Deep Learning Ventricle Segmentation
Zach, 05 January 2022
This project was a part of my Master’s research where the goal was to create an automatic deep learning segmentation approach which performed quickly and provided accurate segmentation results. We used the U-Net model as our baseline which is useful for small datasets like our own and well-suited for 3D ultrasound data. Some of our earlier preliminary work can be seen in our conference paper here which looked into 2D multiplane models for solving our problem.
We later published a journal paper describing a fully automated method that exceeds the SPIE work and current state of the art work for ventricle segmentation. This work uses a 3D U-Net ensemble and can predict results in ~5 seconds. This work is described here with a link to the github repository here that will be updated as the code is ready.