pakheiyeung / PlaneInVol

A network, trained with just registered volumes, to predict the 3D location of 2D freehand ultrasound fetal brain images and video

Learning to Map 2D Ultrasound Images into 3D Space with Minimal Human Annotation

This repository contains the codes (in PyTorch) for the framework introduced in the following paper:

Learning to Map 2D Ultrasound Images into 3D Space with Minimal Human Annotation [Paper] [Project Page]

@article{yeung2021learning,
	title = {Learning to map 2D ultrasound images into 3D space with minimal human annotation},
	author = {Yeung, Pak-Hei and Aliasi, Moska and Papageorghiou, Aris T and Haak, Monique and Xie, Weidi and Namburete, Ana IL},
	journal = {Medical Image Analysis},
	volume = {70},
	pages = {101998},
	year = {2021},
	publisher = {Elsevier}
}

Contents

1. Dependencies
2. Train
3. Inference on Video Data

Dependencies

• Python (3.6), other versions should also work
• PyTorch (1.6), other versions should also work
• scipy
• cv2
• skimage
• imgaug

Train

1. Modify the train_github.py for importing your dataset according to the instructions in the file.
2. Use the following command to train:

   python train_github.py
   

Inference on Video Data

Post (predict after the whole video is acquired)

1. Modify the inference_github.py according to the instructions in the file.
2. Use the following command to train:

   python inference_github.py
   

Real-time

To do, but it would be an easy modification based on the inference_github.py.