Fetal-head-segmentation-and-circumference-measurement-from-ultrasound-images
This repo is a PyTorch implementation of a U-Net model for the segmentation of ultrasound images in order to estimate the circumference of the head of the fetus. The dataset used for the project was taken from the HC18 challenge.
U-Net
The U-Net implemented is shown in the figure below:
The encoder contains 3x3 convolutions with same padding followed by maxpool layers. The used activation function was ReLU. Resnet like addition connections were incorporated in the network to promote gradient flow during backpropagation. Decoder was implemented using Transverse convolutions with sigmoid activation. Binary corss entropy loss was used as the loss function and the model reached a Dice score of 0.95 for the validation data in just 30 epochs. The summary of the model is as following:
A model trained for 30 epochs can be found here.
Problem Statement
The problem statement of the HC18 challenge was to model the head of the fetus as an ellipse. The approach employed to solve the problem was to segment the fetal head followed by using Hough transform to fit the ellipse. The Unet model was trained to segment the ultrasound images and ellipse fitting was performed using the OpenCV implementation of Hough transforms. The results are as follows:
Segmentation
| Input image | Segmentation result |
|---|---|
 |
 |
Hough Trasform For Ellipse Fitting
| Ellipse fitted on the segmented image | Ellipse with reference image |
|---|---|
 |
 |
References
- Thomas L. A. van den Heuvel, Dagmar de Bruijn, Chris L. de Korte and Bram van Ginneken. Automated measurement of fetal head circumference using 2D ultrasound images. PloS one, 13.8 (2018): e0200412.
- Thomas L. A. van den Heuvel, Dagmar de Bruijn, Chris L. de Korte and Bram van Ginneken. Automated measurement of fetal head circumference using 2D ultrasound images [Data set]. Zenodo. http://doi.org/10.5281/zenodo.1322001
- Olaf Ronneberger, Philipp Fischer, and Thomas Brox, U-Net: Convolutional Networks for Biomedical Image Segmentation. https://arxiv.org/pdf/1505.04597.pdf