README.md

Standard Plane Localisation in 3D Ultrasound

This software implements a Convolutional Neural Network (CNN) for the automatic localisation of standard scan planes in 3D ultrasound of the fetal head.

Tensorflow implementation of the MICCAI 2018 paper Standard Plane Detection in 3D Fetal Ultrasound Using an Iterative Transformation Network.

Prerequisites

• Python 2.7
• Tensorflow 1.8.0
• SciPy
• (Optional) NiBabel (for reading NIfTI input)
• (Optional) Matplotlib (for visualization)
• (Optional) scikit-image (for computing SSIM)
• (Optional) srmg (for computing Riemannian mean. Already included in this repo)
• (Optional) transformations.py (for computations on rotation matrices, Euler angles and quaternions. Already included in this repo)

Usage

To train a CNN model:

$ python train.py

To test with an existing CNN model:

$ python infer.py

Data

We are not able to share the dataset we used for the paper due to sensitive patient information. However, we have provided a dummy data which is a random noise matrix to mimic the actual data we used.

Results

Ground truth (red) and predicted (green) transcerebellar standard plane

Path taken by the plane over 10 iterations during inference

Author

Yuanwei Li / yuanwei_li@hotmail.com