aortic wall and dissection flap. Ultrasound images are often noisy and the dissection flap can cause poor visualization of the far wall. We used machine learning to segment the two halves of the aorta to extract the contours for generating a 3D mesh using the truncated signed distance function (TSDF) and marching cubes algorithm. This enhanced visualization should aid surgeons performing aortic dissection operations.
Responsibilities
I developed a CNN model using the UNet architecture to segment the ultrasound images. I then integrated this segmentation code with the Voxblox library TSDF implementation code to create a 3D mesh of the dissected aorta.
